├── 1080Ti Notebook.ipynb
├── 2080Ti.ipynb
└── README.md


/1080Ti Notebook.ipynb:
--------------------------------------------------------------------------------
   1 | {
   2 |  "cells": [
   3 |   {
   4 |    "cell_type": "code",
   5 |    "execution_count": 1,
   6 |    "metadata": {},
   7 |    "outputs": [],
   8 |    "source": [
   9 |     "%reload_ext autoreload\n",
  10 |     "%autoreload 2\n",
  11 |     "%matplotlib inline"
  12 |    ]
  13 |   },
  14 |   {
  15 |    "cell_type": "code",
  16 |    "execution_count": 2,
  17 |    "metadata": {},
  18 |    "outputs": [],
  19 |    "source": [
  20 |     "import functools\n",
  21 |     "import traceback\n",
  22 |     "def get_ref_free_exc_info():\n",
  23 |     "    \"Free traceback from references to locals/globals to avoid circular reference leading to gc.collect() unable to reclaim memory\"\n",
  24 |     "    type, val, tb = sys.exc_info()\n",
  25 |     "    traceback.clear_frames(tb)\n",
  26 |     "    return (type, val, tb)\n",
  27 |     "\n",
  28 |     "def gpu_mem_restore(func):\n",
  29 |     "    \"Reclaim GPU RAM if CUDA out of memory happened, or execution was interrupted\"\n",
  30 |     "    @functools.wraps(func)\n",
  31 |     "    def wrapper(*args, **kwargs):\n",
  32 |     "        try:\n",
  33 |     "            return func(*args, **kwargs)\n",
  34 |     "        except:\n",
  35 |     "            type, val, tb = get_ref_free_exc_info() # must!\n",
  36 |     "            raise type(val).with_traceback(tb) from None\n",
  37 |     "    return wrapper"
  38 |    ]
  39 |   },
  40 |   {
  41 |    "cell_type": "code",
  42 |    "execution_count": 3,
  43 |    "metadata": {},
  44 |    "outputs": [],
  45 |    "source": [
  46 |     "from fastai.vision import *\n",
  47 |     "from fastai.metrics import error_rate"
  48 |    ]
  49 |   },
  50 |   {
  51 |    "cell_type": "code",
  52 |    "execution_count": 4,
  53 |    "metadata": {},
  54 |    "outputs": [],
  55 |    "source": [
  56 |     "path = Path('/home/ekami/workspace/cifar100/')"
  57 |    ]
  58 |   },
  59 |   {
  60 |    "cell_type": "code",
  61 |    "execution_count": 5,
  62 |    "metadata": {},
  63 |    "outputs": [],
  64 |    "source": [
  65 |     "train = path/'train'\n",
  66 |     "test = path/'test'"
  67 |    ]
  68 |   },
  69 |   {
  70 |    "cell_type": "code",
  71 |    "execution_count": 6,
  72 |    "metadata": {},
  73 |    "outputs": [],
  74 |    "source": [
  75 |     "bs = 108"
  76 |    ]
  77 |   },
  78 |   {
  79 |    "cell_type": "code",
  80 |    "execution_count": 7,
  81 |    "metadata": {},
  82 |    "outputs": [
  83 |     {
  84 |      "data": {
  85 |       "text/plain": [
  86 |        "[PosixPath('/home/ekami/workspace/cifar100/test'),\n",
  87 |        " PosixPath('/home/ekami/workspace/cifar100/models'),\n",
  88 |        " PosixPath('/home/ekami/workspace/cifar100/train')]"
  89 |       ]
  90 |      },
  91 |      "execution_count": 7,
  92 |      "metadata": {},
  93 |      "output_type": "execute_result"
  94 |     }
  95 |    ],
  96 |    "source": [
  97 |     "path.ls()"
  98 |    ]
  99 |   },
 100 |   {
 101 |    "cell_type": "code",
 102 |    "execution_count": 8,
 103 |    "metadata": {},
 104 |    "outputs": [],
 105 |    "source": [
 106 |     "np.random.seed(42)\n",
 107 |     "data = ImageDataBunch.from_folder(path, train=train, valid_pct=0.2,\n",
 108 |     "        ds_tfms=get_transforms(), size=224, num_workers=4, bs=bs).normalize(cifar_stats)"
 109 |    ]
 110 |   },
 111 |   {
 112 |    "cell_type": "code",
 113 |    "execution_count": 9,
 114 |    "metadata": {},
 115 |    "outputs": [],
 116 |    "source": [
 117 |     "class gpu_mem_restore_ctx():\n",
 118 |     "    \" context manager to reclaim GPU RAM if CUDA out of memory happened, or execution was interrupted\"\n",
 119 |     "    def __enter__(self): return self\n",
 120 |     "    def __exit__(self, exc_type, exc_val, exc_tb):\n",
 121 |     "        if not exc_val: return True\n",
 122 |     "        traceback.clear_frames(exc_tb)\n",
 123 |     "        raise exc_type(exc_val).with_traceback(exc_tb) from None"
 124 |    ]
 125 |   },
 126 |   {
 127 |    "cell_type": "code",
 128 |    "execution_count": 10,
 129 |    "metadata": {},
 130 |    "outputs": [],
 131 |    "source": [
 132 |     "#Allow crashing\n",
 133 |     "learn = create_cnn(data, models.resnet18 , metrics=accuracy)"
 134 |    ]
 135 |   },
 136 |   {
 137 |    "cell_type": "markdown",
 138 |    "metadata": {},
 139 |    "source": [
 140 |     "## Resnet 18"
 141 |    ]
 142 |   },
 143 |   {
 144 |    "cell_type": "code",
 145 |    "execution_count": 11,
 146 |    "metadata": {},
 147 |    "outputs": [
 148 |     {
 149 |      "data": {
 150 |       "text/plain": [
 151 |        "405"
 152 |       ]
 153 |      },
 154 |      "execution_count": 11,
 155 |      "metadata": {},
 156 |      "output_type": "execute_result"
 157 |     }
 158 |    ],
 159 |    "source": [
 160 |     "int(bs*3.75)"
 161 |    ]
 162 |   },
 163 |   {
 164 |    "cell_type": "code",
 165 |    "execution_count": 12,
 166 |    "metadata": {},
 167 |    "outputs": [],
 168 |    "source": [
 169 |     "np.random.seed(42)\n",
 170 |     "data = ImageDataBunch.from_folder(path, train=train, valid_pct=0.2,\n",
 171 |     "        ds_tfms=get_transforms(), size=224, num_workers=4, bs=408).normalize(cifar_stats)\n",
 172 |     "learn = create_cnn(data, models.resnet18, metrics=accuracy)"
 173 |    ]
 174 |   },
 175 |   {
 176 |    "cell_type": "code",
 177 |    "execution_count": 13,
 178 |    "metadata": {},
 179 |    "outputs": [
 180 |     {
 181 |      "data": {
 182 |       "text/html": [
 183 |        "Total time: 01:14 <p><table style='width:300px; margin-bottom:10px'>\n",
 184 |        "  <tr>\n",
 185 |        "    <th>epoch</th>\n",
 186 |        "    <th>train_loss</th>\n",
 187 |        "    <th>valid_loss</th>\n",
 188 |        "    <th>accuracy</th>\n",
 189 |        "  </tr>\n",
 190 |        "  <tr>\n",
 191 |        "    <th>1</th>\n",
 192 |        "    <th>2.441905</th>\n",
 193 |        "    <th>1.621814</th>\n",
 194 |        "    <th>0.561000</th>\n",
 195 |        "  </tr>\n",
 196 |        "</table>\n"
 197 |       ],
 198 |       "text/plain": [
 199 |        "<IPython.core.display.HTML object>"
 200 |       ]
 201 |      },
 202 |      "metadata": {},
 203 |      "output_type": "display_data"
 204 |     }
 205 |    ],
 206 |    "source": [
 207 |     "with gpu_mem_restore_ctx():\n",
 208 |     "    learn.fit_one_cycle(1)"
 209 |    ]
 210 |   },
 211 |   {
 212 |    "cell_type": "code",
 213 |    "execution_count": 14,
 214 |    "metadata": {},
 215 |    "outputs": [
 216 |     {
 217 |      "data": {
 218 |       "text/html": [
 219 |        "Total time: 35:26 <p><table style='width:300px; margin-bottom:10px'>\n",
 220 |        "  <tr>\n",
 221 |        "    <th>epoch</th>\n",
 222 |        "    <th>train_loss</th>\n",
 223 |        "    <th>valid_loss</th>\n",
 224 |        "    <th>accuracy</th>\n",
 225 |        "  </tr>\n",
 226 |        "  <tr>\n",
 227 |        "    <th>1</th>\n",
 228 |        "    <th>1.686131</th>\n",
 229 |        "    <th>1.220904</th>\n",
 230 |        "    <th>0.651333</th>\n",
 231 |        "  </tr>\n",
 232 |        "  <tr>\n",
 233 |        "    <th>2</th>\n",
 234 |        "    <th>1.388124</th>\n",
 235 |        "    <th>0.996638</th>\n",
 236 |        "    <th>0.703500</th>\n",
 237 |        "  </tr>\n",
 238 |        "  <tr>\n",
 239 |        "    <th>3</th>\n",
 240 |        "    <th>1.216391</th>\n",
 241 |        "    <th>0.915040</th>\n",
 242 |        "    <th>0.721250</th>\n",
 243 |        "  </tr>\n",
 244 |        "  <tr>\n",
 245 |        "    <th>4</th>\n",
 246 |        "    <th>1.165618</th>\n",
 247 |        "    <th>0.937397</th>\n",
 248 |        "    <th>0.716500</th>\n",
 249 |        "  </tr>\n",
 250 |        "  <tr>\n",
 251 |        "    <th>5</th>\n",
 252 |        "    <th>1.166939</th>\n",
 253 |        "    <th>1.005142</th>\n",
 254 |        "    <th>0.696083</th>\n",
 255 |        "  </tr>\n",
 256 |        "  <tr>\n",
 257 |        "    <th>6</th>\n",
 258 |        "    <th>1.167172</th>\n",
 259 |        "    <th>1.036181</th>\n",
 260 |        "    <th>0.690250</th>\n",
 261 |        "  </tr>\n",
 262 |        "  <tr>\n",
 263 |        "    <th>7</th>\n",
 264 |        "    <th>1.155099</th>\n",
 265 |        "    <th>1.098619</th>\n",
 266 |        "    <th>0.678833</th>\n",
 267 |        "  </tr>\n",
 268 |        "  <tr>\n",
 269 |        "    <th>8</th>\n",
 270 |        "    <th>1.141750</th>\n",
 271 |        "    <th>1.031859</th>\n",
 272 |        "    <th>0.692500</th>\n",
 273 |        "  </tr>\n",
 274 |        "  <tr>\n",
 275 |        "    <th>9</th>\n",
 276 |        "    <th>1.122253</th>\n",
 277 |        "    <th>0.973910</th>\n",
 278 |        "    <th>0.700917</th>\n",
 279 |        "  </tr>\n",
 280 |        "  <tr>\n",
 281 |        "    <th>10</th>\n",
 282 |        "    <th>1.078584</th>\n",
 283 |        "    <th>0.987744</th>\n",
 284 |        "    <th>0.700667</th>\n",
 285 |        "  </tr>\n",
 286 |        "  <tr>\n",
 287 |        "    <th>11</th>\n",
 288 |        "    <th>1.049187</th>\n",
 289 |        "    <th>0.925535</th>\n",
 290 |        "    <th>0.718917</th>\n",
 291 |        "  </tr>\n",
 292 |        "  <tr>\n",
 293 |        "    <th>12</th>\n",
 294 |        "    <th>1.023786</th>\n",
 295 |        "    <th>0.923716</th>\n",
 296 |        "    <th>0.723000</th>\n",
 297 |        "  </tr>\n",
 298 |        "  <tr>\n",
 299 |        "    <th>13</th>\n",
 300 |        "    <th>0.994283</th>\n",
 301 |        "    <th>0.950193</th>\n",
 302 |        "    <th>0.714333</th>\n",
 303 |        "  </tr>\n",
 304 |        "  <tr>\n",
 305 |        "    <th>14</th>\n",
 306 |        "    <th>0.978886</th>\n",
 307 |        "    <th>0.939824</th>\n",
 308 |        "    <th>0.715583</th>\n",
 309 |        "  </tr>\n",
 310 |        "  <tr>\n",
 311 |        "    <th>15</th>\n",
 312 |        "    <th>0.947734</th>\n",
 313 |        "    <th>0.900096</th>\n",
 314 |        "    <th>0.729167</th>\n",
 315 |        "  </tr>\n",
 316 |        "  <tr>\n",
 317 |        "    <th>16</th>\n",
 318 |        "    <th>0.912281</th>\n",
 319 |        "    <th>0.827112</th>\n",
 320 |        "    <th>0.747167</th>\n",
 321 |        "  </tr>\n",
 322 |        "  <tr>\n",
 323 |        "    <th>17</th>\n",
 324 |        "    <th>0.877508</th>\n",
 325 |        "    <th>0.794671</th>\n",
 326 |        "    <th>0.759000</th>\n",
 327 |        "  </tr>\n",
 328 |        "  <tr>\n",
 329 |        "    <th>18</th>\n",
 330 |        "    <th>0.848763</th>\n",
 331 |        "    <th>0.807713</th>\n",
 332 |        "    <th>0.751250</th>\n",
 333 |        "  </tr>\n",
 334 |        "  <tr>\n",
 335 |        "    <th>19</th>\n",
 336 |        "    <th>0.832954</th>\n",
 337 |        "    <th>0.764302</th>\n",
 338 |        "    <th>0.766583</th>\n",
 339 |        "  </tr>\n",
 340 |        "  <tr>\n",
 341 |        "    <th>20</th>\n",
 342 |        "    <th>0.783972</th>\n",
 343 |        "    <th>0.722574</th>\n",
 344 |        "    <th>0.776667</th>\n",
 345 |        "  </tr>\n",
 346 |        "  <tr>\n",
 347 |        "    <th>21</th>\n",
 348 |        "    <th>0.765427</th>\n",
 349 |        "    <th>0.726920</th>\n",
 350 |        "    <th>0.779167</th>\n",
 351 |        "  </tr>\n",
 352 |        "  <tr>\n",
 353 |        "    <th>22</th>\n",
 354 |        "    <th>0.740291</th>\n",
 355 |        "    <th>0.716551</th>\n",
 356 |        "    <th>0.779667</th>\n",
 357 |        "  </tr>\n",
 358 |        "  <tr>\n",
 359 |        "    <th>23</th>\n",
 360 |        "    <th>0.711279</th>\n",
 361 |        "    <th>0.689046</th>\n",
 362 |        "    <th>0.785083</th>\n",
 363 |        "  </tr>\n",
 364 |        "  <tr>\n",
 365 |        "    <th>24</th>\n",
 366 |        "    <th>0.683742</th>\n",
 367 |        "    <th>0.683559</th>\n",
 368 |        "    <th>0.788750</th>\n",
 369 |        "  </tr>\n",
 370 |        "  <tr>\n",
 371 |        "    <th>25</th>\n",
 372 |        "    <th>0.657032</th>\n",
 373 |        "    <th>0.672859</th>\n",
 374 |        "    <th>0.791583</th>\n",
 375 |        "  </tr>\n",
 376 |        "  <tr>\n",
 377 |        "    <th>26</th>\n",
 378 |        "    <th>0.634756</th>\n",
 379 |        "    <th>0.666527</th>\n",
 380 |        "    <th>0.795333</th>\n",
 381 |        "  </tr>\n",
 382 |        "  <tr>\n",
 383 |        "    <th>27</th>\n",
 384 |        "    <th>0.622170</th>\n",
 385 |        "    <th>0.663841</th>\n",
 386 |        "    <th>0.796250</th>\n",
 387 |        "  </tr>\n",
 388 |        "  <tr>\n",
 389 |        "    <th>28</th>\n",
 390 |        "    <th>0.613146</th>\n",
 391 |        "    <th>0.658232</th>\n",
 392 |        "    <th>0.797417</th>\n",
 393 |        "  </tr>\n",
 394 |        "  <tr>\n",
 395 |        "    <th>29</th>\n",
 396 |        "    <th>0.613548</th>\n",
 397 |        "    <th>0.658717</th>\n",
 398 |        "    <th>0.797667</th>\n",
 399 |        "  </tr>\n",
 400 |        "  <tr>\n",
 401 |        "    <th>30</th>\n",
 402 |        "    <th>0.612183</th>\n",
 403 |        "    <th>0.658733</th>\n",
 404 |        "    <th>0.798000</th>\n",
 405 |        "  </tr>\n",
 406 |        "</table>\n"
 407 |       ],
 408 |       "text/plain": [
 409 |        "<IPython.core.display.HTML object>"
 410 |       ]
 411 |      },
 412 |      "metadata": {},
 413 |      "output_type": "display_data"
 414 |     }
 415 |    ],
 416 |    "source": [
 417 |     "with gpu_mem_restore_ctx():\n",
 418 |     "    learn.fit_one_cycle(30, max_lr=1e-2)"
 419 |    ]
 420 |   },
 421 |   {
 422 |    "cell_type": "markdown",
 423 |    "metadata": {},
 424 |    "source": [
 425 |     "## Resnet 18 (Mixed Prec)"
 426 |    ]
 427 |   },
 428 |   {
 429 |    "cell_type": "code",
 430 |    "execution_count": 15,
 431 |    "metadata": {},
 432 |    "outputs": [
 433 |     {
 434 |      "data": {
 435 |       "text/plain": [
 436 |        "108"
 437 |       ]
 438 |      },
 439 |      "execution_count": 15,
 440 |      "metadata": {},
 441 |      "output_type": "execute_result"
 442 |     }
 443 |    ],
 444 |    "source": [
 445 |     "bs"
 446 |    ]
 447 |   },
 448 |   {
 449 |    "cell_type": "code",
 450 |    "execution_count": 16,
 451 |    "metadata": {},
 452 |    "outputs": [],
 453 |    "source": [
 454 |     "np.random.seed(42)\n",
 455 |     "data = ImageDataBunch.from_folder(path, train=train, valid_pct=0.2,\n",
 456 |     "        ds_tfms=get_transforms(), size=224, num_workers=4, bs=720).normalize(cifar_stats)"
 457 |    ]
 458 |   },
 459 |   {
 460 |    "cell_type": "code",
 461 |    "execution_count": 17,
 462 |    "metadata": {},
 463 |    "outputs": [],
 464 |    "source": [
 465 |     "learn = to_fp16(create_cnn(data, models.resnet18, metrics=accuracy))"
 466 |    ]
 467 |   },
 468 |   {
 469 |    "cell_type": "code",
 470 |    "execution_count": 18,
 471 |    "metadata": {},
 472 |    "outputs": [
 473 |     {
 474 |      "data": {
 475 |       "text/html": [
 476 |        "Total time: 01:12 <p><table style='width:300px; margin-bottom:10px'>\n",
 477 |        "  <tr>\n",
 478 |        "    <th>epoch</th>\n",
 479 |        "    <th>train_loss</th>\n",
 480 |        "    <th>valid_loss</th>\n",
 481 |        "    <th>accuracy</th>\n",
 482 |        "  </tr>\n",
 483 |        "  <tr>\n",
 484 |        "    <th>1</th>\n",
 485 |        "    <th>2.870623</th>\n",
 486 |        "    <th>1.839730</th>\n",
 487 |        "    <th>0.519083</th>\n",
 488 |        "  </tr>\n",
 489 |        "</table>\n"
 490 |       ],
 491 |       "text/plain": [
 492 |        "<IPython.core.display.HTML object>"
 493 |       ]
 494 |      },
 495 |      "metadata": {},
 496 |      "output_type": "display_data"
 497 |     }
 498 |    ],
 499 |    "source": [
 500 |     "with gpu_mem_restore_ctx():\n",
 501 |     "    learn.fit_one_cycle(1)"
 502 |    ]
 503 |   },
 504 |   {
 505 |    "cell_type": "code",
 506 |    "execution_count": 19,
 507 |    "metadata": {},
 508 |    "outputs": [
 509 |     {
 510 |      "data": {
 511 |       "text/html": [
 512 |        "Total time: 34:08 <p><table style='width:300px; margin-bottom:10px'>\n",
 513 |        "  <tr>\n",
 514 |        "    <th>epoch</th>\n",
 515 |        "    <th>train_loss</th>\n",
 516 |        "    <th>valid_loss</th>\n",
 517 |        "    <th>accuracy</th>\n",
 518 |        "  </tr>\n",
 519 |        "  <tr>\n",
 520 |        "    <th>1</th>\n",
 521 |        "    <th>2.207675</th>\n",
 522 |        "    <th>1.754365</th>\n",
 523 |        "    <th>0.537417</th>\n",
 524 |        "  </tr>\n",
 525 |        "  <tr>\n",
 526 |        "    <th>2</th>\n",
 527 |        "    <th>2.115157</th>\n",
 528 |        "    <th>1.612026</th>\n",
 529 |        "    <th>0.565667</th>\n",
 530 |        "  </tr>\n",
 531 |        "  <tr>\n",
 532 |        "    <th>3</th>\n",
 533 |        "    <th>1.955724</th>\n",
 534 |        "    <th>1.431037</th>\n",
 535 |        "    <th>0.600417</th>\n",
 536 |        "  </tr>\n",
 537 |        "  <tr>\n",
 538 |        "    <th>4</th>\n",
 539 |        "    <th>1.766447</th>\n",
 540 |        "    <th>1.264467</th>\n",
 541 |        "    <th>0.639083</th>\n",
 542 |        "  </tr>\n",
 543 |        "  <tr>\n",
 544 |        "    <th>5</th>\n",
 545 |        "    <th>1.590725</th>\n",
 546 |        "    <th>1.155779</th>\n",
 547 |        "    <th>0.663833</th>\n",
 548 |        "  </tr>\n",
 549 |        "  <tr>\n",
 550 |        "    <th>6</th>\n",
 551 |        "    <th>1.441305</th>\n",
 552 |        "    <th>1.066461</th>\n",
 553 |        "    <th>0.682167</th>\n",
 554 |        "  </tr>\n",
 555 |        "  <tr>\n",
 556 |        "    <th>7</th>\n",
 557 |        "    <th>1.334857</th>\n",
 558 |        "    <th>1.007020</th>\n",
 559 |        "    <th>0.702500</th>\n",
 560 |        "  </tr>\n",
 561 |        "  <tr>\n",
 562 |        "    <th>8</th>\n",
 563 |        "    <th>1.246409</th>\n",
 564 |        "    <th>0.961123</th>\n",
 565 |        "    <th>0.711833</th>\n",
 566 |        "  </tr>\n",
 567 |        "  <tr>\n",
 568 |        "    <th>9</th>\n",
 569 |        "    <th>1.181972</th>\n",
 570 |        "    <th>0.911834</th>\n",
 571 |        "    <th>0.725500</th>\n",
 572 |        "  </tr>\n",
 573 |        "  <tr>\n",
 574 |        "    <th>10</th>\n",
 575 |        "    <th>1.128325</th>\n",
 576 |        "    <th>0.881794</th>\n",
 577 |        "    <th>0.732333</th>\n",
 578 |        "  </tr>\n",
 579 |        "  <tr>\n",
 580 |        "    <th>11</th>\n",
 581 |        "    <th>1.078243</th>\n",
 582 |        "    <th>0.861121</th>\n",
 583 |        "    <th>0.739583</th>\n",
 584 |        "  </tr>\n",
 585 |        "  <tr>\n",
 586 |        "    <th>12</th>\n",
 587 |        "    <th>1.031876</th>\n",
 588 |        "    <th>0.839541</th>\n",
 589 |        "    <th>0.744167</th>\n",
 590 |        "  </tr>\n",
 591 |        "  <tr>\n",
 592 |        "    <th>13</th>\n",
 593 |        "    <th>1.008928</th>\n",
 594 |        "    <th>0.831407</th>\n",
 595 |        "    <th>0.744000</th>\n",
 596 |        "  </tr>\n",
 597 |        "  <tr>\n",
 598 |        "    <th>14</th>\n",
 599 |        "    <th>0.977578</th>\n",
 600 |        "    <th>0.809504</th>\n",
 601 |        "    <th>0.751833</th>\n",
 602 |        "  </tr>\n",
 603 |        "  <tr>\n",
 604 |        "    <th>15</th>\n",
 605 |        "    <th>0.945696</th>\n",
 606 |        "    <th>0.797486</th>\n",
 607 |        "    <th>0.757000</th>\n",
 608 |        "  </tr>\n",
 609 |        "  <tr>\n",
 610 |        "    <th>16</th>\n",
 611 |        "    <th>0.919172</th>\n",
 612 |        "    <th>0.782975</th>\n",
 613 |        "    <th>0.761917</th>\n",
 614 |        "  </tr>\n",
 615 |        "  <tr>\n",
 616 |        "    <th>17</th>\n",
 617 |        "    <th>0.899146</th>\n",
 618 |        "    <th>0.776590</th>\n",
 619 |        "    <th>0.766583</th>\n",
 620 |        "  </tr>\n",
 621 |        "  <tr>\n",
 622 |        "    <th>18</th>\n",
 623 |        "    <th>0.882939</th>\n",
 624 |        "    <th>0.767809</th>\n",
 625 |        "    <th>0.766000</th>\n",
 626 |        "  </tr>\n",
 627 |        "  <tr>\n",
 628 |        "    <th>19</th>\n",
 629 |        "    <th>0.856347</th>\n",
 630 |        "    <th>0.763404</th>\n",
 631 |        "    <th>0.767417</th>\n",
 632 |        "  </tr>\n",
 633 |        "  <tr>\n",
 634 |        "    <th>20</th>\n",
 635 |        "    <th>0.841534</th>\n",
 636 |        "    <th>0.754744</th>\n",
 637 |        "    <th>0.767250</th>\n",
 638 |        "  </tr>\n",
 639 |        "  <tr>\n",
 640 |        "    <th>21</th>\n",
 641 |        "    <th>0.817957</th>\n",
 642 |        "    <th>0.747933</th>\n",
 643 |        "    <th>0.771083</th>\n",
 644 |        "  </tr>\n",
 645 |        "  <tr>\n",
 646 |        "    <th>22</th>\n",
 647 |        "    <th>0.803184</th>\n",
 648 |        "    <th>0.746400</th>\n",
 649 |        "    <th>0.774833</th>\n",
 650 |        "  </tr>\n",
 651 |        "  <tr>\n",
 652 |        "    <th>23</th>\n",
 653 |        "    <th>0.791340</th>\n",
 654 |        "    <th>0.743390</th>\n",
 655 |        "    <th>0.773417</th>\n",
 656 |        "  </tr>\n",
 657 |        "  <tr>\n",
 658 |        "    <th>24</th>\n",
 659 |        "    <th>0.781753</th>\n",
 660 |        "    <th>0.739683</th>\n",
 661 |        "    <th>0.774750</th>\n",
 662 |        "  </tr>\n",
 663 |        "  <tr>\n",
 664 |        "    <th>25</th>\n",
 665 |        "    <th>0.769866</th>\n",
 666 |        "    <th>0.738879</th>\n",
 667 |        "    <th>0.776167</th>\n",
 668 |        "  </tr>\n",
 669 |        "  <tr>\n",
 670 |        "    <th>26</th>\n",
 671 |        "    <th>0.767070</th>\n",
 672 |        "    <th>0.737335</th>\n",
 673 |        "    <th>0.775583</th>\n",
 674 |        "  </tr>\n",
 675 |        "  <tr>\n",
 676 |        "    <th>27</th>\n",
 677 |        "    <th>0.755852</th>\n",
 678 |        "    <th>0.735069</th>\n",
 679 |        "    <th>0.775833</th>\n",
 680 |        "  </tr>\n",
 681 |        "  <tr>\n",
 682 |        "    <th>28</th>\n",
 683 |        "    <th>0.747134</th>\n",
 684 |        "    <th>0.734977</th>\n",
 685 |        "    <th>0.775833</th>\n",
 686 |        "  </tr>\n",
 687 |        "  <tr>\n",
 688 |        "    <th>29</th>\n",
 689 |        "    <th>0.744960</th>\n",
 690 |        "    <th>0.733925</th>\n",
 691 |        "    <th>0.776667</th>\n",
 692 |        "  </tr>\n",
 693 |        "  <tr>\n",
 694 |        "    <th>30</th>\n",
 695 |        "    <th>0.742359</th>\n",
 696 |        "    <th>0.735230</th>\n",
 697 |        "    <th>0.776667</th>\n",
 698 |        "  </tr>\n",
 699 |        "</table>\n"
 700 |       ],
 701 |       "text/plain": [
 702 |        "<IPython.core.display.HTML object>"
 703 |       ]
 704 |      },
 705 |      "metadata": {},
 706 |      "output_type": "display_data"
 707 |     }
 708 |    ],
 709 |    "source": [
 710 |     "with gpu_mem_restore_ctx():\n",
 711 |     "    learn.fit_one_cycle(30)"
 712 |    ]
 713 |   },
 714 |   {
 715 |    "cell_type": "markdown",
 716 |    "metadata": {},
 717 |    "source": [
 718 |     "## Resnet 34 "
 719 |    ]
 720 |   },
 721 |   {
 722 |    "cell_type": "code",
 723 |    "execution_count": 20,
 724 |    "metadata": {},
 725 |    "outputs": [],
 726 |    "source": [
 727 |     "np.random.seed(42)\n",
 728 |     "data = ImageDataBunch.from_folder(path, train=train, valid_pct=0.2,\n",
 729 |     "        ds_tfms=get_transforms(), size=224, num_workers=4, bs=248).normalize(cifar_stats)"
 730 |    ]
 731 |   },
 732 |   {
 733 |    "cell_type": "code",
 734 |    "execution_count": 21,
 735 |    "metadata": {},
 736 |    "outputs": [
 737 |     {
 738 |      "name": "stderr",
 739 |      "output_type": "stream",
 740 |      "text": [
 741 |       "Downloading: \"https://download.pytorch.org/models/resnet34-333f7ec4.pth\" to /home/ekami/.torch/models/resnet34-333f7ec4.pth\n",
 742 |       "100%|██████████| 87306240/87306240 [00:07<00:00, 11767995.32it/s]\n"
 743 |      ]
 744 |     }
 745 |    ],
 746 |    "source": [
 747 |     "learn = create_cnn(data, models.resnet34, metrics=accuracy)"
 748 |    ]
 749 |   },
 750 |   {
 751 |    "cell_type": "code",
 752 |    "execution_count": 22,
 753 |    "metadata": {},
 754 |    "outputs": [
 755 |     {
 756 |      "data": {
 757 |       "text/html": [
 758 |        "Total time: 51:19 <p><table style='width:300px; margin-bottom:10px'>\n",
 759 |        "  <tr>\n",
 760 |        "    <th>epoch</th>\n",
 761 |        "    <th>train_loss</th>\n",
 762 |        "    <th>valid_loss</th>\n",
 763 |        "    <th>accuracy</th>\n",
 764 |        "  </tr>\n",
 765 |        "  <tr>\n",
 766 |        "    <th>1</th>\n",
 767 |        "    <th>4.100102</th>\n",
 768 |        "    <th>3.264899</th>\n",
 769 |        "    <th>0.285417</th>\n",
 770 |        "  </tr>\n",
 771 |        "  <tr>\n",
 772 |        "    <th>2</th>\n",
 773 |        "    <th>2.793747</th>\n",
 774 |        "    <th>1.978519</th>\n",
 775 |        "    <th>0.517667</th>\n",
 776 |        "  </tr>\n",
 777 |        "  <tr>\n",
 778 |        "    <th>3</th>\n",
 779 |        "    <th>1.910982</th>\n",
 780 |        "    <th>1.300082</th>\n",
 781 |        "    <th>0.636833</th>\n",
 782 |        "  </tr>\n",
 783 |        "  <tr>\n",
 784 |        "    <th>4</th>\n",
 785 |        "    <th>1.455095</th>\n",
 786 |        "    <th>0.999775</th>\n",
 787 |        "    <th>0.702500</th>\n",
 788 |        "  </tr>\n",
 789 |        "  <tr>\n",
 790 |        "    <th>5</th>\n",
 791 |        "    <th>1.241369</th>\n",
 792 |        "    <th>0.867122</th>\n",
 793 |        "    <th>0.738250</th>\n",
 794 |        "  </tr>\n",
 795 |        "  <tr>\n",
 796 |        "    <th>6</th>\n",
 797 |        "    <th>1.109905</th>\n",
 798 |        "    <th>0.796427</th>\n",
 799 |        "    <th>0.754583</th>\n",
 800 |        "  </tr>\n",
 801 |        "  <tr>\n",
 802 |        "    <th>7</th>\n",
 803 |        "    <th>1.048591</th>\n",
 804 |        "    <th>0.770618</th>\n",
 805 |        "    <th>0.762250</th>\n",
 806 |        "  </tr>\n",
 807 |        "  <tr>\n",
 808 |        "    <th>8</th>\n",
 809 |        "    <th>0.979058</th>\n",
 810 |        "    <th>0.730820</th>\n",
 811 |        "    <th>0.773833</th>\n",
 812 |        "  </tr>\n",
 813 |        "  <tr>\n",
 814 |        "    <th>9</th>\n",
 815 |        "    <th>0.928035</th>\n",
 816 |        "    <th>0.709470</th>\n",
 817 |        "    <th>0.780250</th>\n",
 818 |        "  </tr>\n",
 819 |        "  <tr>\n",
 820 |        "    <th>10</th>\n",
 821 |        "    <th>0.890598</th>\n",
 822 |        "    <th>0.684087</th>\n",
 823 |        "    <th>0.786833</th>\n",
 824 |        "  </tr>\n",
 825 |        "  <tr>\n",
 826 |        "    <th>11</th>\n",
 827 |        "    <th>0.864331</th>\n",
 828 |        "    <th>0.679980</th>\n",
 829 |        "    <th>0.785333</th>\n",
 830 |        "  </tr>\n",
 831 |        "  <tr>\n",
 832 |        "    <th>12</th>\n",
 833 |        "    <th>0.813269</th>\n",
 834 |        "    <th>0.656503</th>\n",
 835 |        "    <th>0.796083</th>\n",
 836 |        "  </tr>\n",
 837 |        "  <tr>\n",
 838 |        "    <th>13</th>\n",
 839 |        "    <th>0.793111</th>\n",
 840 |        "    <th>0.645735</th>\n",
 841 |        "    <th>0.799333</th>\n",
 842 |        "  </tr>\n",
 843 |        "  <tr>\n",
 844 |        "    <th>14</th>\n",
 845 |        "    <th>0.758749</th>\n",
 846 |        "    <th>0.637417</th>\n",
 847 |        "    <th>0.799167</th>\n",
 848 |        "  </tr>\n",
 849 |        "  <tr>\n",
 850 |        "    <th>15</th>\n",
 851 |        "    <th>0.733792</th>\n",
 852 |        "    <th>0.624778</th>\n",
 853 |        "    <th>0.803500</th>\n",
 854 |        "  </tr>\n",
 855 |        "  <tr>\n",
 856 |        "    <th>16</th>\n",
 857 |        "    <th>0.713488</th>\n",
 858 |        "    <th>0.622848</th>\n",
 859 |        "    <th>0.804250</th>\n",
 860 |        "  </tr>\n",
 861 |        "  <tr>\n",
 862 |        "    <th>17</th>\n",
 863 |        "    <th>0.695918</th>\n",
 864 |        "    <th>0.618520</th>\n",
 865 |        "    <th>0.805333</th>\n",
 866 |        "  </tr>\n",
 867 |        "  <tr>\n",
 868 |        "    <th>18</th>\n",
 869 |        "    <th>0.670478</th>\n",
 870 |        "    <th>0.600957</th>\n",
 871 |        "    <th>0.812333</th>\n",
 872 |        "  </tr>\n",
 873 |        "  <tr>\n",
 874 |        "    <th>19</th>\n",
 875 |        "    <th>0.655789</th>\n",
 876 |        "    <th>0.600830</th>\n",
 877 |        "    <th>0.811750</th>\n",
 878 |        "  </tr>\n",
 879 |        "  <tr>\n",
 880 |        "    <th>20</th>\n",
 881 |        "    <th>0.638686</th>\n",
 882 |        "    <th>0.596192</th>\n",
 883 |        "    <th>0.814667</th>\n",
 884 |        "  </tr>\n",
 885 |        "  <tr>\n",
 886 |        "    <th>21</th>\n",
 887 |        "    <th>0.617020</th>\n",
 888 |        "    <th>0.593554</th>\n",
 889 |        "    <th>0.813500</th>\n",
 890 |        "  </tr>\n",
 891 |        "  <tr>\n",
 892 |        "    <th>22</th>\n",
 893 |        "    <th>0.598800</th>\n",
 894 |        "    <th>0.587362</th>\n",
 895 |        "    <th>0.814333</th>\n",
 896 |        "  </tr>\n",
 897 |        "  <tr>\n",
 898 |        "    <th>23</th>\n",
 899 |        "    <th>0.588485</th>\n",
 900 |        "    <th>0.585536</th>\n",
 901 |        "    <th>0.816500</th>\n",
 902 |        "  </tr>\n",
 903 |        "  <tr>\n",
 904 |        "    <th>24</th>\n",
 905 |        "    <th>0.560206</th>\n",
 906 |        "    <th>0.583934</th>\n",
 907 |        "    <th>0.818000</th>\n",
 908 |        "  </tr>\n",
 909 |        "  <tr>\n",
 910 |        "    <th>25</th>\n",
 911 |        "    <th>0.552242</th>\n",
 912 |        "    <th>0.579802</th>\n",
 913 |        "    <th>0.818750</th>\n",
 914 |        "  </tr>\n",
 915 |        "  <tr>\n",
 916 |        "    <th>26</th>\n",
 917 |        "    <th>0.544388</th>\n",
 918 |        "    <th>0.576987</th>\n",
 919 |        "    <th>0.820917</th>\n",
 920 |        "  </tr>\n",
 921 |        "  <tr>\n",
 922 |        "    <th>27</th>\n",
 923 |        "    <th>0.544016</th>\n",
 924 |        "    <th>0.577123</th>\n",
 925 |        "    <th>0.820667</th>\n",
 926 |        "  </tr>\n",
 927 |        "  <tr>\n",
 928 |        "    <th>28</th>\n",
 929 |        "    <th>0.544756</th>\n",
 930 |        "    <th>0.576735</th>\n",
 931 |        "    <th>0.819333</th>\n",
 932 |        "  </tr>\n",
 933 |        "  <tr>\n",
 934 |        "    <th>29</th>\n",
 935 |        "    <th>0.535235</th>\n",
 936 |        "    <th>0.576290</th>\n",
 937 |        "    <th>0.820167</th>\n",
 938 |        "  </tr>\n",
 939 |        "  <tr>\n",
 940 |        "    <th>30</th>\n",
 941 |        "    <th>0.533571</th>\n",
 942 |        "    <th>0.576551</th>\n",
 943 |        "    <th>0.819833</th>\n",
 944 |        "  </tr>\n",
 945 |        "</table>\n"
 946 |       ],
 947 |       "text/plain": [
 948 |        "<IPython.core.display.HTML object>"
 949 |       ]
 950 |      },
 951 |      "metadata": {},
 952 |      "output_type": "display_data"
 953 |     }
 954 |    ],
 955 |    "source": [
 956 |     "with gpu_mem_restore_ctx():\n",
 957 |     "    learn.fit_one_cycle(30)"
 958 |    ]
 959 |   },
 960 |   {
 961 |    "cell_type": "markdown",
 962 |    "metadata": {},
 963 |    "source": [
 964 |     "## Resnet 34 (Mixed Precision)"
 965 |    ]
 966 |   },
 967 |   {
 968 |    "cell_type": "code",
 969 |    "execution_count": 23,
 970 |    "metadata": {},
 971 |    "outputs": [],
 972 |    "source": [
 973 |     "np.random.seed(42)\n",
 974 |     "data = ImageDataBunch.from_folder(path, train=train, valid_pct=0.2,\n",
 975 |     "        ds_tfms=get_transforms(), size=224, num_workers=4, bs=496).normalize(cifar_stats)"
 976 |    ]
 977 |   },
 978 |   {
 979 |    "cell_type": "code",
 980 |    "execution_count": 24,
 981 |    "metadata": {},
 982 |    "outputs": [],
 983 |    "source": [
 984 |     "learn = to_fp16(create_cnn(data, models.resnet34, metrics=accuracy))"
 985 |    ]
 986 |   },
 987 |   {
 988 |    "cell_type": "code",
 989 |    "execution_count": 25,
 990 |    "metadata": {},
 991 |    "outputs": [
 992 |     {
 993 |      "data": {
 994 |       "text/html": [
 995 |        "Total time: 44:14 <p><table style='width:300px; margin-bottom:10px'>\n",
 996 |        "  <tr>\n",
 997 |        "    <th>epoch</th>\n",
 998 |        "    <th>train_loss</th>\n",
 999 |        "    <th>valid_loss</th>\n",
1000 |        "    <th>accuracy</th>\n",
1001 |        "  </tr>\n",
1002 |        "  <tr>\n",
1003 |        "    <th>1</th>\n",
1004 |        "    <th>4.522862</th>\n",
1005 |        "    <th>3.640730</th>\n",
1006 |        "    <th>0.200583</th>\n",
1007 |        "  </tr>\n",
1008 |        "  <tr>\n",
1009 |        "    <th>2</th>\n",
1010 |        "    <th>3.459070</th>\n",
1011 |        "    <th>2.374401</th>\n",
1012 |        "    <th>0.443583</th>\n",
1013 |        "  </tr>\n",
1014 |        "  <tr>\n",
1015 |        "    <th>3</th>\n",
1016 |        "    <th>2.469919</th>\n",
1017 |        "    <th>1.560894</th>\n",
1018 |        "    <th>0.582833</th>\n",
1019 |        "  </tr>\n",
1020 |        "  <tr>\n",
1021 |        "    <th>4</th>\n",
1022 |        "    <th>1.817846</th>\n",
1023 |        "    <th>1.165402</th>\n",
1024 |        "    <th>0.662000</th>\n",
1025 |        "  </tr>\n",
1026 |        "  <tr>\n",
1027 |        "    <th>5</th>\n",
1028 |        "    <th>1.449373</th>\n",
1029 |        "    <th>0.959060</th>\n",
1030 |        "    <th>0.714667</th>\n",
1031 |        "  </tr>\n",
1032 |        "  <tr>\n",
1033 |        "    <th>6</th>\n",
1034 |        "    <th>1.236690</th>\n",
1035 |        "    <th>0.863116</th>\n",
1036 |        "    <th>0.740500</th>\n",
1037 |        "  </tr>\n",
1038 |        "  <tr>\n",
1039 |        "    <th>7</th>\n",
1040 |        "    <th>1.113157</th>\n",
1041 |        "    <th>0.795895</th>\n",
1042 |        "    <th>0.756417</th>\n",
1043 |        "  </tr>\n",
1044 |        "  <tr>\n",
1045 |        "    <th>8</th>\n",
1046 |        "    <th>1.027290</th>\n",
1047 |        "    <th>0.765271</th>\n",
1048 |        "    <th>0.763417</th>\n",
1049 |        "  </tr>\n",
1050 |        "  <tr>\n",
1051 |        "    <th>9</th>\n",
1052 |        "    <th>0.967804</th>\n",
1053 |        "    <th>0.721662</th>\n",
1054 |        "    <th>0.774667</th>\n",
1055 |        "  </tr>\n",
1056 |        "  <tr>\n",
1057 |        "    <th>10</th>\n",
1058 |        "    <th>0.904266</th>\n",
1059 |        "    <th>0.705837</th>\n",
1060 |        "    <th>0.780333</th>\n",
1061 |        "  </tr>\n",
1062 |        "  <tr>\n",
1063 |        "    <th>11</th>\n",
1064 |        "    <th>0.865761</th>\n",
1065 |        "    <th>0.687858</th>\n",
1066 |        "    <th>0.786917</th>\n",
1067 |        "  </tr>\n",
1068 |        "  <tr>\n",
1069 |        "    <th>12</th>\n",
1070 |        "    <th>0.844098</th>\n",
1071 |        "    <th>0.668408</th>\n",
1072 |        "    <th>0.791083</th>\n",
1073 |        "  </tr>\n",
1074 |        "  <tr>\n",
1075 |        "    <th>13</th>\n",
1076 |        "    <th>0.802036</th>\n",
1077 |        "    <th>0.665113</th>\n",
1078 |        "    <th>0.791667</th>\n",
1079 |        "  </tr>\n",
1080 |        "  <tr>\n",
1081 |        "    <th>14</th>\n",
1082 |        "    <th>0.783970</th>\n",
1083 |        "    <th>0.649133</th>\n",
1084 |        "    <th>0.798250</th>\n",
1085 |        "  </tr>\n",
1086 |        "  <tr>\n",
1087 |        "    <th>15</th>\n",
1088 |        "    <th>0.745470</th>\n",
1089 |        "    <th>0.634941</th>\n",
1090 |        "    <th>0.805250</th>\n",
1091 |        "  </tr>\n",
1092 |        "  <tr>\n",
1093 |        "    <th>16</th>\n",
1094 |        "    <th>0.727000</th>\n",
1095 |        "    <th>0.629780</th>\n",
1096 |        "    <th>0.804333</th>\n",
1097 |        "  </tr>\n",
1098 |        "  <tr>\n",
1099 |        "    <th>17</th>\n",
1100 |        "    <th>0.709905</th>\n",
1101 |        "    <th>0.619154</th>\n",
1102 |        "    <th>0.805417</th>\n",
1103 |        "  </tr>\n",
1104 |        "  <tr>\n",
1105 |        "    <th>18</th>\n",
1106 |        "    <th>0.688805</th>\n",
1107 |        "    <th>0.614638</th>\n",
1108 |        "    <th>0.810250</th>\n",
1109 |        "  </tr>\n",
1110 |        "  <tr>\n",
1111 |        "    <th>19</th>\n",
1112 |        "    <th>0.673002</th>\n",
1113 |        "    <th>0.612835</th>\n",
1114 |        "    <th>0.808083</th>\n",
1115 |        "  </tr>\n",
1116 |        "  <tr>\n",
1117 |        "    <th>20</th>\n",
1118 |        "    <th>0.647279</th>\n",
1119 |        "    <th>0.606833</th>\n",
1120 |        "    <th>0.811917</th>\n",
1121 |        "  </tr>\n",
1122 |        "  <tr>\n",
1123 |        "    <th>21</th>\n",
1124 |        "    <th>0.630066</th>\n",
1125 |        "    <th>0.601341</th>\n",
1126 |        "    <th>0.813167</th>\n",
1127 |        "  </tr>\n",
1128 |        "  <tr>\n",
1129 |        "    <th>22</th>\n",
1130 |        "    <th>0.615739</th>\n",
1131 |        "    <th>0.599353</th>\n",
1132 |        "    <th>0.814083</th>\n",
1133 |        "  </tr>\n",
1134 |        "  <tr>\n",
1135 |        "    <th>23</th>\n",
1136 |        "    <th>0.600889</th>\n",
1137 |        "    <th>0.592636</th>\n",
1138 |        "    <th>0.815833</th>\n",
1139 |        "  </tr>\n",
1140 |        "  <tr>\n",
1141 |        "    <th>24</th>\n",
1142 |        "    <th>0.595916</th>\n",
1143 |        "    <th>0.593515</th>\n",
1144 |        "    <th>0.817667</th>\n",
1145 |        "  </tr>\n",
1146 |        "  <tr>\n",
1147 |        "    <th>25</th>\n",
1148 |        "    <th>0.578139</th>\n",
1149 |        "    <th>0.589500</th>\n",
1150 |        "    <th>0.815500</th>\n",
1151 |        "  </tr>\n",
1152 |        "  <tr>\n",
1153 |        "    <th>26</th>\n",
1154 |        "    <th>0.576346</th>\n",
1155 |        "    <th>0.590978</th>\n",
1156 |        "    <th>0.815750</th>\n",
1157 |        "  </tr>\n",
1158 |        "  <tr>\n",
1159 |        "    <th>27</th>\n",
1160 |        "    <th>0.571478</th>\n",
1161 |        "    <th>0.586962</th>\n",
1162 |        "    <th>0.818833</th>\n",
1163 |        "  </tr>\n",
1164 |        "  <tr>\n",
1165 |        "    <th>28</th>\n",
1166 |        "    <th>0.560052</th>\n",
1167 |        "    <th>0.587171</th>\n",
1168 |        "    <th>0.818000</th>\n",
1169 |        "  </tr>\n",
1170 |        "  <tr>\n",
1171 |        "    <th>29</th>\n",
1172 |        "    <th>0.556342</th>\n",
1173 |        "    <th>0.586621</th>\n",
1174 |        "    <th>0.818083</th>\n",
1175 |        "  </tr>\n",
1176 |        "  <tr>\n",
1177 |        "    <th>30</th>\n",
1178 |        "    <th>0.552511</th>\n",
1179 |        "    <th>0.587052</th>\n",
1180 |        "    <th>0.818333</th>\n",
1181 |        "  </tr>\n",
1182 |        "</table>\n"
1183 |       ],
1184 |       "text/plain": [
1185 |        "<IPython.core.display.HTML object>"
1186 |       ]
1187 |      },
1188 |      "metadata": {},
1189 |      "output_type": "display_data"
1190 |     }
1191 |    ],
1192 |    "source": [
1193 |     "with gpu_mem_restore_ctx():\n",
1194 |     "    learn.fit_one_cycle(30)"
1195 |    ]
1196 |   },
1197 |   {
1198 |    "cell_type": "markdown",
1199 |    "metadata": {},
1200 |    "source": [
1201 |     "## Resnet 50"
1202 |    ]
1203 |   },
1204 |   {
1205 |    "cell_type": "code",
1206 |    "execution_count": 26,
1207 |    "metadata": {},
1208 |    "outputs": [],
1209 |    "source": [
1210 |     "#bs = 512"
1211 |    ]
1212 |   },
1213 |   {
1214 |    "cell_type": "code",
1215 |    "execution_count": 27,
1216 |    "metadata": {},
1217 |    "outputs": [],
1218 |    "source": [
1219 |     "np.random.seed(42)\n",
1220 |     "data = ImageDataBunch.from_folder(path, train=train, valid_pct=0.2,\n",
1221 |     "        ds_tfms=get_transforms(), size=224, num_workers=4, bs=98).normalize(cifar_stats)"
1222 |    ]
1223 |   },
1224 |   {
1225 |    "cell_type": "code",
1226 |    "execution_count": 28,
1227 |    "metadata": {},
1228 |    "outputs": [
1229 |     {
1230 |      "name": "stderr",
1231 |      "output_type": "stream",
1232 |      "text": [
1233 |       "Downloading: \"https://download.pytorch.org/models/resnet50-19c8e357.pth\" to /home/ekami/.torch/models/resnet50-19c8e357.pth\n",
1234 |       "100%|██████████| 102502400/102502400 [00:19<00:00, 5150001.64it/s]\n"
1235 |      ]
1236 |     }
1237 |    ],
1238 |    "source": [
1239 |     "learn = create_cnn(data, models.resnet50, metrics=accuracy)"
1240 |    ]
1241 |   },
1242 |   {
1243 |    "cell_type": "code",
1244 |    "execution_count": 29,
1245 |    "metadata": {},
1246 |    "outputs": [
1247 |     {
1248 |      "data": {
1249 |       "text/html": [
1250 |        "Total time: 1:45:22 <p><table style='width:300px; margin-bottom:10px'>\n",
1251 |        "  <tr>\n",
1252 |        "    <th>epoch</th>\n",
1253 |        "    <th>train_loss</th>\n",
1254 |        "    <th>valid_loss</th>\n",
1255 |        "    <th>accuracy</th>\n",
1256 |        "  </tr>\n",
1257 |        "  <tr>\n",
1258 |        "    <th>1</th>\n",
1259 |        "    <th>2.515773</th>\n",
1260 |        "    <th>1.886690</th>\n",
1261 |        "    <th>0.541167</th>\n",
1262 |        "  </tr>\n",
1263 |        "  <tr>\n",
1264 |        "    <th>2</th>\n",
1265 |        "    <th>1.696823</th>\n",
1266 |        "    <th>1.202539</th>\n",
1267 |        "    <th>0.663750</th>\n",
1268 |        "  </tr>\n",
1269 |        "  <tr>\n",
1270 |        "    <th>3</th>\n",
1271 |        "    <th>1.346115</th>\n",
1272 |        "    <th>0.940126</th>\n",
1273 |        "    <th>0.719500</th>\n",
1274 |        "  </tr>\n",
1275 |        "  <tr>\n",
1276 |        "    <th>4</th>\n",
1277 |        "    <th>1.105000</th>\n",
1278 |        "    <th>0.807197</th>\n",
1279 |        "    <th>0.751083</th>\n",
1280 |        "  </tr>\n",
1281 |        "  <tr>\n",
1282 |        "    <th>5</th>\n",
1283 |        "    <th>1.012399</th>\n",
1284 |        "    <th>0.757759</th>\n",
1285 |        "    <th>0.765250</th>\n",
1286 |        "  </tr>\n",
1287 |        "  <tr>\n",
1288 |        "    <th>6</th>\n",
1289 |        "    <th>0.941379</th>\n",
1290 |        "    <th>0.734523</th>\n",
1291 |        "    <th>0.770833</th>\n",
1292 |        "  </tr>\n",
1293 |        "  <tr>\n",
1294 |        "    <th>7</th>\n",
1295 |        "    <th>0.891324</th>\n",
1296 |        "    <th>0.690761</th>\n",
1297 |        "    <th>0.783500</th>\n",
1298 |        "  </tr>\n",
1299 |        "  <tr>\n",
1300 |        "    <th>8</th>\n",
1301 |        "    <th>0.819676</th>\n",
1302 |        "    <th>0.679745</th>\n",
1303 |        "    <th>0.790583</th>\n",
1304 |        "  </tr>\n",
1305 |        "  <tr>\n",
1306 |        "    <th>9</th>\n",
1307 |        "    <th>0.785965</th>\n",
1308 |        "    <th>0.657922</th>\n",
1309 |        "    <th>0.794000</th>\n",
1310 |        "  </tr>\n",
1311 |        "  <tr>\n",
1312 |        "    <th>10</th>\n",
1313 |        "    <th>0.737616</th>\n",
1314 |        "    <th>0.631472</th>\n",
1315 |        "    <th>0.806583</th>\n",
1316 |        "  </tr>\n",
1317 |        "  <tr>\n",
1318 |        "    <th>11</th>\n",
1319 |        "    <th>0.716565</th>\n",
1320 |        "    <th>0.626581</th>\n",
1321 |        "    <th>0.805500</th>\n",
1322 |        "  </tr>\n",
1323 |        "  <tr>\n",
1324 |        "    <th>12</th>\n",
1325 |        "    <th>0.703448</th>\n",
1326 |        "    <th>0.611971</th>\n",
1327 |        "    <th>0.810333</th>\n",
1328 |        "  </tr>\n",
1329 |        "  <tr>\n",
1330 |        "    <th>13</th>\n",
1331 |        "    <th>0.643908</th>\n",
1332 |        "    <th>0.610946</th>\n",
1333 |        "    <th>0.809333</th>\n",
1334 |        "  </tr>\n",
1335 |        "  <tr>\n",
1336 |        "    <th>14</th>\n",
1337 |        "    <th>0.619197</th>\n",
1338 |        "    <th>0.596491</th>\n",
1339 |        "    <th>0.818167</th>\n",
1340 |        "  </tr>\n",
1341 |        "  <tr>\n",
1342 |        "    <th>15</th>\n",
1343 |        "    <th>0.588613</th>\n",
1344 |        "    <th>0.582370</th>\n",
1345 |        "    <th>0.821167</th>\n",
1346 |        "  </tr>\n",
1347 |        "  <tr>\n",
1348 |        "    <th>16</th>\n",
1349 |        "    <th>0.572277</th>\n",
1350 |        "    <th>0.584467</th>\n",
1351 |        "    <th>0.823833</th>\n",
1352 |        "  </tr>\n",
1353 |        "  <tr>\n",
1354 |        "    <th>17</th>\n",
1355 |        "    <th>0.537526</th>\n",
1356 |        "    <th>0.581618</th>\n",
1357 |        "    <th>0.819917</th>\n",
1358 |        "  </tr>\n",
1359 |        "  <tr>\n",
1360 |        "    <th>18</th>\n",
1361 |        "    <th>0.520785</th>\n",
1362 |        "    <th>0.573473</th>\n",
1363 |        "    <th>0.824500</th>\n",
1364 |        "  </tr>\n",
1365 |        "  <tr>\n",
1366 |        "    <th>19</th>\n",
1367 |        "    <th>0.479418</th>\n",
1368 |        "    <th>0.570315</th>\n",
1369 |        "    <th>0.828000</th>\n",
1370 |        "  </tr>\n",
1371 |        "  <tr>\n",
1372 |        "    <th>20</th>\n",
1373 |        "    <th>0.462298</th>\n",
1374 |        "    <th>0.565643</th>\n",
1375 |        "    <th>0.826917</th>\n",
1376 |        "  </tr>\n",
1377 |        "  <tr>\n",
1378 |        "    <th>21</th>\n",
1379 |        "    <th>0.447708</th>\n",
1380 |        "    <th>0.562472</th>\n",
1381 |        "    <th>0.830500</th>\n",
1382 |        "  </tr>\n",
1383 |        "  <tr>\n",
1384 |        "    <th>22</th>\n",
1385 |        "    <th>0.413717</th>\n",
1386 |        "    <th>0.560359</th>\n",
1387 |        "    <th>0.833417</th>\n",
1388 |        "  </tr>\n",
1389 |        "  <tr>\n",
1390 |        "    <th>23</th>\n",
1391 |        "    <th>0.376912</th>\n",
1392 |        "    <th>0.557939</th>\n",
1393 |        "    <th>0.834167</th>\n",
1394 |        "  </tr>\n",
1395 |        "  <tr>\n",
1396 |        "    <th>24</th>\n",
1397 |        "    <th>0.372460</th>\n",
1398 |        "    <th>0.559411</th>\n",
1399 |        "    <th>0.835167</th>\n",
1400 |        "  </tr>\n",
1401 |        "  <tr>\n",
1402 |        "    <th>25</th>\n",
1403 |        "    <th>0.367571</th>\n",
1404 |        "    <th>0.556380</th>\n",
1405 |        "    <th>0.837167</th>\n",
1406 |        "  </tr>\n",
1407 |        "  <tr>\n",
1408 |        "    <th>26</th>\n",
1409 |        "    <th>0.347707</th>\n",
1410 |        "    <th>0.554721</th>\n",
1411 |        "    <th>0.837667</th>\n",
1412 |        "  </tr>\n",
1413 |        "  <tr>\n",
1414 |        "    <th>27</th>\n",
1415 |        "    <th>0.334381</th>\n",
1416 |        "    <th>0.551504</th>\n",
1417 |        "    <th>0.838417</th>\n",
1418 |        "  </tr>\n",
1419 |        "  <tr>\n",
1420 |        "    <th>28</th>\n",
1421 |        "    <th>0.351300</th>\n",
1422 |        "    <th>0.549640</th>\n",
1423 |        "    <th>0.838250</th>\n",
1424 |        "  </tr>\n",
1425 |        "  <tr>\n",
1426 |        "    <th>29</th>\n",
1427 |        "    <th>0.335910</th>\n",
1428 |        "    <th>0.549918</th>\n",
1429 |        "    <th>0.837000</th>\n",
1430 |        "  </tr>\n",
1431 |        "  <tr>\n",
1432 |        "    <th>30</th>\n",
1433 |        "    <th>0.340140</th>\n",
1434 |        "    <th>0.553000</th>\n",
1435 |        "    <th>0.836417</th>\n",
1436 |        "  </tr>\n",
1437 |        "</table>\n"
1438 |       ],
1439 |       "text/plain": [
1440 |        "<IPython.core.display.HTML object>"
1441 |       ]
1442 |      },
1443 |      "metadata": {},
1444 |      "output_type": "display_data"
1445 |     }
1446 |    ],
1447 |    "source": [
1448 |     "with gpu_mem_restore_ctx():\n",
1449 |     "    learn.fit_one_cycle(30)"
1450 |    ]
1451 |   },
1452 |   {
1453 |    "cell_type": "markdown",
1454 |    "metadata": {},
1455 |    "source": [
1456 |     "## Resnet 50 (Mixed Precision)"
1457 |    ]
1458 |   },
1459 |   {
1460 |    "cell_type": "code",
1461 |    "execution_count": 30,
1462 |    "metadata": {},
1463 |    "outputs": [],
1464 |    "source": [
1465 |     "np.random.seed(42)\n",
1466 |     "data = ImageDataBunch.from_folder(path, train=train, valid_pct=0.2,\n",
1467 |     "        ds_tfms=get_transforms(), size=224, num_workers=4, bs=164).normalize(cifar_stats)"
1468 |    ]
1469 |   },
1470 |   {
1471 |    "cell_type": "code",
1472 |    "execution_count": 31,
1473 |    "metadata": {},
1474 |    "outputs": [],
1475 |    "source": [
1476 |     "learn = to_fp16(create_cnn(data, models.resnet50, metrics=accuracy))"
1477 |    ]
1478 |   },
1479 |   {
1480 |    "cell_type": "code",
1481 |    "execution_count": 32,
1482 |    "metadata": {},
1483 |    "outputs": [
1484 |     {
1485 |      "data": {
1486 |       "text/html": [
1487 |        "Total time: 1:24:52 <p><table style='width:300px; margin-bottom:10px'>\n",
1488 |        "  <tr>\n",
1489 |        "    <th>epoch</th>\n",
1490 |        "    <th>train_loss</th>\n",
1491 |        "    <th>valid_loss</th>\n",
1492 |        "    <th>accuracy</th>\n",
1493 |        "  </tr>\n",
1494 |        "  <tr>\n",
1495 |        "    <th>1</th>\n",
1496 |        "    <th>2.834446</th>\n",
1497 |        "    <th>2.072671</th>\n",
1498 |        "    <th>0.511667</th>\n",
1499 |        "  </tr>\n",
1500 |        "  <tr>\n",
1501 |        "    <th>2</th>\n",
1502 |        "    <th>1.883764</th>\n",
1503 |        "    <th>1.311698</th>\n",
1504 |        "    <th>0.639583</th>\n",
1505 |        "  </tr>\n",
1506 |        "  <tr>\n",
1507 |        "    <th>3</th>\n",
1508 |        "    <th>1.395265</th>\n",
1509 |        "    <th>0.999120</th>\n",
1510 |        "    <th>0.704250</th>\n",
1511 |        "  </tr>\n",
1512 |        "  <tr>\n",
1513 |        "    <th>4</th>\n",
1514 |        "    <th>1.161582</th>\n",
1515 |        "    <th>0.848280</th>\n",
1516 |        "    <th>0.743583</th>\n",
1517 |        "  </tr>\n",
1518 |        "  <tr>\n",
1519 |        "    <th>5</th>\n",
1520 |        "    <th>1.024427</th>\n",
1521 |        "    <th>0.772155</th>\n",
1522 |        "    <th>0.759583</th>\n",
1523 |        "  </tr>\n",
1524 |        "  <tr>\n",
1525 |        "    <th>6</th>\n",
1526 |        "    <th>0.940368</th>\n",
1527 |        "    <th>0.736896</th>\n",
1528 |        "    <th>0.770417</th>\n",
1529 |        "  </tr>\n",
1530 |        "  <tr>\n",
1531 |        "    <th>7</th>\n",
1532 |        "    <th>0.892585</th>\n",
1533 |        "    <th>0.707394</th>\n",
1534 |        "    <th>0.776083</th>\n",
1535 |        "  </tr>\n",
1536 |        "  <tr>\n",
1537 |        "    <th>8</th>\n",
1538 |        "    <th>0.829782</th>\n",
1539 |        "    <th>0.685450</th>\n",
1540 |        "    <th>0.790000</th>\n",
1541 |        "  </tr>\n",
1542 |        "  <tr>\n",
1543 |        "    <th>9</th>\n",
1544 |        "    <th>0.780883</th>\n",
1545 |        "    <th>0.662696</th>\n",
1546 |        "    <th>0.798583</th>\n",
1547 |        "  </tr>\n",
1548 |        "  <tr>\n",
1549 |        "    <th>10</th>\n",
1550 |        "    <th>0.755269</th>\n",
1551 |        "    <th>0.638769</th>\n",
1552 |        "    <th>0.802083</th>\n",
1553 |        "  </tr>\n",
1554 |        "  <tr>\n",
1555 |        "    <th>11</th>\n",
1556 |        "    <th>0.695746</th>\n",
1557 |        "    <th>0.623059</th>\n",
1558 |        "    <th>0.809167</th>\n",
1559 |        "  </tr>\n",
1560 |        "  <tr>\n",
1561 |        "    <th>12</th>\n",
1562 |        "    <th>0.674482</th>\n",
1563 |        "    <th>0.617339</th>\n",
1564 |        "    <th>0.806833</th>\n",
1565 |        "  </tr>\n",
1566 |        "  <tr>\n",
1567 |        "    <th>13</th>\n",
1568 |        "    <th>0.614091</th>\n",
1569 |        "    <th>0.608190</th>\n",
1570 |        "    <th>0.814583</th>\n",
1571 |        "  </tr>\n",
1572 |        "  <tr>\n",
1573 |        "    <th>14</th>\n",
1574 |        "    <th>0.609434</th>\n",
1575 |        "    <th>0.598253</th>\n",
1576 |        "    <th>0.812250</th>\n",
1577 |        "  </tr>\n",
1578 |        "  <tr>\n",
1579 |        "    <th>15</th>\n",
1580 |        "    <th>0.561976</th>\n",
1581 |        "    <th>0.589217</th>\n",
1582 |        "    <th>0.818333</th>\n",
1583 |        "  </tr>\n",
1584 |        "  <tr>\n",
1585 |        "    <th>16</th>\n",
1586 |        "    <th>0.544676</th>\n",
1587 |        "    <th>0.583386</th>\n",
1588 |        "    <th>0.821833</th>\n",
1589 |        "  </tr>\n",
1590 |        "  <tr>\n",
1591 |        "    <th>17</th>\n",
1592 |        "    <th>0.503154</th>\n",
1593 |        "    <th>0.581094</th>\n",
1594 |        "    <th>0.821750</th>\n",
1595 |        "  </tr>\n",
1596 |        "  <tr>\n",
1597 |        "    <th>18</th>\n",
1598 |        "    <th>0.489639</th>\n",
1599 |        "    <th>0.572993</th>\n",
1600 |        "    <th>0.823667</th>\n",
1601 |        "  </tr>\n",
1602 |        "  <tr>\n",
1603 |        "    <th>19</th>\n",
1604 |        "    <th>0.469337</th>\n",
1605 |        "    <th>0.574861</th>\n",
1606 |        "    <th>0.826917</th>\n",
1607 |        "  </tr>\n",
1608 |        "  <tr>\n",
1609 |        "    <th>20</th>\n",
1610 |        "    <th>0.445635</th>\n",
1611 |        "    <th>0.570555</th>\n",
1612 |        "    <th>0.829500</th>\n",
1613 |        "  </tr>\n",
1614 |        "  <tr>\n",
1615 |        "    <th>21</th>\n",
1616 |        "    <th>0.421942</th>\n",
1617 |        "    <th>0.570226</th>\n",
1618 |        "    <th>0.830333</th>\n",
1619 |        "  </tr>\n",
1620 |        "  <tr>\n",
1621 |        "    <th>22</th>\n",
1622 |        "    <th>0.395235</th>\n",
1623 |        "    <th>0.558666</th>\n",
1624 |        "    <th>0.831833</th>\n",
1625 |        "  </tr>\n",
1626 |        "  <tr>\n",
1627 |        "    <th>23</th>\n",
1628 |        "    <th>0.376425</th>\n",
1629 |        "    <th>0.563551</th>\n",
1630 |        "    <th>0.835083</th>\n",
1631 |        "  </tr>\n",
1632 |        "  <tr>\n",
1633 |        "    <th>24</th>\n",
1634 |        "    <th>0.366422</th>\n",
1635 |        "    <th>0.559915</th>\n",
1636 |        "    <th>0.835250</th>\n",
1637 |        "  </tr>\n",
1638 |        "  <tr>\n",
1639 |        "    <th>25</th>\n",
1640 |        "    <th>0.344141</th>\n",
1641 |        "    <th>0.557956</th>\n",
1642 |        "    <th>0.835000</th>\n",
1643 |        "  </tr>\n",
1644 |        "  <tr>\n",
1645 |        "    <th>26</th>\n",
1646 |        "    <th>0.328442</th>\n",
1647 |        "    <th>0.559216</th>\n",
1648 |        "    <th>0.835250</th>\n",
1649 |        "  </tr>\n",
1650 |        "  <tr>\n",
1651 |        "    <th>27</th>\n",
1652 |        "    <th>0.322673</th>\n",
1653 |        "    <th>0.556995</th>\n",
1654 |        "    <th>0.835167</th>\n",
1655 |        "  </tr>\n",
1656 |        "  <tr>\n",
1657 |        "    <th>28</th>\n",
1658 |        "    <th>0.324144</th>\n",
1659 |        "    <th>0.557831</th>\n",
1660 |        "    <th>0.835250</th>\n",
1661 |        "  </tr>\n",
1662 |        "  <tr>\n",
1663 |        "    <th>29</th>\n",
1664 |        "    <th>0.320713</th>\n",
1665 |        "    <th>0.555050</th>\n",
1666 |        "    <th>0.836333</th>\n",
1667 |        "  </tr>\n",
1668 |        "  <tr>\n",
1669 |        "    <th>30</th>\n",
1670 |        "    <th>0.327241</th>\n",
1671 |        "    <th>0.557459</th>\n",
1672 |        "    <th>0.836417</th>\n",
1673 |        "  </tr>\n",
1674 |        "</table>\n"
1675 |       ],
1676 |       "text/plain": [
1677 |        "<IPython.core.display.HTML object>"
1678 |       ]
1679 |      },
1680 |      "metadata": {},
1681 |      "output_type": "display_data"
1682 |     }
1683 |    ],
1684 |    "source": [
1685 |     "with gpu_mem_restore_ctx():\n",
1686 |     "    learn.fit_one_cycle(30)"
1687 |    ]
1688 |   },
1689 |   {
1690 |    "cell_type": "markdown",
1691 |    "metadata": {},
1692 |    "source": [
1693 |     "## Resnet 101"
1694 |    ]
1695 |   },
1696 |   {
1697 |    "cell_type": "code",
1698 |    "execution_count": 33,
1699 |    "metadata": {},
1700 |    "outputs": [],
1701 |    "source": [
1702 |     "np.random.seed(42)\n",
1703 |     "data = ImageDataBunch.from_folder(path, train=train, valid_pct=0.2,\n",
1704 |     "        ds_tfms=get_transforms(), size=224, num_workers=4, bs=64).normalize(cifar_stats)"
1705 |    ]
1706 |   },
1707 |   {
1708 |    "cell_type": "code",
1709 |    "execution_count": 34,
1710 |    "metadata": {},
1711 |    "outputs": [
1712 |     {
1713 |      "name": "stderr",
1714 |      "output_type": "stream",
1715 |      "text": [
1716 |       "Downloading: \"https://download.pytorch.org/models/resnet101-5d3b4d8f.pth\" to /home/ekami/.torch/models/resnet101-5d3b4d8f.pth\n",
1717 |       "100%|██████████| 178728960/178728960 [00:29<00:00, 6127228.47it/s]\n"
1718 |      ]
1719 |     }
1720 |    ],
1721 |    "source": [
1722 |     "learn = create_cnn(data, models.resnet101, metrics=accuracy)"
1723 |    ]
1724 |   },
1725 |   {
1726 |    "cell_type": "code",
1727 |    "execution_count": 35,
1728 |    "metadata": {},
1729 |    "outputs": [
1730 |     {
1731 |      "data": {
1732 |       "text/html": [
1733 |        "Total time: 2:36:27 <p><table style='width:300px; margin-bottom:10px'>\n",
1734 |        "  <tr>\n",
1735 |        "    <th>epoch</th>\n",
1736 |        "    <th>train_loss</th>\n",
1737 |        "    <th>valid_loss</th>\n",
1738 |        "    <th>accuracy</th>\n",
1739 |        "  </tr>\n",
1740 |        "  <tr>\n",
1741 |        "    <th>1</th>\n",
1742 |        "    <th>2.083457</th>\n",
1743 |        "    <th>1.451530</th>\n",
1744 |        "    <th>0.638333</th>\n",
1745 |        "  </tr>\n",
1746 |        "  <tr>\n",
1747 |        "    <th>2</th>\n",
1748 |        "    <th>1.321773</th>\n",
1749 |        "    <th>0.908532</th>\n",
1750 |        "    <th>0.733333</th>\n",
1751 |        "  </tr>\n",
1752 |        "  <tr>\n",
1753 |        "    <th>3</th>\n",
1754 |        "    <th>1.074472</th>\n",
1755 |        "    <th>0.728461</th>\n",
1756 |        "    <th>0.775750</th>\n",
1757 |        "  </tr>\n",
1758 |        "  <tr>\n",
1759 |        "    <th>4</th>\n",
1760 |        "    <th>0.985465</th>\n",
1761 |        "    <th>0.666169</th>\n",
1762 |        "    <th>0.793667</th>\n",
1763 |        "  </tr>\n",
1764 |        "  <tr>\n",
1765 |        "    <th>5</th>\n",
1766 |        "    <th>0.838859</th>\n",
1767 |        "    <th>0.618405</th>\n",
1768 |        "    <th>0.806750</th>\n",
1769 |        "  </tr>\n",
1770 |        "  <tr>\n",
1771 |        "    <th>6</th>\n",
1772 |        "    <th>0.803868</th>\n",
1773 |        "    <th>0.600917</th>\n",
1774 |        "    <th>0.812500</th>\n",
1775 |        "  </tr>\n",
1776 |        "  <tr>\n",
1777 |        "    <th>7</th>\n",
1778 |        "    <th>0.770296</th>\n",
1779 |        "    <th>0.592337</th>\n",
1780 |        "    <th>0.814750</th>\n",
1781 |        "  </tr>\n",
1782 |        "  <tr>\n",
1783 |        "    <th>8</th>\n",
1784 |        "    <th>0.712573</th>\n",
1785 |        "    <th>0.587856</th>\n",
1786 |        "    <th>0.818417</th>\n",
1787 |        "  </tr>\n",
1788 |        "  <tr>\n",
1789 |        "    <th>9</th>\n",
1790 |        "    <th>0.700638</th>\n",
1791 |        "    <th>0.557925</th>\n",
1792 |        "    <th>0.827667</th>\n",
1793 |        "  </tr>\n",
1794 |        "  <tr>\n",
1795 |        "    <th>10</th>\n",
1796 |        "    <th>0.656998</th>\n",
1797 |        "    <th>0.543478</th>\n",
1798 |        "    <th>0.833500</th>\n",
1799 |        "  </tr>\n",
1800 |        "  <tr>\n",
1801 |        "    <th>11</th>\n",
1802 |        "    <th>0.587333</th>\n",
1803 |        "    <th>0.537264</th>\n",
1804 |        "    <th>0.833167</th>\n",
1805 |        "  </tr>\n",
1806 |        "  <tr>\n",
1807 |        "    <th>12</th>\n",
1808 |        "    <th>0.576772</th>\n",
1809 |        "    <th>0.523525</th>\n",
1810 |        "    <th>0.844917</th>\n",
1811 |        "  </tr>\n",
1812 |        "  <tr>\n",
1813 |        "    <th>13</th>\n",
1814 |        "    <th>0.546539</th>\n",
1815 |        "    <th>0.511617</th>\n",
1816 |        "    <th>0.843750</th>\n",
1817 |        "  </tr>\n",
1818 |        "  <tr>\n",
1819 |        "    <th>14</th>\n",
1820 |        "    <th>0.531197</th>\n",
1821 |        "    <th>0.519306</th>\n",
1822 |        "    <th>0.839500</th>\n",
1823 |        "  </tr>\n",
1824 |        "  <tr>\n",
1825 |        "    <th>15</th>\n",
1826 |        "    <th>0.467777</th>\n",
1827 |        "    <th>0.507615</th>\n",
1828 |        "    <th>0.845667</th>\n",
1829 |        "  </tr>\n",
1830 |        "  <tr>\n",
1831 |        "    <th>16</th>\n",
1832 |        "    <th>0.468377</th>\n",
1833 |        "    <th>0.504536</th>\n",
1834 |        "    <th>0.848250</th>\n",
1835 |        "  </tr>\n",
1836 |        "  <tr>\n",
1837 |        "    <th>17</th>\n",
1838 |        "    <th>0.454281</th>\n",
1839 |        "    <th>0.497895</th>\n",
1840 |        "    <th>0.849583</th>\n",
1841 |        "  </tr>\n",
1842 |        "  <tr>\n",
1843 |        "    <th>18</th>\n",
1844 |        "    <th>0.420571</th>\n",
1845 |        "    <th>0.491803</th>\n",
1846 |        "    <th>0.852833</th>\n",
1847 |        "  </tr>\n",
1848 |        "  <tr>\n",
1849 |        "    <th>19</th>\n",
1850 |        "    <th>0.401252</th>\n",
1851 |        "    <th>0.486161</th>\n",
1852 |        "    <th>0.853583</th>\n",
1853 |        "  </tr>\n",
1854 |        "  <tr>\n",
1855 |        "    <th>20</th>\n",
1856 |        "    <th>0.359169</th>\n",
1857 |        "    <th>0.492218</th>\n",
1858 |        "    <th>0.853583</th>\n",
1859 |        "  </tr>\n",
1860 |        "  <tr>\n",
1861 |        "    <th>21</th>\n",
1862 |        "    <th>0.353262</th>\n",
1863 |        "    <th>0.479224</th>\n",
1864 |        "    <th>0.857250</th>\n",
1865 |        "  </tr>\n",
1866 |        "  <tr>\n",
1867 |        "    <th>22</th>\n",
1868 |        "    <th>0.317846</th>\n",
1869 |        "    <th>0.482238</th>\n",
1870 |        "    <th>0.860833</th>\n",
1871 |        "  </tr>\n",
1872 |        "  <tr>\n",
1873 |        "    <th>23</th>\n",
1874 |        "    <th>0.305395</th>\n",
1875 |        "    <th>0.482287</th>\n",
1876 |        "    <th>0.860000</th>\n",
1877 |        "  </tr>\n",
1878 |        "  <tr>\n",
1879 |        "    <th>24</th>\n",
1880 |        "    <th>0.302323</th>\n",
1881 |        "    <th>0.477879</th>\n",
1882 |        "    <th>0.861250</th>\n",
1883 |        "  </tr>\n",
1884 |        "  <tr>\n",
1885 |        "    <th>25</th>\n",
1886 |        "    <th>0.285679</th>\n",
1887 |        "    <th>0.477347</th>\n",
1888 |        "    <th>0.863667</th>\n",
1889 |        "  </tr>\n",
1890 |        "  <tr>\n",
1891 |        "    <th>26</th>\n",
1892 |        "    <th>0.265691</th>\n",
1893 |        "    <th>0.472855</th>\n",
1894 |        "    <th>0.864000</th>\n",
1895 |        "  </tr>\n",
1896 |        "  <tr>\n",
1897 |        "    <th>27</th>\n",
1898 |        "    <th>0.235859</th>\n",
1899 |        "    <th>0.476183</th>\n",
1900 |        "    <th>0.865417</th>\n",
1901 |        "  </tr>\n",
1902 |        "  <tr>\n",
1903 |        "    <th>28</th>\n",
1904 |        "    <th>0.253295</th>\n",
1905 |        "    <th>0.475753</th>\n",
1906 |        "    <th>0.864083</th>\n",
1907 |        "  </tr>\n",
1908 |        "  <tr>\n",
1909 |        "    <th>29</th>\n",
1910 |        "    <th>0.252339</th>\n",
1911 |        "    <th>0.476051</th>\n",
1912 |        "    <th>0.865167</th>\n",
1913 |        "  </tr>\n",
1914 |        "  <tr>\n",
1915 |        "    <th>30</th>\n",
1916 |        "    <th>0.238589</th>\n",
1917 |        "    <th>0.474209</th>\n",
1918 |        "    <th>0.863583</th>\n",
1919 |        "  </tr>\n",
1920 |        "</table>\n"
1921 |       ],
1922 |       "text/plain": [
1923 |        "<IPython.core.display.HTML object>"
1924 |       ]
1925 |      },
1926 |      "metadata": {},
1927 |      "output_type": "display_data"
1928 |     }
1929 |    ],
1930 |    "source": [
1931 |     "with gpu_mem_restore_ctx():\n",
1932 |     "    learn.fit_one_cycle(30)"
1933 |    ]
1934 |   },
1935 |   {
1936 |    "cell_type": "markdown",
1937 |    "metadata": {},
1938 |    "source": [
1939 |     "## Resnet 101 (Mixed Precision)"
1940 |    ]
1941 |   },
1942 |   {
1943 |    "cell_type": "code",
1944 |    "execution_count": 36,
1945 |    "metadata": {},
1946 |    "outputs": [],
1947 |    "source": [
1948 |     "np.random.seed(42)\n",
1949 |     "data = ImageDataBunch.from_folder(path, train=train, valid_pct=0.2,\n",
1950 |     "        ds_tfms=get_transforms(), size=224, num_workers=4, bs=116).normalize(cifar_stats)"
1951 |    ]
1952 |   },
1953 |   {
1954 |    "cell_type": "code",
1955 |    "execution_count": 37,
1956 |    "metadata": {},
1957 |    "outputs": [],
1958 |    "source": [
1959 |     "learn = to_fp16(create_cnn(data, models.resnet101, metrics=accuracy))"
1960 |    ]
1961 |   },
1962 |   {
1963 |    "cell_type": "code",
1964 |    "execution_count": 38,
1965 |    "metadata": {},
1966 |    "outputs": [
1967 |     {
1968 |      "data": {
1969 |       "text/html": [
1970 |        "Total time: 2:08:00 <p><table style='width:300px; margin-bottom:10px'>\n",
1971 |        "  <tr>\n",
1972 |        "    <th>epoch</th>\n",
1973 |        "    <th>train_loss</th>\n",
1974 |        "    <th>valid_loss</th>\n",
1975 |        "    <th>accuracy</th>\n",
1976 |        "  </tr>\n",
1977 |        "  <tr>\n",
1978 |        "    <th>1</th>\n",
1979 |        "    <th>2.402573</th>\n",
1980 |        "    <th>1.641406</th>\n",
1981 |        "    <th>0.603583</th>\n",
1982 |        "  </tr>\n",
1983 |        "  <tr>\n",
1984 |        "    <th>2</th>\n",
1985 |        "    <th>1.470369</th>\n",
1986 |        "    <th>0.982278</th>\n",
1987 |        "    <th>0.721500</th>\n",
1988 |        "  </tr>\n",
1989 |        "  <tr>\n",
1990 |        "    <th>3</th>\n",
1991 |        "    <th>1.102205</th>\n",
1992 |        "    <th>0.770284</th>\n",
1993 |        "    <th>0.762917</th>\n",
1994 |        "  </tr>\n",
1995 |        "  <tr>\n",
1996 |        "    <th>4</th>\n",
1997 |        "    <th>0.931413</th>\n",
1998 |        "    <th>0.675970</th>\n",
1999 |        "    <th>0.792833</th>\n",
2000 |        "  </tr>\n",
2001 |        "  <tr>\n",
2002 |        "    <th>5</th>\n",
2003 |        "    <th>0.843089</th>\n",
2004 |        "    <th>0.631572</th>\n",
2005 |        "    <th>0.804000</th>\n",
2006 |        "  </tr>\n",
2007 |        "  <tr>\n",
2008 |        "    <th>6</th>\n",
2009 |        "    <th>0.773870</th>\n",
2010 |        "    <th>0.607783</th>\n",
2011 |        "    <th>0.812583</th>\n",
2012 |        "  </tr>\n",
2013 |        "  <tr>\n",
2014 |        "    <th>7</th>\n",
2015 |        "    <th>0.747394</th>\n",
2016 |        "    <th>0.579102</th>\n",
2017 |        "    <th>0.823000</th>\n",
2018 |        "  </tr>\n",
2019 |        "  <tr>\n",
2020 |        "    <th>8</th>\n",
2021 |        "    <th>0.698355</th>\n",
2022 |        "    <th>0.569311</th>\n",
2023 |        "    <th>0.822667</th>\n",
2024 |        "  </tr>\n",
2025 |        "  <tr>\n",
2026 |        "    <th>9</th>\n",
2027 |        "    <th>0.665380</th>\n",
2028 |        "    <th>0.569775</th>\n",
2029 |        "    <th>0.824250</th>\n",
2030 |        "  </tr>\n",
2031 |        "  <tr>\n",
2032 |        "    <th>10</th>\n",
2033 |        "    <th>0.601234</th>\n",
2034 |        "    <th>0.540831</th>\n",
2035 |        "    <th>0.833250</th>\n",
2036 |        "  </tr>\n",
2037 |        "  <tr>\n",
2038 |        "    <th>11</th>\n",
2039 |        "    <th>0.583828</th>\n",
2040 |        "    <th>0.515747</th>\n",
2041 |        "    <th>0.842083</th>\n",
2042 |        "  </tr>\n",
2043 |        "  <tr>\n",
2044 |        "    <th>12</th>\n",
2045 |        "    <th>0.515766</th>\n",
2046 |        "    <th>0.514028</th>\n",
2047 |        "    <th>0.841167</th>\n",
2048 |        "  </tr>\n",
2049 |        "  <tr>\n",
2050 |        "    <th>13</th>\n",
2051 |        "    <th>0.501062</th>\n",
2052 |        "    <th>0.508455</th>\n",
2053 |        "    <th>0.844500</th>\n",
2054 |        "  </tr>\n",
2055 |        "  <tr>\n",
2056 |        "    <th>14</th>\n",
2057 |        "    <th>0.475348</th>\n",
2058 |        "    <th>0.509687</th>\n",
2059 |        "    <th>0.845667</th>\n",
2060 |        "  </tr>\n",
2061 |        "  <tr>\n",
2062 |        "    <th>15</th>\n",
2063 |        "    <th>0.463935</th>\n",
2064 |        "    <th>0.501683</th>\n",
2065 |        "    <th>0.846500</th>\n",
2066 |        "  </tr>\n",
2067 |        "  <tr>\n",
2068 |        "    <th>16</th>\n",
2069 |        "    <th>0.433365</th>\n",
2070 |        "    <th>0.493446</th>\n",
2071 |        "    <th>0.852083</th>\n",
2072 |        "  </tr>\n",
2073 |        "  <tr>\n",
2074 |        "    <th>17</th>\n",
2075 |        "    <th>0.399650</th>\n",
2076 |        "    <th>0.496941</th>\n",
2077 |        "    <th>0.854250</th>\n",
2078 |        "  </tr>\n",
2079 |        "  <tr>\n",
2080 |        "    <th>18</th>\n",
2081 |        "    <th>0.380342</th>\n",
2082 |        "    <th>0.498303</th>\n",
2083 |        "    <th>0.853250</th>\n",
2084 |        "  </tr>\n",
2085 |        "  <tr>\n",
2086 |        "    <th>19</th>\n",
2087 |        "    <th>0.350620</th>\n",
2088 |        "    <th>0.483819</th>\n",
2089 |        "    <th>0.857833</th>\n",
2090 |        "  </tr>\n",
2091 |        "  <tr>\n",
2092 |        "    <th>20</th>\n",
2093 |        "    <th>0.329624</th>\n",
2094 |        "    <th>0.486034</th>\n",
2095 |        "    <th>0.855667</th>\n",
2096 |        "  </tr>\n",
2097 |        "  <tr>\n",
2098 |        "    <th>21</th>\n",
2099 |        "    <th>0.320623</th>\n",
2100 |        "    <th>0.480329</th>\n",
2101 |        "    <th>0.858750</th>\n",
2102 |        "  </tr>\n",
2103 |        "  <tr>\n",
2104 |        "    <th>22</th>\n",
2105 |        "    <th>0.299029</th>\n",
2106 |        "    <th>0.477023</th>\n",
2107 |        "    <th>0.859917</th>\n",
2108 |        "  </tr>\n",
2109 |        "  <tr>\n",
2110 |        "    <th>23</th>\n",
2111 |        "    <th>0.264857</th>\n",
2112 |        "    <th>0.476886</th>\n",
2113 |        "    <th>0.863000</th>\n",
2114 |        "  </tr>\n",
2115 |        "  <tr>\n",
2116 |        "    <th>24</th>\n",
2117 |        "    <th>0.261844</th>\n",
2118 |        "    <th>0.477507</th>\n",
2119 |        "    <th>0.864083</th>\n",
2120 |        "  </tr>\n",
2121 |        "  <tr>\n",
2122 |        "    <th>25</th>\n",
2123 |        "    <th>0.254985</th>\n",
2124 |        "    <th>0.476712</th>\n",
2125 |        "    <th>0.864250</th>\n",
2126 |        "  </tr>\n",
2127 |        "  <tr>\n",
2128 |        "    <th>26</th>\n",
2129 |        "    <th>0.255878</th>\n",
2130 |        "    <th>0.476127</th>\n",
2131 |        "    <th>0.864417</th>\n",
2132 |        "  </tr>\n",
2133 |        "  <tr>\n",
2134 |        "    <th>27</th>\n",
2135 |        "    <th>0.231718</th>\n",
2136 |        "    <th>0.474823</th>\n",
2137 |        "    <th>0.866250</th>\n",
2138 |        "  </tr>\n",
2139 |        "  <tr>\n",
2140 |        "    <th>28</th>\n",
2141 |        "    <th>0.221085</th>\n",
2142 |        "    <th>0.473201</th>\n",
2143 |        "    <th>0.866667</th>\n",
2144 |        "  </tr>\n",
2145 |        "  <tr>\n",
2146 |        "    <th>29</th>\n",
2147 |        "    <th>0.221620</th>\n",
2148 |        "    <th>0.471682</th>\n",
2149 |        "    <th>0.865083</th>\n",
2150 |        "  </tr>\n",
2151 |        "  <tr>\n",
2152 |        "    <th>30</th>\n",
2153 |        "    <th>0.234261</th>\n",
2154 |        "    <th>0.476006</th>\n",
2155 |        "    <th>0.865417</th>\n",
2156 |        "  </tr>\n",
2157 |        "</table>\n"
2158 |       ],
2159 |       "text/plain": [
2160 |        "<IPython.core.display.HTML object>"
2161 |       ]
2162 |      },
2163 |      "metadata": {},
2164 |      "output_type": "display_data"
2165 |     }
2166 |    ],
2167 |    "source": [
2168 |     "with gpu_mem_restore_ctx():\n",
2169 |     "    learn.fit_one_cycle(30)"
2170 |    ]
2171 |   }
2172 |  ],
2173 |  "metadata": {
2174 |   "kernelspec": {
2175 |    "display_name": "Python 3",
2176 |    "language": "python",
2177 |    "name": "python3"
2178 |   },
2179 |   "language_info": {
2180 |    "codemirror_mode": {
2181 |     "name": "ipython",
2182 |     "version": 3
2183 |    },
2184 |    "file_extension": ".py",
2185 |    "mimetype": "text/x-python",
2186 |    "name": "python",
2187 |    "nbconvert_exporter": "python",
2188 |    "pygments_lexer": "ipython3",
2189 |    "version": "3.6.6"
2190 |   }
2191 |  },
2192 |  "nbformat": 4,
2193 |  "nbformat_minor": 2
2194 | }
2195 | 


--------------------------------------------------------------------------------
/2080Ti.ipynb:
--------------------------------------------------------------------------------
   1 | {
   2 |  "cells": [
   3 |   {
   4 |    "cell_type": "code",
   5 |    "execution_count": 1,
   6 |    "metadata": {},
   7 |    "outputs": [],
   8 |    "source": [
   9 |     "%reload_ext autoreload\n",
  10 |     "%autoreload 2\n",
  11 |     "%matplotlib inline"
  12 |    ]
  13 |   },
  14 |   {
  15 |    "cell_type": "code",
  16 |    "execution_count": 2,
  17 |    "metadata": {},
  18 |    "outputs": [],
  19 |    "source": [
  20 |     "import functools\n",
  21 |     "import traceback\n",
  22 |     "def get_ref_free_exc_info():\n",
  23 |     "    \"Free traceback from references to locals/globals to avoid circular reference leading to gc.collect() unable to reclaim memory\"\n",
  24 |     "    type, val, tb = sys.exc_info()\n",
  25 |     "    traceback.clear_frames(tb)\n",
  26 |     "    return (type, val, tb)\n",
  27 |     "\n",
  28 |     "def gpu_mem_restore(func):\n",
  29 |     "    \"Reclaim GPU RAM if CUDA out of memory happened, or execution was interrupted\"\n",
  30 |     "    @functools.wraps(func)\n",
  31 |     "    def wrapper(*args, **kwargs):\n",
  32 |     "        try:\n",
  33 |     "            return func(*args, **kwargs)\n",
  34 |     "        except:\n",
  35 |     "            type, val, tb = get_ref_free_exc_info() # must!\n",
  36 |     "            raise type(val).with_traceback(tb) from None\n",
  37 |     "    return wrapper"
  38 |    ]
  39 |   },
  40 |   {
  41 |    "cell_type": "code",
  42 |    "execution_count": 3,
  43 |    "metadata": {},
  44 |    "outputs": [],
  45 |    "source": [
  46 |     "from fastai.vision import *\n",
  47 |     "from fastai.metrics import error_rate"
  48 |    ]
  49 |   },
  50 |   {
  51 |    "cell_type": "code",
  52 |    "execution_count": 4,
  53 |    "metadata": {},
  54 |    "outputs": [],
  55 |    "source": [
  56 |     "path = Path('/home/init27/Downloads/cifar100/')"
  57 |    ]
  58 |   },
  59 |   {
  60 |    "cell_type": "code",
  61 |    "execution_count": 5,
  62 |    "metadata": {},
  63 |    "outputs": [],
  64 |    "source": [
  65 |     "train = path/'train'\n",
  66 |     "test = path/'test'"
  67 |    ]
  68 |   },
  69 |   {
  70 |    "cell_type": "code",
  71 |    "execution_count": 8,
  72 |    "metadata": {},
  73 |    "outputs": [],
  74 |    "source": [
  75 |     "bs = 108"
  76 |    ]
  77 |   },
  78 |   {
  79 |    "cell_type": "code",
  80 |    "execution_count": 6,
  81 |    "metadata": {},
  82 |    "outputs": [
  83 |     {
  84 |      "data": {
  85 |       "text/plain": [
  86 |        "[PosixPath('/home/init27/Downloads/cifar100/models'),\n",
  87 |        " PosixPath('/home/init27/Downloads/cifar100/train'),\n",
  88 |        " PosixPath('/home/init27/Downloads/cifar100/test')]"
  89 |       ]
  90 |      },
  91 |      "execution_count": 6,
  92 |      "metadata": {},
  93 |      "output_type": "execute_result"
  94 |     }
  95 |    ],
  96 |    "source": [
  97 |     "path.ls()"
  98 |    ]
  99 |   },
 100 |   {
 101 |    "cell_type": "code",
 102 |    "execution_count": 9,
 103 |    "metadata": {},
 104 |    "outputs": [],
 105 |    "source": [
 106 |     "np.random.seed(42)\n",
 107 |     "data = ImageDataBunch.from_folder(path, train=train, valid_pct=0.2,\n",
 108 |     "        ds_tfms=get_transforms(), size=224, num_workers=4, bs=bs).normalize(cifar_stats)"
 109 |    ]
 110 |   },
 111 |   {
 112 |    "cell_type": "code",
 113 |    "execution_count": 10,
 114 |    "metadata": {},
 115 |    "outputs": [],
 116 |    "source": [
 117 |     "class gpu_mem_restore_ctx():\n",
 118 |     "    \" context manager to reclaim GPU RAM if CUDA out of memory happened, or execution was interrupted\"\n",
 119 |     "    def __enter__(self): return self\n",
 120 |     "    def __exit__(self, exc_type, exc_val, exc_tb):\n",
 121 |     "        if not exc_val: return True\n",
 122 |     "        traceback.clear_frames(exc_tb)\n",
 123 |     "        raise exc_type(exc_val).with_traceback(exc_tb) from None"
 124 |    ]
 125 |   },
 126 |   {
 127 |    "cell_type": "code",
 128 |    "execution_count": 11,
 129 |    "metadata": {},
 130 |    "outputs": [],
 131 |    "source": [
 132 |     "#Allow crashing\n",
 133 |     "learn = create_cnn(data, models.resnet18 , metrics=accuracy)"
 134 |    ]
 135 |   },
 136 |   {
 137 |    "cell_type": "markdown",
 138 |    "metadata": {},
 139 |    "source": [
 140 |     "## Resnet 18\n",
 141 |     "\n"
 142 |    ]
 143 |   },
 144 |   {
 145 |    "cell_type": "code",
 146 |    "execution_count": 17,
 147 |    "metadata": {},
 148 |    "outputs": [],
 149 |    "source": [
 150 |     "np.random.seed(42)\n",
 151 |     "data = ImageDataBunch.from_folder(path, train=train, valid_pct=0.2,\n",
 152 |     "        ds_tfms=get_transforms(), size=224, num_workers=4, bs=408).normalize(cifar_stats)\n",
 153 |     "learn = create_cnn(data, models.resnet18, metrics=accuracy)"
 154 |    ]
 155 |   },
 156 |   {
 157 |    "cell_type": "code",
 158 |    "execution_count": 18,
 159 |    "metadata": {},
 160 |    "outputs": [
 161 |     {
 162 |      "data": {
 163 |       "text/html": [
 164 |        "Total time: 00:54 <p><table style='width:300px; margin-bottom:10px'>\n",
 165 |        "  <tr>\n",
 166 |        "    <th>epoch</th>\n",
 167 |        "    <th>train_loss</th>\n",
 168 |        "    <th>valid_loss</th>\n",
 169 |        "    <th>accuracy</th>\n",
 170 |        "  </tr>\n",
 171 |        "  <tr>\n",
 172 |        "    <th>1</th>\n",
 173 |        "    <th>2.425544</th>\n",
 174 |        "    <th>1.632482</th>\n",
 175 |        "    <th>0.559083</th>\n",
 176 |        "  </tr>\n",
 177 |        "</table>\n"
 178 |       ],
 179 |       "text/plain": [
 180 |        "<IPython.core.display.HTML object>"
 181 |       ]
 182 |      },
 183 |      "metadata": {},
 184 |      "output_type": "display_data"
 185 |     }
 186 |    ],
 187 |    "source": [
 188 |     "with gpu_mem_restore_ctx():\n",
 189 |     "    learn.fit_one_cycle(1)"
 190 |    ]
 191 |   },
 192 |   {
 193 |    "cell_type": "code",
 194 |    "execution_count": 19,
 195 |    "metadata": {},
 196 |    "outputs": [
 197 |     {
 198 |      "data": {
 199 |       "text/html": [
 200 |        "Total time: 28:12 <p><table style='width:300px; margin-bottom:10px'>\n",
 201 |        "  <tr>\n",
 202 |        "    <th>epoch</th>\n",
 203 |        "    <th>train_loss</th>\n",
 204 |        "    <th>valid_loss</th>\n",
 205 |        "    <th>accuracy</th>\n",
 206 |        "  </tr>\n",
 207 |        "  <tr>\n",
 208 |        "    <th>1</th>\n",
 209 |        "    <th>1.683162</th>\n",
 210 |        "    <th>1.234888</th>\n",
 211 |        "    <th>0.647083</th>\n",
 212 |        "  </tr>\n",
 213 |        "  <tr>\n",
 214 |        "    <th>2</th>\n",
 215 |        "    <th>1.377793</th>\n",
 216 |        "    <th>1.006688</th>\n",
 217 |        "    <th>0.698833</th>\n",
 218 |        "  </tr>\n",
 219 |        "  <tr>\n",
 220 |        "    <th>3</th>\n",
 221 |        "    <th>1.210670</th>\n",
 222 |        "    <th>0.953726</th>\n",
 223 |        "    <th>0.711000</th>\n",
 224 |        "  </tr>\n",
 225 |        "  <tr>\n",
 226 |        "    <th>4</th>\n",
 227 |        "    <th>1.161485</th>\n",
 228 |        "    <th>0.981398</th>\n",
 229 |        "    <th>0.702750</th>\n",
 230 |        "  </tr>\n",
 231 |        "  <tr>\n",
 232 |        "    <th>5</th>\n",
 233 |        "    <th>1.158831</th>\n",
 234 |        "    <th>0.985615</th>\n",
 235 |        "    <th>0.702833</th>\n",
 236 |        "  </tr>\n",
 237 |        "  <tr>\n",
 238 |        "    <th>6</th>\n",
 239 |        "    <th>1.168251</th>\n",
 240 |        "    <th>1.045662</th>\n",
 241 |        "    <th>0.683667</th>\n",
 242 |        "  </tr>\n",
 243 |        "  <tr>\n",
 244 |        "    <th>7</th>\n",
 245 |        "    <th>1.150126</th>\n",
 246 |        "    <th>1.071930</th>\n",
 247 |        "    <th>0.681250</th>\n",
 248 |        "  </tr>\n",
 249 |        "  <tr>\n",
 250 |        "    <th>8</th>\n",
 251 |        "    <th>1.146979</th>\n",
 252 |        "    <th>1.025527</th>\n",
 253 |        "    <th>0.692417</th>\n",
 254 |        "  </tr>\n",
 255 |        "  <tr>\n",
 256 |        "    <th>9</th>\n",
 257 |        "    <th>1.109232</th>\n",
 258 |        "    <th>1.041167</th>\n",
 259 |        "    <th>0.691250</th>\n",
 260 |        "  </tr>\n",
 261 |        "  <tr>\n",
 262 |        "    <th>10</th>\n",
 263 |        "    <th>1.088920</th>\n",
 264 |        "    <th>1.046639</th>\n",
 265 |        "    <th>0.685083</th>\n",
 266 |        "  </tr>\n",
 267 |        "  <tr>\n",
 268 |        "    <th>11</th>\n",
 269 |        "    <th>1.055113</th>\n",
 270 |        "    <th>1.009589</th>\n",
 271 |        "    <th>0.702417</th>\n",
 272 |        "  </tr>\n",
 273 |        "  <tr>\n",
 274 |        "    <th>12</th>\n",
 275 |        "    <th>1.027755</th>\n",
 276 |        "    <th>0.937168</th>\n",
 277 |        "    <th>0.716500</th>\n",
 278 |        "  </tr>\n",
 279 |        "  <tr>\n",
 280 |        "    <th>13</th>\n",
 281 |        "    <th>1.003785</th>\n",
 282 |        "    <th>0.953145</th>\n",
 283 |        "    <th>0.715917</th>\n",
 284 |        "  </tr>\n",
 285 |        "  <tr>\n",
 286 |        "    <th>14</th>\n",
 287 |        "    <th>0.968761</th>\n",
 288 |        "    <th>0.916469</th>\n",
 289 |        "    <th>0.725417</th>\n",
 290 |        "  </tr>\n",
 291 |        "  <tr>\n",
 292 |        "    <th>15</th>\n",
 293 |        "    <th>0.932007</th>\n",
 294 |        "    <th>0.907113</th>\n",
 295 |        "    <th>0.728667</th>\n",
 296 |        "  </tr>\n",
 297 |        "  <tr>\n",
 298 |        "    <th>16</th>\n",
 299 |        "    <th>0.899529</th>\n",
 300 |        "    <th>0.874359</th>\n",
 301 |        "    <th>0.738333</th>\n",
 302 |        "  </tr>\n",
 303 |        "  <tr>\n",
 304 |        "    <th>17</th>\n",
 305 |        "    <th>0.876744</th>\n",
 306 |        "    <th>0.819828</th>\n",
 307 |        "    <th>0.748167</th>\n",
 308 |        "  </tr>\n",
 309 |        "  <tr>\n",
 310 |        "    <th>18</th>\n",
 311 |        "    <th>0.853781</th>\n",
 312 |        "    <th>0.813430</th>\n",
 313 |        "    <th>0.750667</th>\n",
 314 |        "  </tr>\n",
 315 |        "  <tr>\n",
 316 |        "    <th>19</th>\n",
 317 |        "    <th>0.821074</th>\n",
 318 |        "    <th>0.790245</th>\n",
 319 |        "    <th>0.762583</th>\n",
 320 |        "  </tr>\n",
 321 |        "  <tr>\n",
 322 |        "    <th>20</th>\n",
 323 |        "    <th>0.793510</th>\n",
 324 |        "    <th>0.764171</th>\n",
 325 |        "    <th>0.768833</th>\n",
 326 |        "  </tr>\n",
 327 |        "  <tr>\n",
 328 |        "    <th>21</th>\n",
 329 |        "    <th>0.766965</th>\n",
 330 |        "    <th>0.755508</th>\n",
 331 |        "    <th>0.770500</th>\n",
 332 |        "  </tr>\n",
 333 |        "  <tr>\n",
 334 |        "    <th>22</th>\n",
 335 |        "    <th>0.738198</th>\n",
 336 |        "    <th>0.737558</th>\n",
 337 |        "    <th>0.777167</th>\n",
 338 |        "  </tr>\n",
 339 |        "  <tr>\n",
 340 |        "    <th>23</th>\n",
 341 |        "    <th>0.711821</th>\n",
 342 |        "    <th>0.725124</th>\n",
 343 |        "    <th>0.779583</th>\n",
 344 |        "  </tr>\n",
 345 |        "  <tr>\n",
 346 |        "    <th>24</th>\n",
 347 |        "    <th>0.680861</th>\n",
 348 |        "    <th>0.722985</th>\n",
 349 |        "    <th>0.778667</th>\n",
 350 |        "  </tr>\n",
 351 |        "  <tr>\n",
 352 |        "    <th>25</th>\n",
 353 |        "    <th>0.663084</th>\n",
 354 |        "    <th>0.706659</th>\n",
 355 |        "    <th>0.783667</th>\n",
 356 |        "  </tr>\n",
 357 |        "  <tr>\n",
 358 |        "    <th>26</th>\n",
 359 |        "    <th>0.644237</th>\n",
 360 |        "    <th>0.696470</th>\n",
 361 |        "    <th>0.788083</th>\n",
 362 |        "  </tr>\n",
 363 |        "  <tr>\n",
 364 |        "    <th>27</th>\n",
 365 |        "    <th>0.619859</th>\n",
 366 |        "    <th>0.691847</th>\n",
 367 |        "    <th>0.790667</th>\n",
 368 |        "  </tr>\n",
 369 |        "  <tr>\n",
 370 |        "    <th>28</th>\n",
 371 |        "    <th>0.609522</th>\n",
 372 |        "    <th>0.691935</th>\n",
 373 |        "    <th>0.788750</th>\n",
 374 |        "  </tr>\n",
 375 |        "  <tr>\n",
 376 |        "    <th>29</th>\n",
 377 |        "    <th>0.597701</th>\n",
 378 |        "    <th>0.690438</th>\n",
 379 |        "    <th>0.789417</th>\n",
 380 |        "  </tr>\n",
 381 |        "  <tr>\n",
 382 |        "    <th>30</th>\n",
 383 |        "    <th>0.605155</th>\n",
 384 |        "    <th>0.690853</th>\n",
 385 |        "    <th>0.788500</th>\n",
 386 |        "  </tr>\n",
 387 |        "</table>\n"
 388 |       ],
 389 |       "text/plain": [
 390 |        "<IPython.core.display.HTML object>"
 391 |       ]
 392 |      },
 393 |      "metadata": {},
 394 |      "output_type": "display_data"
 395 |     }
 396 |    ],
 397 |    "source": [
 398 |     "with gpu_mem_restore_ctx():\n",
 399 |     "    learn.fit_one_cycle(30, max_lr=1e-2)"
 400 |    ]
 401 |   },
 402 |   {
 403 |    "cell_type": "markdown",
 404 |    "metadata": {},
 405 |    "source": [
 406 |     "## Resnet 18 (Mixed Precision)"
 407 |    ]
 408 |   },
 409 |   {
 410 |    "cell_type": "code",
 411 |    "execution_count": 29,
 412 |    "metadata": {},
 413 |    "outputs": [],
 414 |    "source": [
 415 |     "np.random.seed(42)\n",
 416 |     "data = ImageDataBunch.from_folder(path, train=train, valid_pct=0.2,\n",
 417 |     "        ds_tfms=get_transforms(), size=224, num_workers=4, bs=720).normalize(cifar_stats)"
 418 |    ]
 419 |   },
 420 |   {
 421 |    "cell_type": "code",
 422 |    "execution_count": 30,
 423 |    "metadata": {},
 424 |    "outputs": [],
 425 |    "source": [
 426 |     "learn = to_fp16(create_cnn(data, models.resnet18, metrics=accuracy))"
 427 |    ]
 428 |   },
 429 |   {
 430 |    "cell_type": "code",
 431 |    "execution_count": 31,
 432 |    "metadata": {},
 433 |    "outputs": [
 434 |     {
 435 |      "data": {
 436 |       "text/html": [
 437 |        "Total time: 00:58 <p><table style='width:300px; margin-bottom:10px'>\n",
 438 |        "  <tr>\n",
 439 |        "    <th>epoch</th>\n",
 440 |        "    <th>train_loss</th>\n",
 441 |        "    <th>valid_loss</th>\n",
 442 |        "    <th>accuracy</th>\n",
 443 |        "  </tr>\n",
 444 |        "  <tr>\n",
 445 |        "    <th>1</th>\n",
 446 |        "    <th>2.892367</th>\n",
 447 |        "    <th>1.857484</th>\n",
 448 |        "    <th>0.514833</th>\n",
 449 |        "  </tr>\n",
 450 |        "</table>\n"
 451 |       ],
 452 |       "text/plain": [
 453 |        "<IPython.core.display.HTML object>"
 454 |       ]
 455 |      },
 456 |      "metadata": {},
 457 |      "output_type": "display_data"
 458 |     }
 459 |    ],
 460 |    "source": [
 461 |     "with gpu_mem_restore_ctx():\n",
 462 |     "    learn.fit_one_cycle(1)"
 463 |    ]
 464 |   },
 465 |   {
 466 |    "cell_type": "code",
 467 |    "execution_count": 32,
 468 |    "metadata": {},
 469 |    "outputs": [
 470 |     {
 471 |      "data": {
 472 |       "text/html": [
 473 |        "Total time: 28:28 <p><table style='width:300px; margin-bottom:10px'>\n",
 474 |        "  <tr>\n",
 475 |        "    <th>epoch</th>\n",
 476 |        "    <th>train_loss</th>\n",
 477 |        "    <th>valid_loss</th>\n",
 478 |        "    <th>accuracy</th>\n",
 479 |        "  </tr>\n",
 480 |        "  <tr>\n",
 481 |        "    <th>1</th>\n",
 482 |        "    <th>2.217803</th>\n",
 483 |        "    <th>1.776256</th>\n",
 484 |        "    <th>0.535000</th>\n",
 485 |        "  </tr>\n",
 486 |        "  <tr>\n",
 487 |        "    <th>2</th>\n",
 488 |        "    <th>2.111644</th>\n",
 489 |        "    <th>1.634846</th>\n",
 490 |        "    <th>0.565083</th>\n",
 491 |        "  </tr>\n",
 492 |        "  <tr>\n",
 493 |        "    <th>3</th>\n",
 494 |        "    <th>1.948603</th>\n",
 495 |        "    <th>1.451506</th>\n",
 496 |        "    <th>0.600167</th>\n",
 497 |        "  </tr>\n",
 498 |        "  <tr>\n",
 499 |        "    <th>4</th>\n",
 500 |        "    <th>1.763333</th>\n",
 501 |        "    <th>1.288680</th>\n",
 502 |        "    <th>0.634083</th>\n",
 503 |        "  </tr>\n",
 504 |        "  <tr>\n",
 505 |        "    <th>5</th>\n",
 506 |        "    <th>1.576218</th>\n",
 507 |        "    <th>1.170622</th>\n",
 508 |        "    <th>0.659167</th>\n",
 509 |        "  </tr>\n",
 510 |        "  <tr>\n",
 511 |        "    <th>6</th>\n",
 512 |        "    <th>1.435181</th>\n",
 513 |        "    <th>1.088699</th>\n",
 514 |        "    <th>0.677000</th>\n",
 515 |        "  </tr>\n",
 516 |        "  <tr>\n",
 517 |        "    <th>7</th>\n",
 518 |        "    <th>1.335717</th>\n",
 519 |        "    <th>1.024443</th>\n",
 520 |        "    <th>0.695833</th>\n",
 521 |        "  </tr>\n",
 522 |        "  <tr>\n",
 523 |        "    <th>8</th>\n",
 524 |        "    <th>1.241808</th>\n",
 525 |        "    <th>0.974325</th>\n",
 526 |        "    <th>0.710167</th>\n",
 527 |        "  </tr>\n",
 528 |        "  <tr>\n",
 529 |        "    <th>9</th>\n",
 530 |        "    <th>1.171949</th>\n",
 531 |        "    <th>0.934963</th>\n",
 532 |        "    <th>0.718083</th>\n",
 533 |        "  </tr>\n",
 534 |        "  <tr>\n",
 535 |        "    <th>10</th>\n",
 536 |        "    <th>1.118036</th>\n",
 537 |        "    <th>0.914327</th>\n",
 538 |        "    <th>0.726500</th>\n",
 539 |        "  </tr>\n",
 540 |        "  <tr>\n",
 541 |        "    <th>11</th>\n",
 542 |        "    <th>1.073453</th>\n",
 543 |        "    <th>0.887545</th>\n",
 544 |        "    <th>0.731083</th>\n",
 545 |        "  </tr>\n",
 546 |        "  <tr>\n",
 547 |        "    <th>12</th>\n",
 548 |        "    <th>1.032388</th>\n",
 549 |        "    <th>0.872336</th>\n",
 550 |        "    <th>0.735083</th>\n",
 551 |        "  </tr>\n",
 552 |        "  <tr>\n",
 553 |        "    <th>13</th>\n",
 554 |        "    <th>0.991246</th>\n",
 555 |        "    <th>0.859043</th>\n",
 556 |        "    <th>0.737417</th>\n",
 557 |        "  </tr>\n",
 558 |        "  <tr>\n",
 559 |        "    <th>14</th>\n",
 560 |        "    <th>0.968865</th>\n",
 561 |        "    <th>0.838764</th>\n",
 562 |        "    <th>0.744333</th>\n",
 563 |        "  </tr>\n",
 564 |        "  <tr>\n",
 565 |        "    <th>15</th>\n",
 566 |        "    <th>0.941147</th>\n",
 567 |        "    <th>0.827038</th>\n",
 568 |        "    <th>0.748500</th>\n",
 569 |        "  </tr>\n",
 570 |        "  <tr>\n",
 571 |        "    <th>16</th>\n",
 572 |        "    <th>0.921168</th>\n",
 573 |        "    <th>0.815047</th>\n",
 574 |        "    <th>0.750500</th>\n",
 575 |        "  </tr>\n",
 576 |        "  <tr>\n",
 577 |        "    <th>17</th>\n",
 578 |        "    <th>0.896468</th>\n",
 579 |        "    <th>0.810671</th>\n",
 580 |        "    <th>0.750750</th>\n",
 581 |        "  </tr>\n",
 582 |        "  <tr>\n",
 583 |        "    <th>18</th>\n",
 584 |        "    <th>0.876352</th>\n",
 585 |        "    <th>0.797368</th>\n",
 586 |        "    <th>0.752500</th>\n",
 587 |        "  </tr>\n",
 588 |        "  <tr>\n",
 589 |        "    <th>19</th>\n",
 590 |        "    <th>0.855625</th>\n",
 591 |        "    <th>0.796557</th>\n",
 592 |        "    <th>0.754833</th>\n",
 593 |        "  </tr>\n",
 594 |        "  <tr>\n",
 595 |        "    <th>20</th>\n",
 596 |        "    <th>0.834282</th>\n",
 597 |        "    <th>0.787105</th>\n",
 598 |        "    <th>0.758667</th>\n",
 599 |        "  </tr>\n",
 600 |        "  <tr>\n",
 601 |        "    <th>21</th>\n",
 602 |        "    <th>0.824325</th>\n",
 603 |        "    <th>0.783010</th>\n",
 604 |        "    <th>0.761167</th>\n",
 605 |        "  </tr>\n",
 606 |        "  <tr>\n",
 607 |        "    <th>22</th>\n",
 608 |        "    <th>0.799274</th>\n",
 609 |        "    <th>0.776148</th>\n",
 610 |        "    <th>0.762583</th>\n",
 611 |        "  </tr>\n",
 612 |        "  <tr>\n",
 613 |        "    <th>23</th>\n",
 614 |        "    <th>0.790784</th>\n",
 615 |        "    <th>0.770834</th>\n",
 616 |        "    <th>0.761250</th>\n",
 617 |        "  </tr>\n",
 618 |        "  <tr>\n",
 619 |        "    <th>24</th>\n",
 620 |        "    <th>0.774112</th>\n",
 621 |        "    <th>0.768528</th>\n",
 622 |        "    <th>0.763750</th>\n",
 623 |        "  </tr>\n",
 624 |        "  <tr>\n",
 625 |        "    <th>25</th>\n",
 626 |        "    <th>0.767463</th>\n",
 627 |        "    <th>0.766201</th>\n",
 628 |        "    <th>0.765917</th>\n",
 629 |        "  </tr>\n",
 630 |        "  <tr>\n",
 631 |        "    <th>26</th>\n",
 632 |        "    <th>0.758423</th>\n",
 633 |        "    <th>0.764043</th>\n",
 634 |        "    <th>0.766750</th>\n",
 635 |        "  </tr>\n",
 636 |        "  <tr>\n",
 637 |        "    <th>27</th>\n",
 638 |        "    <th>0.749885</th>\n",
 639 |        "    <th>0.762675</th>\n",
 640 |        "    <th>0.766500</th>\n",
 641 |        "  </tr>\n",
 642 |        "  <tr>\n",
 643 |        "    <th>28</th>\n",
 644 |        "    <th>0.746697</th>\n",
 645 |        "    <th>0.761255</th>\n",
 646 |        "    <th>0.767583</th>\n",
 647 |        "  </tr>\n",
 648 |        "  <tr>\n",
 649 |        "    <th>29</th>\n",
 650 |        "    <th>0.749717</th>\n",
 651 |        "    <th>0.761567</th>\n",
 652 |        "    <th>0.766833</th>\n",
 653 |        "  </tr>\n",
 654 |        "  <tr>\n",
 655 |        "    <th>30</th>\n",
 656 |        "    <th>0.747547</th>\n",
 657 |        "    <th>0.762019</th>\n",
 658 |        "    <th>0.766500</th>\n",
 659 |        "  </tr>\n",
 660 |        "</table>\n"
 661 |       ],
 662 |       "text/plain": [
 663 |        "<IPython.core.display.HTML object>"
 664 |       ]
 665 |      },
 666 |      "metadata": {},
 667 |      "output_type": "display_data"
 668 |     }
 669 |    ],
 670 |    "source": [
 671 |     "with gpu_mem_restore_ctx():\n",
 672 |     "    learn.fit_one_cycle(30)"
 673 |    ]
 674 |   },
 675 |   {
 676 |    "cell_type": "markdown",
 677 |    "metadata": {},
 678 |    "source": [
 679 |     "## Resnet 34 "
 680 |    ]
 681 |   },
 682 |   {
 683 |    "cell_type": "code",
 684 |    "execution_count": 155,
 685 |    "metadata": {},
 686 |    "outputs": [],
 687 |    "source": [
 688 |     "np.random.seed(42)\n",
 689 |     "data = ImageDataBunch.from_folder(path, train=train, valid_pct=0.2,\n",
 690 |     "        ds_tfms=get_transforms(), size=224, num_workers=4, bs=248).normalize(cifar_stats)"
 691 |    ]
 692 |   },
 693 |   {
 694 |    "cell_type": "code",
 695 |    "execution_count": 156,
 696 |    "metadata": {},
 697 |    "outputs": [],
 698 |    "source": [
 699 |     "learn = create_cnn(data, models.resnet34, metrics=accuracy)"
 700 |    ]
 701 |   },
 702 |   {
 703 |    "cell_type": "code",
 704 |    "execution_count": 157,
 705 |    "metadata": {},
 706 |    "outputs": [
 707 |     {
 708 |      "data": {
 709 |       "text/html": [
 710 |        "Total time: 39:37 <p><table style='width:300px; margin-bottom:10px'>\n",
 711 |        "  <tr>\n",
 712 |        "    <th>epoch</th>\n",
 713 |        "    <th>train_loss</th>\n",
 714 |        "    <th>valid_loss</th>\n",
 715 |        "    <th>accuracy</th>\n",
 716 |        "  </tr>\n",
 717 |        "  <tr>\n",
 718 |        "    <th>1</th>\n",
 719 |        "    <th>4.091527</th>\n",
 720 |        "    <th>3.254414</th>\n",
 721 |        "    <th>0.291917</th>\n",
 722 |        "  </tr>\n",
 723 |        "  <tr>\n",
 724 |        "    <th>2</th>\n",
 725 |        "    <th>2.743854</th>\n",
 726 |        "    <th>1.980968</th>\n",
 727 |        "    <th>0.516750</th>\n",
 728 |        "  </tr>\n",
 729 |        "  <tr>\n",
 730 |        "    <th>3</th>\n",
 731 |        "    <th>1.920904</th>\n",
 732 |        "    <th>1.337292</th>\n",
 733 |        "    <th>0.628833</th>\n",
 734 |        "  </tr>\n",
 735 |        "  <tr>\n",
 736 |        "    <th>4</th>\n",
 737 |        "    <th>1.471699</th>\n",
 738 |        "    <th>1.053553</th>\n",
 739 |        "    <th>0.692083</th>\n",
 740 |        "  </tr>\n",
 741 |        "  <tr>\n",
 742 |        "    <th>5</th>\n",
 743 |        "    <th>1.233301</th>\n",
 744 |        "    <th>0.913338</th>\n",
 745 |        "    <th>0.720250</th>\n",
 746 |        "  </tr>\n",
 747 |        "  <tr>\n",
 748 |        "    <th>6</th>\n",
 749 |        "    <th>1.109750</th>\n",
 750 |        "    <th>0.848846</th>\n",
 751 |        "    <th>0.743417</th>\n",
 752 |        "  </tr>\n",
 753 |        "  <tr>\n",
 754 |        "    <th>7</th>\n",
 755 |        "    <th>1.037520</th>\n",
 756 |        "    <th>0.791737</th>\n",
 757 |        "    <th>0.756333</th>\n",
 758 |        "  </tr>\n",
 759 |        "  <tr>\n",
 760 |        "    <th>8</th>\n",
 761 |        "    <th>0.969036</th>\n",
 762 |        "    <th>0.762382</th>\n",
 763 |        "    <th>0.766417</th>\n",
 764 |        "  </tr>\n",
 765 |        "  <tr>\n",
 766 |        "    <th>9</th>\n",
 767 |        "    <th>0.919657</th>\n",
 768 |        "    <th>0.729532</th>\n",
 769 |        "    <th>0.773250</th>\n",
 770 |        "  </tr>\n",
 771 |        "  <tr>\n",
 772 |        "    <th>10</th>\n",
 773 |        "    <th>0.885872</th>\n",
 774 |        "    <th>0.712802</th>\n",
 775 |        "    <th>0.778000</th>\n",
 776 |        "  </tr>\n",
 777 |        "  <tr>\n",
 778 |        "    <th>11</th>\n",
 779 |        "    <th>0.839404</th>\n",
 780 |        "    <th>0.704638</th>\n",
 781 |        "    <th>0.782583</th>\n",
 782 |        "  </tr>\n",
 783 |        "  <tr>\n",
 784 |        "    <th>12</th>\n",
 785 |        "    <th>0.813941</th>\n",
 786 |        "    <th>0.690926</th>\n",
 787 |        "    <th>0.786333</th>\n",
 788 |        "  </tr>\n",
 789 |        "  <tr>\n",
 790 |        "    <th>13</th>\n",
 791 |        "    <th>0.789469</th>\n",
 792 |        "    <th>0.672594</th>\n",
 793 |        "    <th>0.793000</th>\n",
 794 |        "  </tr>\n",
 795 |        "  <tr>\n",
 796 |        "    <th>14</th>\n",
 797 |        "    <th>0.765909</th>\n",
 798 |        "    <th>0.663917</th>\n",
 799 |        "    <th>0.796250</th>\n",
 800 |        "  </tr>\n",
 801 |        "  <tr>\n",
 802 |        "    <th>15</th>\n",
 803 |        "    <th>0.728020</th>\n",
 804 |        "    <th>0.653305</th>\n",
 805 |        "    <th>0.800167</th>\n",
 806 |        "  </tr>\n",
 807 |        "  <tr>\n",
 808 |        "    <th>16</th>\n",
 809 |        "    <th>0.707719</th>\n",
 810 |        "    <th>0.656965</th>\n",
 811 |        "    <th>0.797250</th>\n",
 812 |        "  </tr>\n",
 813 |        "  <tr>\n",
 814 |        "    <th>17</th>\n",
 815 |        "    <th>0.682595</th>\n",
 816 |        "    <th>0.642476</th>\n",
 817 |        "    <th>0.804083</th>\n",
 818 |        "  </tr>\n",
 819 |        "  <tr>\n",
 820 |        "    <th>18</th>\n",
 821 |        "    <th>0.663653</th>\n",
 822 |        "    <th>0.637333</th>\n",
 823 |        "    <th>0.803500</th>\n",
 824 |        "  </tr>\n",
 825 |        "  <tr>\n",
 826 |        "    <th>19</th>\n",
 827 |        "    <th>0.651230</th>\n",
 828 |        "    <th>0.625249</th>\n",
 829 |        "    <th>0.806083</th>\n",
 830 |        "  </tr>\n",
 831 |        "  <tr>\n",
 832 |        "    <th>20</th>\n",
 833 |        "    <th>0.621778</th>\n",
 834 |        "    <th>0.622463</th>\n",
 835 |        "    <th>0.809583</th>\n",
 836 |        "  </tr>\n",
 837 |        "  <tr>\n",
 838 |        "    <th>21</th>\n",
 839 |        "    <th>0.613027</th>\n",
 840 |        "    <th>0.621823</th>\n",
 841 |        "    <th>0.809250</th>\n",
 842 |        "  </tr>\n",
 843 |        "  <tr>\n",
 844 |        "    <th>22</th>\n",
 845 |        "    <th>0.608741</th>\n",
 846 |        "    <th>0.618127</th>\n",
 847 |        "    <th>0.813333</th>\n",
 848 |        "  </tr>\n",
 849 |        "  <tr>\n",
 850 |        "    <th>23</th>\n",
 851 |        "    <th>0.577575</th>\n",
 852 |        "    <th>0.613892</th>\n",
 853 |        "    <th>0.814583</th>\n",
 854 |        "  </tr>\n",
 855 |        "  <tr>\n",
 856 |        "    <th>24</th>\n",
 857 |        "    <th>0.570914</th>\n",
 858 |        "    <th>0.609038</th>\n",
 859 |        "    <th>0.814833</th>\n",
 860 |        "  </tr>\n",
 861 |        "  <tr>\n",
 862 |        "    <th>25</th>\n",
 863 |        "    <th>0.544643</th>\n",
 864 |        "    <th>0.609252</th>\n",
 865 |        "    <th>0.816750</th>\n",
 866 |        "  </tr>\n",
 867 |        "  <tr>\n",
 868 |        "    <th>26</th>\n",
 869 |        "    <th>0.548220</th>\n",
 870 |        "    <th>0.606148</th>\n",
 871 |        "    <th>0.817583</th>\n",
 872 |        "  </tr>\n",
 873 |        "  <tr>\n",
 874 |        "    <th>27</th>\n",
 875 |        "    <th>0.535698</th>\n",
 876 |        "    <th>0.604082</th>\n",
 877 |        "    <th>0.817750</th>\n",
 878 |        "  </tr>\n",
 879 |        "  <tr>\n",
 880 |        "    <th>28</th>\n",
 881 |        "    <th>0.525107</th>\n",
 882 |        "    <th>0.604753</th>\n",
 883 |        "    <th>0.817667</th>\n",
 884 |        "  </tr>\n",
 885 |        "  <tr>\n",
 886 |        "    <th>29</th>\n",
 887 |        "    <th>0.531150</th>\n",
 888 |        "    <th>0.604243</th>\n",
 889 |        "    <th>0.817250</th>\n",
 890 |        "  </tr>\n",
 891 |        "  <tr>\n",
 892 |        "    <th>30</th>\n",
 893 |        "    <th>0.533026</th>\n",
 894 |        "    <th>0.604470</th>\n",
 895 |        "    <th>0.817167</th>\n",
 896 |        "  </tr>\n",
 897 |        "</table>\n"
 898 |       ],
 899 |       "text/plain": [
 900 |        "<IPython.core.display.HTML object>"
 901 |       ]
 902 |      },
 903 |      "metadata": {},
 904 |      "output_type": "display_data"
 905 |     }
 906 |    ],
 907 |    "source": [
 908 |     "with gpu_mem_restore_ctx():\n",
 909 |     "    learn.fit_one_cycle(30)"
 910 |    ]
 911 |   },
 912 |   {
 913 |    "cell_type": "markdown",
 914 |    "metadata": {},
 915 |    "source": [
 916 |     "## Resnet 34 (Mixed Precision)"
 917 |    ]
 918 |   },
 919 |   {
 920 |    "cell_type": "code",
 921 |    "execution_count": 158,
 922 |    "metadata": {},
 923 |    "outputs": [],
 924 |    "source": [
 925 |     "np.random.seed(42)\n",
 926 |     "data = ImageDataBunch.from_folder(path, train=train, valid_pct=0.2,\n",
 927 |     "        ds_tfms=get_transforms(), size=224, num_workers=4, bs=496).normalize(cifar_stats)"
 928 |    ]
 929 |   },
 930 |   {
 931 |    "cell_type": "code",
 932 |    "execution_count": 159,
 933 |    "metadata": {},
 934 |    "outputs": [],
 935 |    "source": [
 936 |     "learn = to_fp16(create_cnn(data, models.resnet34, metrics=accuracy))"
 937 |    ]
 938 |   },
 939 |   {
 940 |    "cell_type": "code",
 941 |    "execution_count": 160,
 942 |    "metadata": {},
 943 |    "outputs": [
 944 |     {
 945 |      "data": {
 946 |       "text/html": [
 947 |        "Total time: 32:13 <p><table style='width:300px; margin-bottom:10px'>\n",
 948 |        "  <tr>\n",
 949 |        "    <th>epoch</th>\n",
 950 |        "    <th>train_loss</th>\n",
 951 |        "    <th>valid_loss</th>\n",
 952 |        "    <th>accuracy</th>\n",
 953 |        "  </tr>\n",
 954 |        "  <tr>\n",
 955 |        "    <th>1</th>\n",
 956 |        "    <th>4.527595</th>\n",
 957 |        "    <th>3.723982</th>\n",
 958 |        "    <th>0.192333</th>\n",
 959 |        "  </tr>\n",
 960 |        "  <tr>\n",
 961 |        "    <th>2</th>\n",
 962 |        "    <th>3.496515</th>\n",
 963 |        "    <th>2.463351</th>\n",
 964 |        "    <th>0.443083</th>\n",
 965 |        "  </tr>\n",
 966 |        "  <tr>\n",
 967 |        "    <th>3</th>\n",
 968 |        "    <th>2.486472</th>\n",
 969 |        "    <th>1.608410</th>\n",
 970 |        "    <th>0.577167</th>\n",
 971 |        "  </tr>\n",
 972 |        "  <tr>\n",
 973 |        "    <th>4</th>\n",
 974 |        "    <th>1.810522</th>\n",
 975 |        "    <th>1.204057</th>\n",
 976 |        "    <th>0.655750</th>\n",
 977 |        "  </tr>\n",
 978 |        "  <tr>\n",
 979 |        "    <th>5</th>\n",
 980 |        "    <th>1.430669</th>\n",
 981 |        "    <th>1.008747</th>\n",
 982 |        "    <th>0.701667</th>\n",
 983 |        "  </tr>\n",
 984 |        "  <tr>\n",
 985 |        "    <th>6</th>\n",
 986 |        "    <th>1.225085</th>\n",
 987 |        "    <th>0.905269</th>\n",
 988 |        "    <th>0.727500</th>\n",
 989 |        "  </tr>\n",
 990 |        "  <tr>\n",
 991 |        "    <th>7</th>\n",
 992 |        "    <th>1.092392</th>\n",
 993 |        "    <th>0.835543</th>\n",
 994 |        "    <th>0.743750</th>\n",
 995 |        "  </tr>\n",
 996 |        "  <tr>\n",
 997 |        "    <th>8</th>\n",
 998 |        "    <th>1.013207</th>\n",
 999 |        "    <th>0.793349</th>\n",
1000 |        "    <th>0.753833</th>\n",
1001 |        "  </tr>\n",
1002 |        "  <tr>\n",
1003 |        "    <th>9</th>\n",
1004 |        "    <th>0.954900</th>\n",
1005 |        "    <th>0.760904</th>\n",
1006 |        "    <th>0.762250</th>\n",
1007 |        "  </tr>\n",
1008 |        "  <tr>\n",
1009 |        "    <th>10</th>\n",
1010 |        "    <th>0.897951</th>\n",
1011 |        "    <th>0.739107</th>\n",
1012 |        "    <th>0.773417</th>\n",
1013 |        "  </tr>\n",
1014 |        "  <tr>\n",
1015 |        "    <th>11</th>\n",
1016 |        "    <th>0.860628</th>\n",
1017 |        "    <th>0.723568</th>\n",
1018 |        "    <th>0.776000</th>\n",
1019 |        "  </tr>\n",
1020 |        "  <tr>\n",
1021 |        "    <th>12</th>\n",
1022 |        "    <th>0.823930</th>\n",
1023 |        "    <th>0.703144</th>\n",
1024 |        "    <th>0.784000</th>\n",
1025 |        "  </tr>\n",
1026 |        "  <tr>\n",
1027 |        "    <th>13</th>\n",
1028 |        "    <th>0.792410</th>\n",
1029 |        "    <th>0.681215</th>\n",
1030 |        "    <th>0.790583</th>\n",
1031 |        "  </tr>\n",
1032 |        "  <tr>\n",
1033 |        "    <th>14</th>\n",
1034 |        "    <th>0.767022</th>\n",
1035 |        "    <th>0.683587</th>\n",
1036 |        "    <th>0.789167</th>\n",
1037 |        "  </tr>\n",
1038 |        "  <tr>\n",
1039 |        "    <th>15</th>\n",
1040 |        "    <th>0.741968</th>\n",
1041 |        "    <th>0.668285</th>\n",
1042 |        "    <th>0.796083</th>\n",
1043 |        "  </tr>\n",
1044 |        "  <tr>\n",
1045 |        "    <th>16</th>\n",
1046 |        "    <th>0.718596</th>\n",
1047 |        "    <th>0.661959</th>\n",
1048 |        "    <th>0.797583</th>\n",
1049 |        "  </tr>\n",
1050 |        "  <tr>\n",
1051 |        "    <th>17</th>\n",
1052 |        "    <th>0.693880</th>\n",
1053 |        "    <th>0.657565</th>\n",
1054 |        "    <th>0.796500</th>\n",
1055 |        "  </tr>\n",
1056 |        "  <tr>\n",
1057 |        "    <th>18</th>\n",
1058 |        "    <th>0.676531</th>\n",
1059 |        "    <th>0.652226</th>\n",
1060 |        "    <th>0.804333</th>\n",
1061 |        "  </tr>\n",
1062 |        "  <tr>\n",
1063 |        "    <th>19</th>\n",
1064 |        "    <th>0.661541</th>\n",
1065 |        "    <th>0.650092</th>\n",
1066 |        "    <th>0.803000</th>\n",
1067 |        "  </tr>\n",
1068 |        "  <tr>\n",
1069 |        "    <th>20</th>\n",
1070 |        "    <th>0.641904</th>\n",
1071 |        "    <th>0.632877</th>\n",
1072 |        "    <th>0.808167</th>\n",
1073 |        "  </tr>\n",
1074 |        "  <tr>\n",
1075 |        "    <th>21</th>\n",
1076 |        "    <th>0.628384</th>\n",
1077 |        "    <th>0.633768</th>\n",
1078 |        "    <th>0.807500</th>\n",
1079 |        "  </tr>\n",
1080 |        "  <tr>\n",
1081 |        "    <th>22</th>\n",
1082 |        "    <th>0.607901</th>\n",
1083 |        "    <th>0.630894</th>\n",
1084 |        "    <th>0.808167</th>\n",
1085 |        "  </tr>\n",
1086 |        "  <tr>\n",
1087 |        "    <th>23</th>\n",
1088 |        "    <th>0.597407</th>\n",
1089 |        "    <th>0.630145</th>\n",
1090 |        "    <th>0.809083</th>\n",
1091 |        "  </tr>\n",
1092 |        "  <tr>\n",
1093 |        "    <th>24</th>\n",
1094 |        "    <th>0.580347</th>\n",
1095 |        "    <th>0.625942</th>\n",
1096 |        "    <th>0.809583</th>\n",
1097 |        "  </tr>\n",
1098 |        "  <tr>\n",
1099 |        "    <th>25</th>\n",
1100 |        "    <th>0.576550</th>\n",
1101 |        "    <th>0.624290</th>\n",
1102 |        "    <th>0.813000</th>\n",
1103 |        "  </tr>\n",
1104 |        "  <tr>\n",
1105 |        "    <th>26</th>\n",
1106 |        "    <th>0.564124</th>\n",
1107 |        "    <th>0.620901</th>\n",
1108 |        "    <th>0.811500</th>\n",
1109 |        "  </tr>\n",
1110 |        "  <tr>\n",
1111 |        "    <th>27</th>\n",
1112 |        "    <th>0.561751</th>\n",
1113 |        "    <th>0.620830</th>\n",
1114 |        "    <th>0.811000</th>\n",
1115 |        "  </tr>\n",
1116 |        "  <tr>\n",
1117 |        "    <th>28</th>\n",
1118 |        "    <th>0.549595</th>\n",
1119 |        "    <th>0.620113</th>\n",
1120 |        "    <th>0.811500</th>\n",
1121 |        "  </tr>\n",
1122 |        "  <tr>\n",
1123 |        "    <th>29</th>\n",
1124 |        "    <th>0.549034</th>\n",
1125 |        "    <th>0.619881</th>\n",
1126 |        "    <th>0.812917</th>\n",
1127 |        "  </tr>\n",
1128 |        "  <tr>\n",
1129 |        "    <th>30</th>\n",
1130 |        "    <th>0.553179</th>\n",
1131 |        "    <th>0.618634</th>\n",
1132 |        "    <th>0.812083</th>\n",
1133 |        "  </tr>\n",
1134 |        "</table>\n"
1135 |       ],
1136 |       "text/plain": [
1137 |        "<IPython.core.display.HTML object>"
1138 |       ]
1139 |      },
1140 |      "metadata": {},
1141 |      "output_type": "display_data"
1142 |     }
1143 |    ],
1144 |    "source": [
1145 |     "with gpu_mem_restore_ctx():\n",
1146 |     "    learn.fit_one_cycle(30)"
1147 |    ]
1148 |   },
1149 |   {
1150 |    "cell_type": "markdown",
1151 |    "metadata": {},
1152 |    "source": [
1153 |     "## Training Resnet 50"
1154 |    ]
1155 |   },
1156 |   {
1157 |    "cell_type": "code",
1158 |    "execution_count": 44,
1159 |    "metadata": {},
1160 |    "outputs": [],
1161 |    "source": [
1162 |     "np.random.seed(42)\n",
1163 |     "data = ImageDataBunch.from_folder(path, train=train, valid_pct=0.2,\n",
1164 |     "        ds_tfms=get_transforms(), size=224, num_workers=4, bs=98).normalize(cifar_stats)"
1165 |    ]
1166 |   },
1167 |   {
1168 |    "cell_type": "code",
1169 |    "execution_count": 45,
1170 |    "metadata": {},
1171 |    "outputs": [],
1172 |    "source": [
1173 |     "learn = create_cnn(data, models.resnet50, metrics=accuracy)"
1174 |    ]
1175 |   },
1176 |   {
1177 |    "cell_type": "code",
1178 |    "execution_count": 46,
1179 |    "metadata": {},
1180 |    "outputs": [
1181 |     {
1182 |      "data": {
1183 |       "text/html": [
1184 |        "Total time: 1:16:18 <p><table style='width:300px; margin-bottom:10px'>\n",
1185 |        "  <tr>\n",
1186 |        "    <th>epoch</th>\n",
1187 |        "    <th>train_loss</th>\n",
1188 |        "    <th>valid_loss</th>\n",
1189 |        "    <th>accuracy</th>\n",
1190 |        "  </tr>\n",
1191 |        "  <tr>\n",
1192 |        "    <th>1</th>\n",
1193 |        "    <th>2.538904</th>\n",
1194 |        "    <th>1.896686</th>\n",
1195 |        "    <th>0.537917</th>\n",
1196 |        "  </tr>\n",
1197 |        "  <tr>\n",
1198 |        "    <th>2</th>\n",
1199 |        "    <th>1.700987</th>\n",
1200 |        "    <th>1.219014</th>\n",
1201 |        "    <th>0.660167</th>\n",
1202 |        "  </tr>\n",
1203 |        "  <tr>\n",
1204 |        "    <th>3</th>\n",
1205 |        "    <th>1.304718</th>\n",
1206 |        "    <th>0.952666</th>\n",
1207 |        "    <th>0.716250</th>\n",
1208 |        "  </tr>\n",
1209 |        "  <tr>\n",
1210 |        "    <th>4</th>\n",
1211 |        "    <th>1.097150</th>\n",
1212 |        "    <th>0.850275</th>\n",
1213 |        "    <th>0.743417</th>\n",
1214 |        "  </tr>\n",
1215 |        "  <tr>\n",
1216 |        "    <th>5</th>\n",
1217 |        "    <th>1.008206</th>\n",
1218 |        "    <th>0.781421</th>\n",
1219 |        "    <th>0.758083</th>\n",
1220 |        "  </tr>\n",
1221 |        "  <tr>\n",
1222 |        "    <th>6</th>\n",
1223 |        "    <th>0.947150</th>\n",
1224 |        "    <th>0.740677</th>\n",
1225 |        "    <th>0.770500</th>\n",
1226 |        "  </tr>\n",
1227 |        "  <tr>\n",
1228 |        "    <th>7</th>\n",
1229 |        "    <th>0.891749</th>\n",
1230 |        "    <th>0.720329</th>\n",
1231 |        "    <th>0.778250</th>\n",
1232 |        "  </tr>\n",
1233 |        "  <tr>\n",
1234 |        "    <th>8</th>\n",
1235 |        "    <th>0.800058</th>\n",
1236 |        "    <th>0.686394</th>\n",
1237 |        "    <th>0.791250</th>\n",
1238 |        "  </tr>\n",
1239 |        "  <tr>\n",
1240 |        "    <th>9</th>\n",
1241 |        "    <th>0.771210</th>\n",
1242 |        "    <th>0.674660</th>\n",
1243 |        "    <th>0.793167</th>\n",
1244 |        "  </tr>\n",
1245 |        "  <tr>\n",
1246 |        "    <th>10</th>\n",
1247 |        "    <th>0.744009</th>\n",
1248 |        "    <th>0.653407</th>\n",
1249 |        "    <th>0.798000</th>\n",
1250 |        "  </tr>\n",
1251 |        "  <tr>\n",
1252 |        "    <th>11</th>\n",
1253 |        "    <th>0.710490</th>\n",
1254 |        "    <th>0.635136</th>\n",
1255 |        "    <th>0.803083</th>\n",
1256 |        "  </tr>\n",
1257 |        "  <tr>\n",
1258 |        "    <th>12</th>\n",
1259 |        "    <th>0.653651</th>\n",
1260 |        "    <th>0.618544</th>\n",
1261 |        "    <th>0.810833</th>\n",
1262 |        "  </tr>\n",
1263 |        "  <tr>\n",
1264 |        "    <th>13</th>\n",
1265 |        "    <th>0.654817</th>\n",
1266 |        "    <th>0.612365</th>\n",
1267 |        "    <th>0.811500</th>\n",
1268 |        "  </tr>\n",
1269 |        "  <tr>\n",
1270 |        "    <th>14</th>\n",
1271 |        "    <th>0.604655</th>\n",
1272 |        "    <th>0.610485</th>\n",
1273 |        "    <th>0.818000</th>\n",
1274 |        "  </tr>\n",
1275 |        "  <tr>\n",
1276 |        "    <th>15</th>\n",
1277 |        "    <th>0.593437</th>\n",
1278 |        "    <th>0.591317</th>\n",
1279 |        "    <th>0.820833</th>\n",
1280 |        "  </tr>\n",
1281 |        "  <tr>\n",
1282 |        "    <th>16</th>\n",
1283 |        "    <th>0.553768</th>\n",
1284 |        "    <th>0.596443</th>\n",
1285 |        "    <th>0.819417</th>\n",
1286 |        "  </tr>\n",
1287 |        "  <tr>\n",
1288 |        "    <th>17</th>\n",
1289 |        "    <th>0.507188</th>\n",
1290 |        "    <th>0.585333</th>\n",
1291 |        "    <th>0.825250</th>\n",
1292 |        "  </tr>\n",
1293 |        "  <tr>\n",
1294 |        "    <th>18</th>\n",
1295 |        "    <th>0.493085</th>\n",
1296 |        "    <th>0.583739</th>\n",
1297 |        "    <th>0.829417</th>\n",
1298 |        "  </tr>\n",
1299 |        "  <tr>\n",
1300 |        "    <th>19</th>\n",
1301 |        "    <th>0.476964</th>\n",
1302 |        "    <th>0.575371</th>\n",
1303 |        "    <th>0.828667</th>\n",
1304 |        "  </tr>\n",
1305 |        "  <tr>\n",
1306 |        "    <th>20</th>\n",
1307 |        "    <th>0.455830</th>\n",
1308 |        "    <th>0.569629</th>\n",
1309 |        "    <th>0.831167</th>\n",
1310 |        "  </tr>\n",
1311 |        "  <tr>\n",
1312 |        "    <th>21</th>\n",
1313 |        "    <th>0.447361</th>\n",
1314 |        "    <th>0.572533</th>\n",
1315 |        "    <th>0.831417</th>\n",
1316 |        "  </tr>\n",
1317 |        "  <tr>\n",
1318 |        "    <th>22</th>\n",
1319 |        "    <th>0.407300</th>\n",
1320 |        "    <th>0.567695</th>\n",
1321 |        "    <th>0.833667</th>\n",
1322 |        "  </tr>\n",
1323 |        "  <tr>\n",
1324 |        "    <th>23</th>\n",
1325 |        "    <th>0.383044</th>\n",
1326 |        "    <th>0.565551</th>\n",
1327 |        "    <th>0.833667</th>\n",
1328 |        "  </tr>\n",
1329 |        "  <tr>\n",
1330 |        "    <th>24</th>\n",
1331 |        "    <th>0.369885</th>\n",
1332 |        "    <th>0.563926</th>\n",
1333 |        "    <th>0.835917</th>\n",
1334 |        "  </tr>\n",
1335 |        "  <tr>\n",
1336 |        "    <th>25</th>\n",
1337 |        "    <th>0.376316</th>\n",
1338 |        "    <th>0.561316</th>\n",
1339 |        "    <th>0.836833</th>\n",
1340 |        "  </tr>\n",
1341 |        "  <tr>\n",
1342 |        "    <th>26</th>\n",
1343 |        "    <th>0.335686</th>\n",
1344 |        "    <th>0.558788</th>\n",
1345 |        "    <th>0.838000</th>\n",
1346 |        "  </tr>\n",
1347 |        "  <tr>\n",
1348 |        "    <th>27</th>\n",
1349 |        "    <th>0.346769</th>\n",
1350 |        "    <th>0.560210</th>\n",
1351 |        "    <th>0.838083</th>\n",
1352 |        "  </tr>\n",
1353 |        "  <tr>\n",
1354 |        "    <th>28</th>\n",
1355 |        "    <th>0.340222</th>\n",
1356 |        "    <th>0.561173</th>\n",
1357 |        "    <th>0.839667</th>\n",
1358 |        "  </tr>\n",
1359 |        "  <tr>\n",
1360 |        "    <th>29</th>\n",
1361 |        "    <th>0.327940</th>\n",
1362 |        "    <th>0.558264</th>\n",
1363 |        "    <th>0.839667</th>\n",
1364 |        "  </tr>\n",
1365 |        "  <tr>\n",
1366 |        "    <th>30</th>\n",
1367 |        "    <th>0.316947</th>\n",
1368 |        "    <th>0.559535</th>\n",
1369 |        "    <th>0.839333</th>\n",
1370 |        "  </tr>\n",
1371 |        "</table>\n"
1372 |       ],
1373 |       "text/plain": [
1374 |        "<IPython.core.display.HTML object>"
1375 |       ]
1376 |      },
1377 |      "metadata": {},
1378 |      "output_type": "display_data"
1379 |     }
1380 |    ],
1381 |    "source": [
1382 |     "with gpu_mem_restore_ctx():\n",
1383 |     "    learn.fit_one_cycle(30)"
1384 |    ]
1385 |   },
1386 |   {
1387 |    "cell_type": "markdown",
1388 |    "metadata": {},
1389 |    "source": [
1390 |     "## Resnet 50 (Mixed Precision)"
1391 |    ]
1392 |   },
1393 |   {
1394 |    "cell_type": "code",
1395 |    "execution_count": 57,
1396 |    "metadata": {},
1397 |    "outputs": [],
1398 |    "source": [
1399 |     "np.random.seed(42)\n",
1400 |     "data = ImageDataBunch.from_folder(path, train=train, valid_pct=0.2,\n",
1401 |     "        ds_tfms=get_transforms(), size=224, num_workers=4, bs=164).normalize(cifar_stats)"
1402 |    ]
1403 |   },
1404 |   {
1405 |    "cell_type": "code",
1406 |    "execution_count": 58,
1407 |    "metadata": {},
1408 |    "outputs": [],
1409 |    "source": [
1410 |     "learn = to_fp16(create_cnn(data, models.resnet50, metrics=accuracy))"
1411 |    ]
1412 |   },
1413 |   {
1414 |    "cell_type": "code",
1415 |    "execution_count": 59,
1416 |    "metadata": {},
1417 |    "outputs": [
1418 |     {
1419 |      "data": {
1420 |       "text/html": [
1421 |        "Total time: 53:38 <p><table style='width:300px; margin-bottom:10px'>\n",
1422 |        "  <tr>\n",
1423 |        "    <th>epoch</th>\n",
1424 |        "    <th>train_loss</th>\n",
1425 |        "    <th>valid_loss</th>\n",
1426 |        "    <th>accuracy</th>\n",
1427 |        "  </tr>\n",
1428 |        "  <tr>\n",
1429 |        "    <th>1</th>\n",
1430 |        "    <th>2.853479</th>\n",
1431 |        "    <th>2.087002</th>\n",
1432 |        "    <th>0.499583</th>\n",
1433 |        "  </tr>\n",
1434 |        "  <tr>\n",
1435 |        "    <th>2</th>\n",
1436 |        "    <th>1.851662</th>\n",
1437 |        "    <th>1.342598</th>\n",
1438 |        "    <th>0.631250</th>\n",
1439 |        "  </tr>\n",
1440 |        "  <tr>\n",
1441 |        "    <th>3</th>\n",
1442 |        "    <th>1.387896</th>\n",
1443 |        "    <th>1.018817</th>\n",
1444 |        "    <th>0.699417</th>\n",
1445 |        "  </tr>\n",
1446 |        "  <tr>\n",
1447 |        "    <th>4</th>\n",
1448 |        "    <th>1.165870</th>\n",
1449 |        "    <th>0.866866</th>\n",
1450 |        "    <th>0.736500</th>\n",
1451 |        "  </tr>\n",
1452 |        "  <tr>\n",
1453 |        "    <th>5</th>\n",
1454 |        "    <th>1.031915</th>\n",
1455 |        "    <th>0.790143</th>\n",
1456 |        "    <th>0.755833</th>\n",
1457 |        "  </tr>\n",
1458 |        "  <tr>\n",
1459 |        "    <th>6</th>\n",
1460 |        "    <th>0.921515</th>\n",
1461 |        "    <th>0.759719</th>\n",
1462 |        "    <th>0.765500</th>\n",
1463 |        "  </tr>\n",
1464 |        "  <tr>\n",
1465 |        "    <th>7</th>\n",
1466 |        "    <th>0.880313</th>\n",
1467 |        "    <th>0.717190</th>\n",
1468 |        "    <th>0.778917</th>\n",
1469 |        "  </tr>\n",
1470 |        "  <tr>\n",
1471 |        "    <th>8</th>\n",
1472 |        "    <th>0.838310</th>\n",
1473 |        "    <th>0.697754</th>\n",
1474 |        "    <th>0.784250</th>\n",
1475 |        "  </tr>\n",
1476 |        "  <tr>\n",
1477 |        "    <th>9</th>\n",
1478 |        "    <th>0.775729</th>\n",
1479 |        "    <th>0.681422</th>\n",
1480 |        "    <th>0.791750</th>\n",
1481 |        "  </tr>\n",
1482 |        "  <tr>\n",
1483 |        "    <th>10</th>\n",
1484 |        "    <th>0.725294</th>\n",
1485 |        "    <th>0.649651</th>\n",
1486 |        "    <th>0.801583</th>\n",
1487 |        "  </tr>\n",
1488 |        "  <tr>\n",
1489 |        "    <th>11</th>\n",
1490 |        "    <th>0.692380</th>\n",
1491 |        "    <th>0.643033</th>\n",
1492 |        "    <th>0.801167</th>\n",
1493 |        "  </tr>\n",
1494 |        "  <tr>\n",
1495 |        "    <th>12</th>\n",
1496 |        "    <th>0.661506</th>\n",
1497 |        "    <th>0.633204</th>\n",
1498 |        "    <th>0.805750</th>\n",
1499 |        "  </tr>\n",
1500 |        "  <tr>\n",
1501 |        "    <th>13</th>\n",
1502 |        "    <th>0.616202</th>\n",
1503 |        "    <th>0.624044</th>\n",
1504 |        "    <th>0.810333</th>\n",
1505 |        "  </tr>\n",
1506 |        "  <tr>\n",
1507 |        "    <th>14</th>\n",
1508 |        "    <th>0.591845</th>\n",
1509 |        "    <th>0.612394</th>\n",
1510 |        "    <th>0.812417</th>\n",
1511 |        "  </tr>\n",
1512 |        "  <tr>\n",
1513 |        "    <th>15</th>\n",
1514 |        "    <th>0.566731</th>\n",
1515 |        "    <th>0.603592</th>\n",
1516 |        "    <th>0.818500</th>\n",
1517 |        "  </tr>\n",
1518 |        "  <tr>\n",
1519 |        "    <th>16</th>\n",
1520 |        "    <th>0.542707</th>\n",
1521 |        "    <th>0.597049</th>\n",
1522 |        "    <th>0.818333</th>\n",
1523 |        "  </tr>\n",
1524 |        "  <tr>\n",
1525 |        "    <th>17</th>\n",
1526 |        "    <th>0.506298</th>\n",
1527 |        "    <th>0.596841</th>\n",
1528 |        "    <th>0.822417</th>\n",
1529 |        "  </tr>\n",
1530 |        "  <tr>\n",
1531 |        "    <th>18</th>\n",
1532 |        "    <th>0.492856</th>\n",
1533 |        "    <th>0.588816</th>\n",
1534 |        "    <th>0.824667</th>\n",
1535 |        "  </tr>\n",
1536 |        "  <tr>\n",
1537 |        "    <th>19</th>\n",
1538 |        "    <th>0.450351</th>\n",
1539 |        "    <th>0.585315</th>\n",
1540 |        "    <th>0.826667</th>\n",
1541 |        "  </tr>\n",
1542 |        "  <tr>\n",
1543 |        "    <th>20</th>\n",
1544 |        "    <th>0.441241</th>\n",
1545 |        "    <th>0.580710</th>\n",
1546 |        "    <th>0.827167</th>\n",
1547 |        "  </tr>\n",
1548 |        "  <tr>\n",
1549 |        "    <th>21</th>\n",
1550 |        "    <th>0.420925</th>\n",
1551 |        "    <th>0.578927</th>\n",
1552 |        "    <th>0.828500</th>\n",
1553 |        "  </tr>\n",
1554 |        "  <tr>\n",
1555 |        "    <th>22</th>\n",
1556 |        "    <th>0.392566</th>\n",
1557 |        "    <th>0.581997</th>\n",
1558 |        "    <th>0.831417</th>\n",
1559 |        "  </tr>\n",
1560 |        "  <tr>\n",
1561 |        "    <th>23</th>\n",
1562 |        "    <th>0.374037</th>\n",
1563 |        "    <th>0.571375</th>\n",
1564 |        "    <th>0.832500</th>\n",
1565 |        "  </tr>\n",
1566 |        "  <tr>\n",
1567 |        "    <th>24</th>\n",
1568 |        "    <th>0.366736</th>\n",
1569 |        "    <th>0.572550</th>\n",
1570 |        "    <th>0.832583</th>\n",
1571 |        "  </tr>\n",
1572 |        "  <tr>\n",
1573 |        "    <th>25</th>\n",
1574 |        "    <th>0.339871</th>\n",
1575 |        "    <th>0.569914</th>\n",
1576 |        "    <th>0.831500</th>\n",
1577 |        "  </tr>\n",
1578 |        "  <tr>\n",
1579 |        "    <th>26</th>\n",
1580 |        "    <th>0.344734</th>\n",
1581 |        "    <th>0.568628</th>\n",
1582 |        "    <th>0.833083</th>\n",
1583 |        "  </tr>\n",
1584 |        "  <tr>\n",
1585 |        "    <th>27</th>\n",
1586 |        "    <th>0.322793</th>\n",
1587 |        "    <th>0.567510</th>\n",
1588 |        "    <th>0.832667</th>\n",
1589 |        "  </tr>\n",
1590 |        "  <tr>\n",
1591 |        "    <th>28</th>\n",
1592 |        "    <th>0.307898</th>\n",
1593 |        "    <th>0.565839</th>\n",
1594 |        "    <th>0.835750</th>\n",
1595 |        "  </tr>\n",
1596 |        "  <tr>\n",
1597 |        "    <th>29</th>\n",
1598 |        "    <th>0.322832</th>\n",
1599 |        "    <th>0.566465</th>\n",
1600 |        "    <th>0.832583</th>\n",
1601 |        "  </tr>\n",
1602 |        "  <tr>\n",
1603 |        "    <th>30</th>\n",
1604 |        "    <th>0.317137</th>\n",
1605 |        "    <th>0.568857</th>\n",
1606 |        "    <th>0.832000</th>\n",
1607 |        "  </tr>\n",
1608 |        "</table>\n"
1609 |       ],
1610 |       "text/plain": [
1611 |        "<IPython.core.display.HTML object>"
1612 |       ]
1613 |      },
1614 |      "metadata": {},
1615 |      "output_type": "display_data"
1616 |     }
1617 |    ],
1618 |    "source": [
1619 |     "with gpu_mem_restore_ctx():\n",
1620 |     "    learn.fit_one_cycle(30)"
1621 |    ]
1622 |   },
1623 |   {
1624 |    "cell_type": "markdown",
1625 |    "metadata": {},
1626 |    "source": [
1627 |     "## Resnet 101"
1628 |    ]
1629 |   },
1630 |   {
1631 |    "cell_type": "code",
1632 |    "execution_count": 161,
1633 |    "metadata": {},
1634 |    "outputs": [],
1635 |    "source": [
1636 |     "np.random.seed(42)\n",
1637 |     "data = ImageDataBunch.from_folder(path, train=train, valid_pct=0.2,\n",
1638 |     "        ds_tfms=get_transforms(), size=224, num_workers=4, bs=64).normalize(cifar_stats)"
1639 |    ]
1640 |   },
1641 |   {
1642 |    "cell_type": "code",
1643 |    "execution_count": 162,
1644 |    "metadata": {},
1645 |    "outputs": [],
1646 |    "source": [
1647 |     "learn = create_cnn(data, models.resnet101, metrics=accuracy)"
1648 |    ]
1649 |   },
1650 |   {
1651 |    "cell_type": "code",
1652 |    "execution_count": 163,
1653 |    "metadata": {},
1654 |    "outputs": [
1655 |     {
1656 |      "data": {
1657 |       "text/html": [
1658 |        "Total time: 2:01:58 <p><table style='width:300px; margin-bottom:10px'>\n",
1659 |        "  <tr>\n",
1660 |        "    <th>epoch</th>\n",
1661 |        "    <th>train_loss</th>\n",
1662 |        "    <th>valid_loss</th>\n",
1663 |        "    <th>accuracy</th>\n",
1664 |        "  </tr>\n",
1665 |        "  <tr>\n",
1666 |        "    <th>1</th>\n",
1667 |        "    <th>2.055326</th>\n",
1668 |        "    <th>1.447209</th>\n",
1669 |        "    <th>0.630000</th>\n",
1670 |        "  </tr>\n",
1671 |        "  <tr>\n",
1672 |        "    <th>2</th>\n",
1673 |        "    <th>1.328482</th>\n",
1674 |        "    <th>0.938083</th>\n",
1675 |        "    <th>0.723333</th>\n",
1676 |        "  </tr>\n",
1677 |        "  <tr>\n",
1678 |        "    <th>3</th>\n",
1679 |        "    <th>1.065412</th>\n",
1680 |        "    <th>0.762804</th>\n",
1681 |        "    <th>0.767833</th>\n",
1682 |        "  </tr>\n",
1683 |        "  <tr>\n",
1684 |        "    <th>4</th>\n",
1685 |        "    <th>0.893859</th>\n",
1686 |        "    <th>0.681791</th>\n",
1687 |        "    <th>0.788333</th>\n",
1688 |        "  </tr>\n",
1689 |        "  <tr>\n",
1690 |        "    <th>5</th>\n",
1691 |        "    <th>0.857969</th>\n",
1692 |        "    <th>0.640077</th>\n",
1693 |        "    <th>0.801417</th>\n",
1694 |        "  </tr>\n",
1695 |        "  <tr>\n",
1696 |        "    <th>6</th>\n",
1697 |        "    <th>0.766642</th>\n",
1698 |        "    <th>0.637175</th>\n",
1699 |        "    <th>0.802417</th>\n",
1700 |        "  </tr>\n",
1701 |        "  <tr>\n",
1702 |        "    <th>7</th>\n",
1703 |        "    <th>0.781614</th>\n",
1704 |        "    <th>0.625195</th>\n",
1705 |        "    <th>0.804750</th>\n",
1706 |        "  </tr>\n",
1707 |        "  <tr>\n",
1708 |        "    <th>8</th>\n",
1709 |        "    <th>0.731761</th>\n",
1710 |        "    <th>0.596706</th>\n",
1711 |        "    <th>0.817333</th>\n",
1712 |        "  </tr>\n",
1713 |        "  <tr>\n",
1714 |        "    <th>9</th>\n",
1715 |        "    <th>0.682865</th>\n",
1716 |        "    <th>0.582054</th>\n",
1717 |        "    <th>0.822250</th>\n",
1718 |        "  </tr>\n",
1719 |        "  <tr>\n",
1720 |        "    <th>10</th>\n",
1721 |        "    <th>0.648590</th>\n",
1722 |        "    <th>0.560301</th>\n",
1723 |        "    <th>0.828667</th>\n",
1724 |        "  </tr>\n",
1725 |        "  <tr>\n",
1726 |        "    <th>11</th>\n",
1727 |        "    <th>0.618241</th>\n",
1728 |        "    <th>0.556258</th>\n",
1729 |        "    <th>0.829667</th>\n",
1730 |        "  </tr>\n",
1731 |        "  <tr>\n",
1732 |        "    <th>12</th>\n",
1733 |        "    <th>0.574654</th>\n",
1734 |        "    <th>0.547130</th>\n",
1735 |        "    <th>0.835917</th>\n",
1736 |        "  </tr>\n",
1737 |        "  <tr>\n",
1738 |        "    <th>13</th>\n",
1739 |        "    <th>0.530491</th>\n",
1740 |        "    <th>0.540169</th>\n",
1741 |        "    <th>0.837667</th>\n",
1742 |        "  </tr>\n",
1743 |        "  <tr>\n",
1744 |        "    <th>14</th>\n",
1745 |        "    <th>0.528848</th>\n",
1746 |        "    <th>0.534272</th>\n",
1747 |        "    <th>0.839333</th>\n",
1748 |        "  </tr>\n",
1749 |        "  <tr>\n",
1750 |        "    <th>15</th>\n",
1751 |        "    <th>0.491567</th>\n",
1752 |        "    <th>0.526254</th>\n",
1753 |        "    <th>0.840750</th>\n",
1754 |        "  </tr>\n",
1755 |        "  <tr>\n",
1756 |        "    <th>16</th>\n",
1757 |        "    <th>0.451276</th>\n",
1758 |        "    <th>0.531170</th>\n",
1759 |        "    <th>0.842167</th>\n",
1760 |        "  </tr>\n",
1761 |        "  <tr>\n",
1762 |        "    <th>17</th>\n",
1763 |        "    <th>0.438837</th>\n",
1764 |        "    <th>0.518594</th>\n",
1765 |        "    <th>0.847167</th>\n",
1766 |        "  </tr>\n",
1767 |        "  <tr>\n",
1768 |        "    <th>18</th>\n",
1769 |        "    <th>0.433088</th>\n",
1770 |        "    <th>0.524159</th>\n",
1771 |        "    <th>0.845750</th>\n",
1772 |        "  </tr>\n",
1773 |        "  <tr>\n",
1774 |        "    <th>19</th>\n",
1775 |        "    <th>0.381678</th>\n",
1776 |        "    <th>0.522843</th>\n",
1777 |        "    <th>0.847667</th>\n",
1778 |        "  </tr>\n",
1779 |        "  <tr>\n",
1780 |        "    <th>20</th>\n",
1781 |        "    <th>0.365216</th>\n",
1782 |        "    <th>0.521657</th>\n",
1783 |        "    <th>0.848917</th>\n",
1784 |        "  </tr>\n",
1785 |        "  <tr>\n",
1786 |        "    <th>21</th>\n",
1787 |        "    <th>0.343347</th>\n",
1788 |        "    <th>0.517553</th>\n",
1789 |        "    <th>0.852500</th>\n",
1790 |        "  </tr>\n",
1791 |        "  <tr>\n",
1792 |        "    <th>22</th>\n",
1793 |        "    <th>0.339596</th>\n",
1794 |        "    <th>0.507800</th>\n",
1795 |        "    <th>0.854000</th>\n",
1796 |        "  </tr>\n",
1797 |        "  <tr>\n",
1798 |        "    <th>23</th>\n",
1799 |        "    <th>0.309928</th>\n",
1800 |        "    <th>0.514493</th>\n",
1801 |        "    <th>0.854917</th>\n",
1802 |        "  </tr>\n",
1803 |        "  <tr>\n",
1804 |        "    <th>24</th>\n",
1805 |        "    <th>0.297399</th>\n",
1806 |        "    <th>0.506818</th>\n",
1807 |        "    <th>0.856000</th>\n",
1808 |        "  </tr>\n",
1809 |        "  <tr>\n",
1810 |        "    <th>25</th>\n",
1811 |        "    <th>0.268865</th>\n",
1812 |        "    <th>0.504412</th>\n",
1813 |        "    <th>0.857667</th>\n",
1814 |        "  </tr>\n",
1815 |        "  <tr>\n",
1816 |        "    <th>26</th>\n",
1817 |        "    <th>0.262993</th>\n",
1818 |        "    <th>0.505606</th>\n",
1819 |        "    <th>0.860333</th>\n",
1820 |        "  </tr>\n",
1821 |        "  <tr>\n",
1822 |        "    <th>27</th>\n",
1823 |        "    <th>0.244982</th>\n",
1824 |        "    <th>0.503122</th>\n",
1825 |        "    <th>0.858667</th>\n",
1826 |        "  </tr>\n",
1827 |        "  <tr>\n",
1828 |        "    <th>28</th>\n",
1829 |        "    <th>0.249251</th>\n",
1830 |        "    <th>0.501970</th>\n",
1831 |        "    <th>0.860833</th>\n",
1832 |        "  </tr>\n",
1833 |        "  <tr>\n",
1834 |        "    <th>29</th>\n",
1835 |        "    <th>0.254412</th>\n",
1836 |        "    <th>0.504455</th>\n",
1837 |        "    <th>0.861500</th>\n",
1838 |        "  </tr>\n",
1839 |        "  <tr>\n",
1840 |        "    <th>30</th>\n",
1841 |        "    <th>0.244284</th>\n",
1842 |        "    <th>0.502431</th>\n",
1843 |        "    <th>0.859750</th>\n",
1844 |        "  </tr>\n",
1845 |        "</table>\n"
1846 |       ],
1847 |       "text/plain": [
1848 |        "<IPython.core.display.HTML object>"
1849 |       ]
1850 |      },
1851 |      "metadata": {},
1852 |      "output_type": "display_data"
1853 |     }
1854 |    ],
1855 |    "source": [
1856 |     "with gpu_mem_restore_ctx():\n",
1857 |     "    learn.fit_one_cycle(30)"
1858 |    ]
1859 |   },
1860 |   {
1861 |    "cell_type": "markdown",
1862 |    "metadata": {},
1863 |    "source": [
1864 |     "## Resnet 101 (Mixed Precision)"
1865 |    ]
1866 |   },
1867 |   {
1868 |    "cell_type": "code",
1869 |    "execution_count": 164,
1870 |    "metadata": {},
1871 |    "outputs": [],
1872 |    "source": [
1873 |     "np.random.seed(42)\n",
1874 |     "data = ImageDataBunch.from_folder(path, train=train, valid_pct=0.2,\n",
1875 |     "        ds_tfms=get_transforms(), size=224, num_workers=4, bs=116).normalize(cifar_stats)"
1876 |    ]
1877 |   },
1878 |   {
1879 |    "cell_type": "code",
1880 |    "execution_count": 165,
1881 |    "metadata": {},
1882 |    "outputs": [],
1883 |    "source": [
1884 |     "learn = to_fp16(create_cnn(data, models.resnet101, metrics=accuracy))"
1885 |    ]
1886 |   },
1887 |   {
1888 |    "cell_type": "code",
1889 |    "execution_count": 166,
1890 |    "metadata": {},
1891 |    "outputs": [
1892 |     {
1893 |      "data": {
1894 |       "text/html": [
1895 |        "Total time: 1:22:32 <p><table style='width:300px; margin-bottom:10px'>\n",
1896 |        "  <tr>\n",
1897 |        "    <th>epoch</th>\n",
1898 |        "    <th>train_loss</th>\n",
1899 |        "    <th>valid_loss</th>\n",
1900 |        "    <th>accuracy</th>\n",
1901 |        "  </tr>\n",
1902 |        "  <tr>\n",
1903 |        "    <th>1</th>\n",
1904 |        "    <th>2.356006</th>\n",
1905 |        "    <th>1.653988</th>\n",
1906 |        "    <th>0.594750</th>\n",
1907 |        "  </tr>\n",
1908 |        "  <tr>\n",
1909 |        "    <th>2</th>\n",
1910 |        "    <th>1.458173</th>\n",
1911 |        "    <th>1.015118</th>\n",
1912 |        "    <th>0.709333</th>\n",
1913 |        "  </tr>\n",
1914 |        "  <tr>\n",
1915 |        "    <th>3</th>\n",
1916 |        "    <th>1.113660</th>\n",
1917 |        "    <th>0.803704</th>\n",
1918 |        "    <th>0.758167</th>\n",
1919 |        "  </tr>\n",
1920 |        "  <tr>\n",
1921 |        "    <th>4</th>\n",
1922 |        "    <th>0.931843</th>\n",
1923 |        "    <th>0.710339</th>\n",
1924 |        "    <th>0.782333</th>\n",
1925 |        "  </tr>\n",
1926 |        "  <tr>\n",
1927 |        "    <th>5</th>\n",
1928 |        "    <th>0.844236</th>\n",
1929 |        "    <th>0.666915</th>\n",
1930 |        "    <th>0.791833</th>\n",
1931 |        "  </tr>\n",
1932 |        "  <tr>\n",
1933 |        "    <th>6</th>\n",
1934 |        "    <th>0.785052</th>\n",
1935 |        "    <th>0.625276</th>\n",
1936 |        "    <th>0.807167</th>\n",
1937 |        "  </tr>\n",
1938 |        "  <tr>\n",
1939 |        "    <th>7</th>\n",
1940 |        "    <th>0.725867</th>\n",
1941 |        "    <th>0.601363</th>\n",
1942 |        "    <th>0.815750</th>\n",
1943 |        "  </tr>\n",
1944 |        "  <tr>\n",
1945 |        "    <th>8</th>\n",
1946 |        "    <th>0.704728</th>\n",
1947 |        "    <th>0.592559</th>\n",
1948 |        "    <th>0.817833</th>\n",
1949 |        "  </tr>\n",
1950 |        "  <tr>\n",
1951 |        "    <th>9</th>\n",
1952 |        "    <th>0.656651</th>\n",
1953 |        "    <th>0.576617</th>\n",
1954 |        "    <th>0.821583</th>\n",
1955 |        "  </tr>\n",
1956 |        "  <tr>\n",
1957 |        "    <th>10</th>\n",
1958 |        "    <th>0.599608</th>\n",
1959 |        "    <th>0.572227</th>\n",
1960 |        "    <th>0.822667</th>\n",
1961 |        "  </tr>\n",
1962 |        "  <tr>\n",
1963 |        "    <th>11</th>\n",
1964 |        "    <th>0.569434</th>\n",
1965 |        "    <th>0.566142</th>\n",
1966 |        "    <th>0.828750</th>\n",
1967 |        "  </tr>\n",
1968 |        "  <tr>\n",
1969 |        "    <th>12</th>\n",
1970 |        "    <th>0.538815</th>\n",
1971 |        "    <th>0.539515</th>\n",
1972 |        "    <th>0.835917</th>\n",
1973 |        "  </tr>\n",
1974 |        "  <tr>\n",
1975 |        "    <th>13</th>\n",
1976 |        "    <th>0.508781</th>\n",
1977 |        "    <th>0.536017</th>\n",
1978 |        "    <th>0.835250</th>\n",
1979 |        "  </tr>\n",
1980 |        "  <tr>\n",
1981 |        "    <th>14</th>\n",
1982 |        "    <th>0.471497</th>\n",
1983 |        "    <th>0.535257</th>\n",
1984 |        "    <th>0.839833</th>\n",
1985 |        "  </tr>\n",
1986 |        "  <tr>\n",
1987 |        "    <th>15</th>\n",
1988 |        "    <th>0.455695</th>\n",
1989 |        "    <th>0.540152</th>\n",
1990 |        "    <th>0.836750</th>\n",
1991 |        "  </tr>\n",
1992 |        "  <tr>\n",
1993 |        "    <th>16</th>\n",
1994 |        "    <th>0.437731</th>\n",
1995 |        "    <th>0.531233</th>\n",
1996 |        "    <th>0.841583</th>\n",
1997 |        "  </tr>\n",
1998 |        "  <tr>\n",
1999 |        "    <th>17</th>\n",
2000 |        "    <th>0.398460</th>\n",
2001 |        "    <th>0.528875</th>\n",
2002 |        "    <th>0.847250</th>\n",
2003 |        "  </tr>\n",
2004 |        "  <tr>\n",
2005 |        "    <th>18</th>\n",
2006 |        "    <th>0.370818</th>\n",
2007 |        "    <th>0.521328</th>\n",
2008 |        "    <th>0.851000</th>\n",
2009 |        "  </tr>\n",
2010 |        "  <tr>\n",
2011 |        "    <th>19</th>\n",
2012 |        "    <th>0.361273</th>\n",
2013 |        "    <th>0.524604</th>\n",
2014 |        "    <th>0.851583</th>\n",
2015 |        "  </tr>\n",
2016 |        "  <tr>\n",
2017 |        "    <th>20</th>\n",
2018 |        "    <th>0.328099</th>\n",
2019 |        "    <th>0.516841</th>\n",
2020 |        "    <th>0.851167</th>\n",
2021 |        "  </tr>\n",
2022 |        "  <tr>\n",
2023 |        "    <th>21</th>\n",
2024 |        "    <th>0.309971</th>\n",
2025 |        "    <th>0.510228</th>\n",
2026 |        "    <th>0.854833</th>\n",
2027 |        "  </tr>\n",
2028 |        "  <tr>\n",
2029 |        "    <th>22</th>\n",
2030 |        "    <th>0.300888</th>\n",
2031 |        "    <th>0.518774</th>\n",
2032 |        "    <th>0.852333</th>\n",
2033 |        "  </tr>\n",
2034 |        "  <tr>\n",
2035 |        "    <th>23</th>\n",
2036 |        "    <th>0.289590</th>\n",
2037 |        "    <th>0.506502</th>\n",
2038 |        "    <th>0.856250</th>\n",
2039 |        "  </tr>\n",
2040 |        "  <tr>\n",
2041 |        "    <th>24</th>\n",
2042 |        "    <th>0.258556</th>\n",
2043 |        "    <th>0.513164</th>\n",
2044 |        "    <th>0.855417</th>\n",
2045 |        "  </tr>\n",
2046 |        "  <tr>\n",
2047 |        "    <th>25</th>\n",
2048 |        "    <th>0.254689</th>\n",
2049 |        "    <th>0.512996</th>\n",
2050 |        "    <th>0.857583</th>\n",
2051 |        "  </tr>\n",
2052 |        "  <tr>\n",
2053 |        "    <th>26</th>\n",
2054 |        "    <th>0.232998</th>\n",
2055 |        "    <th>0.516241</th>\n",
2056 |        "    <th>0.857750</th>\n",
2057 |        "  </tr>\n",
2058 |        "  <tr>\n",
2059 |        "    <th>27</th>\n",
2060 |        "    <th>0.238141</th>\n",
2061 |        "    <th>0.514106</th>\n",
2062 |        "    <th>0.857417</th>\n",
2063 |        "  </tr>\n",
2064 |        "  <tr>\n",
2065 |        "    <th>28</th>\n",
2066 |        "    <th>0.217220</th>\n",
2067 |        "    <th>0.514888</th>\n",
2068 |        "    <th>0.856250</th>\n",
2069 |        "  </tr>\n",
2070 |        "  <tr>\n",
2071 |        "    <th>29</th>\n",
2072 |        "    <th>0.224265</th>\n",
2073 |        "    <th>0.513912</th>\n",
2074 |        "    <th>0.856167</th>\n",
2075 |        "  </tr>\n",
2076 |        "  <tr>\n",
2077 |        "    <th>30</th>\n",
2078 |        "    <th>0.220724</th>\n",
2079 |        "    <th>0.514018</th>\n",
2080 |        "    <th>0.856750</th>\n",
2081 |        "  </tr>\n",
2082 |        "</table>\n"
2083 |       ],
2084 |       "text/plain": [
2085 |        "<IPython.core.display.HTML object>"
2086 |       ]
2087 |      },
2088 |      "metadata": {},
2089 |      "output_type": "display_data"
2090 |     }
2091 |    ],
2092 |    "source": [
2093 |     "with gpu_mem_restore_ctx():\n",
2094 |     "    learn.fit_one_cycle(30)"
2095 |    ]
2096 |   },
2097 |   {
2098 |    "cell_type": "markdown",
2099 |    "metadata": {},
2100 |    "source": [
2101 |     "## Resnet 152"
2102 |    ]
2103 |   },
2104 |   {
2105 |    "cell_type": "code",
2106 |    "execution_count": 65,
2107 |    "metadata": {},
2108 |    "outputs": [],
2109 |    "source": [
2110 |     "np.random.seed(42)\n",
2111 |     "data = ImageDataBunch.from_folder(path, train=train, valid_pct=0.2,\n",
2112 |     "        ds_tfms=get_transforms(), size=224, num_workers=4, bs=48).normalize(cifar_stats)"
2113 |    ]
2114 |   },
2115 |   {
2116 |    "cell_type": "code",
2117 |    "execution_count": 66,
2118 |    "metadata": {},
2119 |    "outputs": [],
2120 |    "source": [
2121 |     "learn = create_cnn(data, models.resnet152, metrics=accuracy)"
2122 |    ]
2123 |   },
2124 |   {
2125 |    "cell_type": "code",
2126 |    "execution_count": 67,
2127 |    "metadata": {},
2128 |    "outputs": [
2129 |     {
2130 |      "data": {
2131 |       "text/html": [
2132 |        "Total time: 2:58:27 <p><table style='width:300px; margin-bottom:10px'>\n",
2133 |        "  <tr>\n",
2134 |        "    <th>epoch</th>\n",
2135 |        "    <th>train_loss</th>\n",
2136 |        "    <th>valid_loss</th>\n",
2137 |        "    <th>accuracy</th>\n",
2138 |        "  </tr>\n",
2139 |        "  <tr>\n",
2140 |        "    <th>1</th>\n",
2141 |        "    <th>1.898828</th>\n",
2142 |        "    <th>1.305966</th>\n",
2143 |        "    <th>0.653833</th>\n",
2144 |        "  </tr>\n",
2145 |        "  <tr>\n",
2146 |        "    <th>2</th>\n",
2147 |        "    <th>1.175181</th>\n",
2148 |        "    <th>0.842290</th>\n",
2149 |        "    <th>0.747250</th>\n",
2150 |        "  </tr>\n",
2151 |        "  <tr>\n",
2152 |        "    <th>3</th>\n",
2153 |        "    <th>0.966224</th>\n",
2154 |        "    <th>0.699221</th>\n",
2155 |        "    <th>0.786167</th>\n",
2156 |        "  </tr>\n",
2157 |        "  <tr>\n",
2158 |        "    <th>4</th>\n",
2159 |        "    <th>0.884034</th>\n",
2160 |        "    <th>0.651086</th>\n",
2161 |        "    <th>0.795667</th>\n",
2162 |        "  </tr>\n",
2163 |        "  <tr>\n",
2164 |        "    <th>5</th>\n",
2165 |        "    <th>0.792795</th>\n",
2166 |        "    <th>0.621682</th>\n",
2167 |        "    <th>0.804750</th>\n",
2168 |        "  </tr>\n",
2169 |        "  <tr>\n",
2170 |        "    <th>6</th>\n",
2171 |        "    <th>0.795135</th>\n",
2172 |        "    <th>0.593315</th>\n",
2173 |        "    <th>0.819083</th>\n",
2174 |        "  </tr>\n",
2175 |        "  <tr>\n",
2176 |        "    <th>7</th>\n",
2177 |        "    <th>0.723622</th>\n",
2178 |        "    <th>0.596839</th>\n",
2179 |        "    <th>0.816833</th>\n",
2180 |        "  </tr>\n",
2181 |        "  <tr>\n",
2182 |        "    <th>8</th>\n",
2183 |        "    <th>0.658714</th>\n",
2184 |        "    <th>0.581674</th>\n",
2185 |        "    <th>0.821417</th>\n",
2186 |        "  </tr>\n",
2187 |        "  <tr>\n",
2188 |        "    <th>9</th>\n",
2189 |        "    <th>0.670455</th>\n",
2190 |        "    <th>0.553189</th>\n",
2191 |        "    <th>0.833000</th>\n",
2192 |        "  </tr>\n",
2193 |        "  <tr>\n",
2194 |        "    <th>10</th>\n",
2195 |        "    <th>0.622607</th>\n",
2196 |        "    <th>0.560935</th>\n",
2197 |        "    <th>0.830250</th>\n",
2198 |        "  </tr>\n",
2199 |        "  <tr>\n",
2200 |        "    <th>11</th>\n",
2201 |        "    <th>0.587018</th>\n",
2202 |        "    <th>0.533542</th>\n",
2203 |        "    <th>0.836250</th>\n",
2204 |        "  </tr>\n",
2205 |        "  <tr>\n",
2206 |        "    <th>12</th>\n",
2207 |        "    <th>0.580308</th>\n",
2208 |        "    <th>0.520549</th>\n",
2209 |        "    <th>0.841167</th>\n",
2210 |        "  </tr>\n",
2211 |        "  <tr>\n",
2212 |        "    <th>13</th>\n",
2213 |        "    <th>0.521300</th>\n",
2214 |        "    <th>0.532823</th>\n",
2215 |        "    <th>0.840583</th>\n",
2216 |        "  </tr>\n",
2217 |        "  <tr>\n",
2218 |        "    <th>14</th>\n",
2219 |        "    <th>0.487249</th>\n",
2220 |        "    <th>0.526109</th>\n",
2221 |        "    <th>0.842750</th>\n",
2222 |        "  </tr>\n",
2223 |        "  <tr>\n",
2224 |        "    <th>15</th>\n",
2225 |        "    <th>0.448812</th>\n",
2226 |        "    <th>0.513343</th>\n",
2227 |        "    <th>0.850167</th>\n",
2228 |        "  </tr>\n",
2229 |        "  <tr>\n",
2230 |        "    <th>16</th>\n",
2231 |        "    <th>0.431717</th>\n",
2232 |        "    <th>0.506566</th>\n",
2233 |        "    <th>0.850667</th>\n",
2234 |        "  </tr>\n",
2235 |        "  <tr>\n",
2236 |        "    <th>17</th>\n",
2237 |        "    <th>0.401086</th>\n",
2238 |        "    <th>0.506412</th>\n",
2239 |        "    <th>0.851917</th>\n",
2240 |        "  </tr>\n",
2241 |        "  <tr>\n",
2242 |        "    <th>18</th>\n",
2243 |        "    <th>0.398720</th>\n",
2244 |        "    <th>0.505510</th>\n",
2245 |        "    <th>0.853083</th>\n",
2246 |        "  </tr>\n",
2247 |        "  <tr>\n",
2248 |        "    <th>19</th>\n",
2249 |        "    <th>0.361149</th>\n",
2250 |        "    <th>0.500569</th>\n",
2251 |        "    <th>0.855833</th>\n",
2252 |        "  </tr>\n",
2253 |        "  <tr>\n",
2254 |        "    <th>20</th>\n",
2255 |        "    <th>0.357157</th>\n",
2256 |        "    <th>0.493713</th>\n",
2257 |        "    <th>0.856417</th>\n",
2258 |        "  </tr>\n",
2259 |        "  <tr>\n",
2260 |        "    <th>21</th>\n",
2261 |        "    <th>0.319523</th>\n",
2262 |        "    <th>0.496027</th>\n",
2263 |        "    <th>0.860000</th>\n",
2264 |        "  </tr>\n",
2265 |        "  <tr>\n",
2266 |        "    <th>22</th>\n",
2267 |        "    <th>0.284045</th>\n",
2268 |        "    <th>0.494504</th>\n",
2269 |        "    <th>0.861000</th>\n",
2270 |        "  </tr>\n",
2271 |        "  <tr>\n",
2272 |        "    <th>23</th>\n",
2273 |        "    <th>0.253644</th>\n",
2274 |        "    <th>0.493373</th>\n",
2275 |        "    <th>0.861333</th>\n",
2276 |        "  </tr>\n",
2277 |        "  <tr>\n",
2278 |        "    <th>24</th>\n",
2279 |        "    <th>0.275702</th>\n",
2280 |        "    <th>0.500611</th>\n",
2281 |        "    <th>0.861167</th>\n",
2282 |        "  </tr>\n",
2283 |        "  <tr>\n",
2284 |        "    <th>25</th>\n",
2285 |        "    <th>0.248939</th>\n",
2286 |        "    <th>0.488999</th>\n",
2287 |        "    <th>0.863083</th>\n",
2288 |        "  </tr>\n",
2289 |        "  <tr>\n",
2290 |        "    <th>26</th>\n",
2291 |        "    <th>0.233902</th>\n",
2292 |        "    <th>0.496497</th>\n",
2293 |        "    <th>0.862333</th>\n",
2294 |        "  </tr>\n",
2295 |        "  <tr>\n",
2296 |        "    <th>27</th>\n",
2297 |        "    <th>0.218012</th>\n",
2298 |        "    <th>0.491988</th>\n",
2299 |        "    <th>0.864417</th>\n",
2300 |        "  </tr>\n",
2301 |        "  <tr>\n",
2302 |        "    <th>28</th>\n",
2303 |        "    <th>0.225783</th>\n",
2304 |        "    <th>0.493372</th>\n",
2305 |        "    <th>0.864667</th>\n",
2306 |        "  </tr>\n",
2307 |        "  <tr>\n",
2308 |        "    <th>29</th>\n",
2309 |        "    <th>0.217687</th>\n",
2310 |        "    <th>0.493124</th>\n",
2311 |        "    <th>0.866333</th>\n",
2312 |        "  </tr>\n",
2313 |        "  <tr>\n",
2314 |        "    <th>30</th>\n",
2315 |        "    <th>0.202206</th>\n",
2316 |        "    <th>0.491947</th>\n",
2317 |        "    <th>0.863083</th>\n",
2318 |        "  </tr>\n",
2319 |        "</table>\n"
2320 |       ],
2321 |       "text/plain": [
2322 |        "<IPython.core.display.HTML object>"
2323 |       ]
2324 |      },
2325 |      "metadata": {},
2326 |      "output_type": "display_data"
2327 |     }
2328 |    ],
2329 |    "source": [
2330 |     "with gpu_mem_restore_ctx():\n",
2331 |     "    learn.fit_one_cycle(30)"
2332 |    ]
2333 |   },
2334 |   {
2335 |    "cell_type": "markdown",
2336 |    "metadata": {},
2337 |    "source": [
2338 |     "## Resnet 152 (Mixed Precision)"
2339 |    ]
2340 |   },
2341 |   {
2342 |    "cell_type": "code",
2343 |    "execution_count": 167,
2344 |    "metadata": {},
2345 |    "outputs": [],
2346 |    "source": [
2347 |     "np.random.seed(42)\n",
2348 |     "data = ImageDataBunch.from_folder(path, train=train, valid_pct=0.2,\n",
2349 |     "        ds_tfms=get_transforms(), size=224, num_workers=4, bs=88).normalize(cifar_stats)"
2350 |    ]
2351 |   },
2352 |   {
2353 |    "cell_type": "code",
2354 |    "execution_count": 168,
2355 |    "metadata": {},
2356 |    "outputs": [],
2357 |    "source": [
2358 |     "learn = to_fp16(create_cnn(data, models.resnet152, metrics=accuracy))"
2359 |    ]
2360 |   },
2361 |   {
2362 |    "cell_type": "code",
2363 |    "execution_count": 169,
2364 |    "metadata": {},
2365 |    "outputs": [
2366 |     {
2367 |      "data": {
2368 |       "text/html": [
2369 |        "Total time: 1:51:21 <p><table style='width:300px; margin-bottom:10px'>\n",
2370 |        "  <tr>\n",
2371 |        "    <th>epoch</th>\n",
2372 |        "    <th>train_loss</th>\n",
2373 |        "    <th>valid_loss</th>\n",
2374 |        "    <th>accuracy</th>\n",
2375 |        "  </tr>\n",
2376 |        "  <tr>\n",
2377 |        "    <th>1</th>\n",
2378 |        "    <th>2.160135</th>\n",
2379 |        "    <th>1.495938</th>\n",
2380 |        "    <th>0.625750</th>\n",
2381 |        "  </tr>\n",
2382 |        "  <tr>\n",
2383 |        "    <th>2</th>\n",
2384 |        "    <th>1.294585</th>\n",
2385 |        "    <th>0.920020</th>\n",
2386 |        "    <th>0.733083</th>\n",
2387 |        "  </tr>\n",
2388 |        "  <tr>\n",
2389 |        "    <th>3</th>\n",
2390 |        "    <th>0.984379</th>\n",
2391 |        "    <th>0.727062</th>\n",
2392 |        "    <th>0.776750</th>\n",
2393 |        "  </tr>\n",
2394 |        "  <tr>\n",
2395 |        "    <th>4</th>\n",
2396 |        "    <th>0.865696</th>\n",
2397 |        "    <th>0.641753</th>\n",
2398 |        "    <th>0.800250</th>\n",
2399 |        "  </tr>\n",
2400 |        "  <tr>\n",
2401 |        "    <th>5</th>\n",
2402 |        "    <th>0.777403</th>\n",
2403 |        "    <th>0.599795</th>\n",
2404 |        "    <th>0.815000</th>\n",
2405 |        "  </tr>\n",
2406 |        "  <tr>\n",
2407 |        "    <th>6</th>\n",
2408 |        "    <th>0.726824</th>\n",
2409 |        "    <th>0.607037</th>\n",
2410 |        "    <th>0.811833</th>\n",
2411 |        "  </tr>\n",
2412 |        "  <tr>\n",
2413 |        "    <th>7</th>\n",
2414 |        "    <th>0.695808</th>\n",
2415 |        "    <th>0.577306</th>\n",
2416 |        "    <th>0.824917</th>\n",
2417 |        "  </tr>\n",
2418 |        "  <tr>\n",
2419 |        "    <th>8</th>\n",
2420 |        "    <th>0.663318</th>\n",
2421 |        "    <th>0.561540</th>\n",
2422 |        "    <th>0.828833</th>\n",
2423 |        "  </tr>\n",
2424 |        "  <tr>\n",
2425 |        "    <th>9</th>\n",
2426 |        "    <th>0.601016</th>\n",
2427 |        "    <th>0.554505</th>\n",
2428 |        "    <th>0.833417</th>\n",
2429 |        "  </tr>\n",
2430 |        "  <tr>\n",
2431 |        "    <th>10</th>\n",
2432 |        "    <th>0.576079</th>\n",
2433 |        "    <th>0.546127</th>\n",
2434 |        "    <th>0.834667</th>\n",
2435 |        "  </tr>\n",
2436 |        "  <tr>\n",
2437 |        "    <th>11</th>\n",
2438 |        "    <th>0.549800</th>\n",
2439 |        "    <th>0.524947</th>\n",
2440 |        "    <th>0.842250</th>\n",
2441 |        "  </tr>\n",
2442 |        "  <tr>\n",
2443 |        "    <th>12</th>\n",
2444 |        "    <th>0.494807</th>\n",
2445 |        "    <th>0.513583</th>\n",
2446 |        "    <th>0.844250</th>\n",
2447 |        "  </tr>\n",
2448 |        "  <tr>\n",
2449 |        "    <th>13</th>\n",
2450 |        "    <th>0.475365</th>\n",
2451 |        "    <th>0.516716</th>\n",
2452 |        "    <th>0.846250</th>\n",
2453 |        "  </tr>\n",
2454 |        "  <tr>\n",
2455 |        "    <th>14</th>\n",
2456 |        "    <th>0.443925</th>\n",
2457 |        "    <th>0.506661</th>\n",
2458 |        "    <th>0.849750</th>\n",
2459 |        "  </tr>\n",
2460 |        "  <tr>\n",
2461 |        "    <th>15</th>\n",
2462 |        "    <th>0.410093</th>\n",
2463 |        "    <th>0.509516</th>\n",
2464 |        "    <th>0.849250</th>\n",
2465 |        "  </tr>\n",
2466 |        "  <tr>\n",
2467 |        "    <th>16</th>\n",
2468 |        "    <th>0.378922</th>\n",
2469 |        "    <th>0.499594</th>\n",
2470 |        "    <th>0.855167</th>\n",
2471 |        "  </tr>\n",
2472 |        "  <tr>\n",
2473 |        "    <th>17</th>\n",
2474 |        "    <th>0.344542</th>\n",
2475 |        "    <th>0.489011</th>\n",
2476 |        "    <th>0.856333</th>\n",
2477 |        "  </tr>\n",
2478 |        "  <tr>\n",
2479 |        "    <th>18</th>\n",
2480 |        "    <th>0.340006</th>\n",
2481 |        "    <th>0.507246</th>\n",
2482 |        "    <th>0.852333</th>\n",
2483 |        "  </tr>\n",
2484 |        "  <tr>\n",
2485 |        "    <th>19</th>\n",
2486 |        "    <th>0.325086</th>\n",
2487 |        "    <th>0.486156</th>\n",
2488 |        "    <th>0.861083</th>\n",
2489 |        "  </tr>\n",
2490 |        "  <tr>\n",
2491 |        "    <th>20</th>\n",
2492 |        "    <th>0.302048</th>\n",
2493 |        "    <th>0.490937</th>\n",
2494 |        "    <th>0.857667</th>\n",
2495 |        "  </tr>\n",
2496 |        "  <tr>\n",
2497 |        "    <th>21</th>\n",
2498 |        "    <th>0.288165</th>\n",
2499 |        "    <th>0.486860</th>\n",
2500 |        "    <th>0.864250</th>\n",
2501 |        "  </tr>\n",
2502 |        "  <tr>\n",
2503 |        "    <th>22</th>\n",
2504 |        "    <th>0.256561</th>\n",
2505 |        "    <th>0.492568</th>\n",
2506 |        "    <th>0.860917</th>\n",
2507 |        "  </tr>\n",
2508 |        "  <tr>\n",
2509 |        "    <th>23</th>\n",
2510 |        "    <th>0.237763</th>\n",
2511 |        "    <th>0.485718</th>\n",
2512 |        "    <th>0.862583</th>\n",
2513 |        "  </tr>\n",
2514 |        "  <tr>\n",
2515 |        "    <th>24</th>\n",
2516 |        "    <th>0.224336</th>\n",
2517 |        "    <th>0.486326</th>\n",
2518 |        "    <th>0.863333</th>\n",
2519 |        "  </tr>\n",
2520 |        "  <tr>\n",
2521 |        "    <th>25</th>\n",
2522 |        "    <th>0.209552</th>\n",
2523 |        "    <th>0.485967</th>\n",
2524 |        "    <th>0.863500</th>\n",
2525 |        "  </tr>\n",
2526 |        "  <tr>\n",
2527 |        "    <th>26</th>\n",
2528 |        "    <th>0.210889</th>\n",
2529 |        "    <th>0.485575</th>\n",
2530 |        "    <th>0.865250</th>\n",
2531 |        "  </tr>\n",
2532 |        "  <tr>\n",
2533 |        "    <th>27</th>\n",
2534 |        "    <th>0.202123</th>\n",
2535 |        "    <th>0.481421</th>\n",
2536 |        "    <th>0.865833</th>\n",
2537 |        "  </tr>\n",
2538 |        "  <tr>\n",
2539 |        "    <th>28</th>\n",
2540 |        "    <th>0.205988</th>\n",
2541 |        "    <th>0.486022</th>\n",
2542 |        "    <th>0.867167</th>\n",
2543 |        "  </tr>\n",
2544 |        "  <tr>\n",
2545 |        "    <th>29</th>\n",
2546 |        "    <th>0.173248</th>\n",
2547 |        "    <th>0.485957</th>\n",
2548 |        "    <th>0.865583</th>\n",
2549 |        "  </tr>\n",
2550 |        "  <tr>\n",
2551 |        "    <th>30</th>\n",
2552 |        "    <th>0.182509</th>\n",
2553 |        "    <th>0.484172</th>\n",
2554 |        "    <th>0.866750</th>\n",
2555 |        "  </tr>\n",
2556 |        "</table>\n"
2557 |       ],
2558 |       "text/plain": [
2559 |        "<IPython.core.display.HTML object>"
2560 |       ]
2561 |      },
2562 |      "metadata": {},
2563 |      "output_type": "display_data"
2564 |     }
2565 |    ],
2566 |    "source": [
2567 |     "with gpu_mem_restore_ctx():\n",
2568 |     "    learn.fit_one_cycle(30)"
2569 |    ]
2570 |   }
2571 |  ],
2572 |  "metadata": {
2573 |   "kernelspec": {
2574 |    "display_name": "Python 3",
2575 |    "language": "python",
2576 |    "name": "python3"
2577 |   },
2578 |   "language_info": {
2579 |    "codemirror_mode": {
2580 |     "name": "ipython",
2581 |     "version": 3
2582 |    },
2583 |    "file_extension": ".py",
2584 |    "mimetype": "text/x-python",
2585 |    "name": "python",
2586 |    "nbconvert_exporter": "python",
2587 |    "pygments_lexer": "ipython3",
2588 |    "version": "3.6.6"
2589 |   }
2590 |  },
2591 |  "nbformat": 4,
2592 |  "nbformat_minor": 2
2593 | }
2594 | 


--------------------------------------------------------------------------------
/README.md:
--------------------------------------------------------------------------------
 1 | # RTX-2080Ti-Vs-GTX-1080Ti-CIFAR-100-Benchmarks
 2 | 
 3 | [Scripts for the accompanying blogpost](https://hackernoon.com/rtx-2080ti-vs-gtx-1080ti-fastai-mixed-precision-training-comparisons-on-cifar-100-761d8f615d7f?source=user_profile---------4------------------)
 4 | 
 5 | For more details, please checkout the blogpost above.
 6 | 
 7 | For details on fast.ai and the latest MOOC (2019), please checkout course.fast.ai
 8 | 
 9 | ### Special Thanks 
10 | 
11 | Special Thanks to Tuatini Godard for helping me run the tests on his 1080Ti
12 | 


--------------------------------------------------------------------------------