├── .gitignore
├── 01_the_foundation_of_deep_learning.ipynb
├── 02_pytorch_tutorial.ipynb
├── 03_tensorflow_tutorial.ipynb
├── 04_neural_network_foundation.ipynb
├── 05_neural_network.ipynb
├── 06_deep_neural_network.ipynb
├── 08_cnn_pytorch.ipynb
├── 08_cnn_tensorflow.ipynb
├── 09_rnn_pytorch.ipynb
├── 09_rnn_tensorflow.ipynb
├── LICENSE
├── README.md
├── datasets
    ├── ch01
    │   └── cat.jpg
    ├── ch03
    │   └── 3-3.png
    ├── ch06
    │   ├── test.rar
    │   └── train.rar
    ├── ch08
    │   ├── pytorch
    │   │   └── MNIST
    │   │   │   ├── processed
    │   │   │       ├── test.pt
    │   │   │       └── training.pt
    │   │   │   └── raw
    │   │   │       ├── t10k-images-idx3-ubyte
    │   │   │       ├── t10k-labels-idx1-ubyte
    │   │   │       ├── train-images-idx3-ubyte
    │   │   │       └── train-labels-idx1-ubyte
    │   └── tensorflow
    │   │   └── MNIST
    │   │       ├── t10k-images-idx3-ubyte.gz
    │   │       ├── t10k-labels-idx1-ubyte.gz
    │   │       ├── train-images-idx3-ubyte.gz
    │   │       └── train-labels-idx1-ubyte.gz
    └── readme.md
├── tf_logs
    ├── ch03
    │   └── run-20190704075350
    │   │   └── events.out.tfevents.1562226830.AB-201810292038
    └── readme.md
└── 微信公众号：AI有道.jpg


/.gitignore:
--------------------------------------------------------------------------------
  1 | # Byte-compiled / optimized / DLL files
  2 | __pycache__/
  3 | *.py[cod]
  4 | *$py.class
  5 | 
  6 | # C extensions
  7 | *.so
  8 | 
  9 | # Distribution / packaging
 10 | .Python
 11 | build/
 12 | develop-eggs/
 13 | dist/
 14 | downloads/
 15 | eggs/
 16 | .eggs/
 17 | lib/
 18 | lib64/
 19 | parts/
 20 | sdist/
 21 | var/
 22 | wheels/
 23 | *.egg-info/
 24 | .installed.cfg
 25 | *.egg
 26 | MANIFEST
 27 | 
 28 | # PyInstaller
 29 | #  Usually these files are written by a python script from a template
 30 | #  before PyInstaller builds the exe, so as to inject date/other infos into it.
 31 | *.manifest
 32 | *.spec
 33 | 
 34 | # Installer logs
 35 | pip-log.txt
 36 | pip-delete-this-directory.txt
 37 | 
 38 | # Unit test / coverage reports
 39 | htmlcov/
 40 | .tox/
 41 | .coverage
 42 | .coverage.*
 43 | .cache
 44 | nosetests.xml
 45 | coverage.xml
 46 | *.cover
 47 | .hypothesis/
 48 | .pytest_cache/
 49 | 
 50 | # Translations
 51 | *.mo
 52 | *.pot
 53 | 
 54 | # Django stuff:
 55 | *.log
 56 | local_settings.py
 57 | db.sqlite3
 58 | 
 59 | # Flask stuff:
 60 | instance/
 61 | .webassets-cache
 62 | 
 63 | # Scrapy stuff:
 64 | .scrapy
 65 | 
 66 | # Sphinx documentation
 67 | docs/_build/
 68 | 
 69 | # PyBuilder
 70 | target/
 71 | 
 72 | # Jupyter Notebook
 73 | .ipynb_checkpoints
 74 | 
 75 | # pyenv
 76 | .python-version
 77 | 
 78 | # celery beat schedule file
 79 | celerybeat-schedule
 80 | 
 81 | # SageMath parsed files
 82 | *.sage.py
 83 | 
 84 | # Environments
 85 | .env
 86 | .venv
 87 | env/
 88 | venv/
 89 | ENV/
 90 | env.bak/
 91 | venv.bak/
 92 | 
 93 | # Spyder project settings
 94 | .spyderproject
 95 | .spyproject
 96 | 
 97 | # Rope project settings
 98 | .ropeproject
 99 | 
100 | # mkdocs documentation
101 | /site
102 | 
103 | # mypy
104 | .mypy_cache/
105 | 


--------------------------------------------------------------------------------
/02_pytorch_tutorial.ipynb:
--------------------------------------------------------------------------------
   1 | {
   2 |  "cells": [
   3 |   {
   4 |    "cell_type": "markdown",
   5 |    "metadata": {},
   6 |    "source": [
   7 |     "# 第 2 章：PyTorch"
   8 |    ]
   9 |   },
  10 |   {
  11 |    "cell_type": "code",
  12 |    "execution_count": 1,
  13 |    "metadata": {},
  14 |    "outputs": [
  15 |     {
  16 |      "data": {
  17 |       "text/plain": [
  18 |        "'1.0.1'"
  19 |       ]
  20 |      },
  21 |      "execution_count": 1,
  22 |      "metadata": {},
  23 |      "output_type": "execute_result"
  24 |     }
  25 |    ],
  26 |    "source": [
  27 |     "# 导入PyTorch库\n",
  28 |     "import torch\n",
  29 |     "import torchvision\n",
  30 |     "\n",
  31 |     "# 查看安装的PyTorch版本\n",
  32 |     "torch.__version__"
  33 |    ]
  34 |   },
  35 |   {
  36 |    "cell_type": "markdown",
  37 |    "metadata": {},
  38 |    "source": [
  39 |     "## 张量 Tensor"
  40 |    ]
  41 |   },
  42 |   {
  43 |    "cell_type": "markdown",
  44 |    "metadata": {},
  45 |    "source": [
  46 |     "### 创建 Tensor"
  47 |    ]
  48 |   },
  49 |   {
  50 |    "cell_type": "markdown",
  51 |    "metadata": {},
  52 |    "source": [
  53 |     "创建一个随机初始化的 Tensor"
  54 |    ]
  55 |   },
  56 |   {
  57 |    "cell_type": "code",
  58 |    "execution_count": 2,
  59 |    "metadata": {},
  60 |    "outputs": [
  61 |     {
  62 |      "name": "stdout",
  63 |      "output_type": "stream",
  64 |      "text": [
  65 |       "tensor([[-0.3249,  0.4365],\n",
  66 |       "        [ 0.3976, -0.4804]])\n"
  67 |      ]
  68 |     }
  69 |    ],
  70 |    "source": [
  71 |     "x = torch.randn(2,2)\n",
  72 |     "print(x)"
  73 |    ]
  74 |   },
  75 |   {
  76 |    "cell_type": "markdown",
  77 |    "metadata": {},
  78 |    "source": [
  79 |     "直接把 Python 列表构建成 Tensor"
  80 |    ]
  81 |   },
  82 |   {
  83 |    "cell_type": "code",
  84 |    "execution_count": 3,
  85 |    "metadata": {},
  86 |    "outputs": [
  87 |     {
  88 |      "name": "stdout",
  89 |      "output_type": "stream",
  90 |      "text": [
  91 |       "tensor([[1, 2],\n",
  92 |       "        [3, 4]])\n"
  93 |      ]
  94 |     }
  95 |    ],
  96 |    "source": [
  97 |     "x = torch.tensor([[1, 2], [3, 4]])\n",
  98 |     "print(x)"
  99 |    ]
 100 |   },
 101 |   {
 102 |    "cell_type": "markdown",
 103 |    "metadata": {},
 104 |    "source": [
 105 |     "创建一个全零 Tensor"
 106 |    ]
 107 |   },
 108 |   {
 109 |    "cell_type": "code",
 110 |    "execution_count": 4,
 111 |    "metadata": {},
 112 |    "outputs": [
 113 |     {
 114 |      "name": "stdout",
 115 |      "output_type": "stream",
 116 |      "text": [
 117 |       "tensor([[0., 0.],\n",
 118 |       "        [0., 0.]])\n"
 119 |      ]
 120 |     }
 121 |    ],
 122 |    "source": [
 123 |     "x = torch.zeros(2,2)\n",
 124 |     "print(x)"
 125 |    ]
 126 |   },
 127 |   {
 128 |    "cell_type": "markdown",
 129 |    "metadata": {},
 130 |    "source": [
 131 |     "基于现有的 Tensor 创建新的 Tensor"
 132 |    ]
 133 |   },
 134 |   {
 135 |    "cell_type": "code",
 136 |    "execution_count": 5,
 137 |    "metadata": {},
 138 |    "outputs": [
 139 |     {
 140 |      "name": "stdout",
 141 |      "output_type": "stream",
 142 |      "text": [
 143 |       "tensor([[0., 0.],\n",
 144 |       "        [0., 0.]])\n",
 145 |       "tensor([[1., 1.],\n",
 146 |       "        [1., 1.]])\n"
 147 |      ]
 148 |     }
 149 |    ],
 150 |    "source": [
 151 |     "x = torch.zeros(2,2)\n",
 152 |     "y = torch.ones_like(x)\n",
 153 |     "print(x)\n",
 154 |     "print(y)"
 155 |    ]
 156 |   },
 157 |   {
 158 |    "cell_type": "markdown",
 159 |    "metadata": {},
 160 |    "source": [
 161 |     "指定 Tensor 数据类型"
 162 |    ]
 163 |   },
 164 |   {
 165 |    "cell_type": "code",
 166 |    "execution_count": 6,
 167 |    "metadata": {},
 168 |    "outputs": [
 169 |     {
 170 |      "name": "stdout",
 171 |      "output_type": "stream",
 172 |      "text": [
 173 |       "tensor([[1, 1],\n",
 174 |       "        [1, 1]])\n"
 175 |      ]
 176 |     }
 177 |    ],
 178 |    "source": [
 179 |     "x = torch.ones(2, 2, dtype=torch.long)\n",
 180 |     "print(x)"
 181 |    ]
 182 |   },
 183 |   {
 184 |    "cell_type": "markdown",
 185 |    "metadata": {},
 186 |    "source": [
 187 |     "### Tensor 的数学运算"
 188 |    ]
 189 |   },
 190 |   {
 191 |    "cell_type": "markdown",
 192 |    "metadata": {},
 193 |    "source": [
 194 |     "两个 Tensor 相加"
 195 |    ]
 196 |   },
 197 |   {
 198 |    "cell_type": "code",
 199 |    "execution_count": 7,
 200 |    "metadata": {},
 201 |    "outputs": [
 202 |     {
 203 |      "name": "stdout",
 204 |      "output_type": "stream",
 205 |      "text": [
 206 |       "tensor([[2., 2.],\n",
 207 |       "        [2., 2.]])\n"
 208 |      ]
 209 |     }
 210 |    ],
 211 |    "source": [
 212 |     "x = torch.ones(2,2)\n",
 213 |     "y = torch.ones(2,2)\n",
 214 |     "z = x + y\n",
 215 |     "print(z)"
 216 |    ]
 217 |   },
 218 |   {
 219 |    "cell_type": "markdown",
 220 |    "metadata": {},
 221 |    "source": [
 222 |     "也可以使用torch.add()实现Tensor相加："
 223 |    ]
 224 |   },
 225 |   {
 226 |    "cell_type": "code",
 227 |    "execution_count": 8,
 228 |    "metadata": {},
 229 |    "outputs": [
 230 |     {
 231 |      "name": "stdout",
 232 |      "output_type": "stream",
 233 |      "text": [
 234 |       "tensor([[2., 2.],\n",
 235 |       "        [2., 2.]])\n"
 236 |      ]
 237 |     }
 238 |    ],
 239 |    "source": [
 240 |     "x = torch.ones(2,2)\n",
 241 |     "y = torch.ones(2,2)\n",
 242 |     "z = torch.add(x, y)\n",
 243 |     "print(z)"
 244 |    ]
 245 |   },
 246 |   {
 247 |    "cell_type": "markdown",
 248 |    "metadata": {},
 249 |    "source": [
 250 |     "还可以使用._add()实现替换："
 251 |    ]
 252 |   },
 253 |   {
 254 |    "cell_type": "code",
 255 |    "execution_count": 9,
 256 |    "metadata": {},
 257 |    "outputs": [
 258 |     {
 259 |      "name": "stdout",
 260 |      "output_type": "stream",
 261 |      "text": [
 262 |       "tensor([[2., 2.],\n",
 263 |       "        [2., 2.]])\n"
 264 |      ]
 265 |     }
 266 |    ],
 267 |    "source": [
 268 |     "x = torch.ones(2,2)\n",
 269 |     "y = torch.ones(2,2)\n",
 270 |     "y.add_(x)\n",
 271 |     "print(y)"
 272 |    ]
 273 |   },
 274 |   {
 275 |    "cell_type": "markdown",
 276 |    "metadata": {},
 277 |    "source": [
 278 |     "Tenosr乘法有两种形式，第一种是对应元素相乘："
 279 |    ]
 280 |   },
 281 |   {
 282 |    "cell_type": "code",
 283 |    "execution_count": 10,
 284 |    "metadata": {},
 285 |    "outputs": [
 286 |     {
 287 |      "data": {
 288 |       "text/plain": [
 289 |        "tensor([[ 1,  4],\n",
 290 |        "        [ 9, 16]])"
 291 |       ]
 292 |      },
 293 |      "execution_count": 10,
 294 |      "metadata": {},
 295 |      "output_type": "execute_result"
 296 |     }
 297 |    ],
 298 |    "source": [
 299 |     "x = torch.tensor([[1, 2], [3, 4]])\n",
 300 |     "y = torch.tensor([[1, 2], [3, 4]])\n",
 301 |     "x.mul(y)"
 302 |    ]
 303 |   },
 304 |   {
 305 |    "cell_type": "markdown",
 306 |    "metadata": {},
 307 |    "source": [
 308 |     "第二种更常用的是矩阵相乘："
 309 |    ]
 310 |   },
 311 |   {
 312 |    "cell_type": "code",
 313 |    "execution_count": 11,
 314 |    "metadata": {},
 315 |    "outputs": [
 316 |     {
 317 |      "data": {
 318 |       "text/plain": [
 319 |        "tensor([[ 7, 10],\n",
 320 |        "        [15, 22]])"
 321 |       ]
 322 |      },
 323 |      "execution_count": 11,
 324 |      "metadata": {},
 325 |      "output_type": "execute_result"
 326 |     }
 327 |    ],
 328 |    "source": [
 329 |     "x = torch.tensor([[1, 2], [3, 4]])\n",
 330 |     "y = torch.tensor([[1, 2], [3, 4]])\n",
 331 |     "x.mm(y)"
 332 |    ]
 333 |   },
 334 |   {
 335 |    "cell_type": "markdown",
 336 |    "metadata": {},
 337 |    "source": [
 338 |     "### Tensor 与 NumPy"
 339 |    ]
 340 |   },
 341 |   {
 342 |    "cell_type": "code",
 343 |    "execution_count": 12,
 344 |    "metadata": {},
 345 |    "outputs": [],
 346 |    "source": [
 347 |     "# 导入numpy\n",
 348 |     "import numpy as np"
 349 |    ]
 350 |   },
 351 |   {
 352 |    "cell_type": "markdown",
 353 |    "metadata": {},
 354 |    "source": [
 355 |     "#### Tensor to NumPy"
 356 |    ]
 357 |   },
 358 |   {
 359 |    "cell_type": "code",
 360 |    "execution_count": 13,
 361 |    "metadata": {},
 362 |    "outputs": [
 363 |     {
 364 |      "name": "stdout",
 365 |      "output_type": "stream",
 366 |      "text": [
 367 |       "<class 'torch.Tensor'>\n",
 368 |       "<class 'numpy.ndarray'>\n"
 369 |      ]
 370 |     }
 371 |    ],
 372 |    "source": [
 373 |     "a = torch.ones(2,2)\n",
 374 |     "b = a.numpy()\n",
 375 |     "print(type(a))\n",
 376 |     "print(type(b))"
 377 |    ]
 378 |   },
 379 |   {
 380 |    "cell_type": "markdown",
 381 |    "metadata": {},
 382 |    "source": [
 383 |     "此时，如果Tensor发生改变，对应的NumPy数组也有相同的变化。"
 384 |    ]
 385 |   },
 386 |   {
 387 |    "cell_type": "code",
 388 |    "execution_count": 14,
 389 |    "metadata": {},
 390 |    "outputs": [
 391 |     {
 392 |      "name": "stdout",
 393 |      "output_type": "stream",
 394 |      "text": [
 395 |       "tensor([[2., 2.],\n",
 396 |       "        [2., 2.]])\n",
 397 |       "[[2. 2.]\n",
 398 |       " [2. 2.]]\n"
 399 |      ]
 400 |     }
 401 |    ],
 402 |    "source": [
 403 |     "a.add_(1)\n",
 404 |     "print(a)\n",
 405 |     "print(b)"
 406 |    ]
 407 |   },
 408 |   {
 409 |    "cell_type": "markdown",
 410 |    "metadata": {},
 411 |    "source": [
 412 |     "#### NumPy to Tensor"
 413 |    ]
 414 |   },
 415 |   {
 416 |    "cell_type": "code",
 417 |    "execution_count": 15,
 418 |    "metadata": {},
 419 |    "outputs": [
 420 |     {
 421 |      "name": "stdout",
 422 |      "output_type": "stream",
 423 |      "text": [
 424 |       "<class 'numpy.ndarray'>\n",
 425 |       "<class 'torch.Tensor'>\n"
 426 |      ]
 427 |     }
 428 |    ],
 429 |    "source": [
 430 |     "a = np.array([[1, 1], [1, 1]])\n",
 431 |     "b = torch.from_numpy(a)\n",
 432 |     "print(type(a))\n",
 433 |     "print(type(b))"
 434 |    ]
 435 |   },
 436 |   {
 437 |    "cell_type": "markdown",
 438 |    "metadata": {},
 439 |    "source": [
 440 |     "如果NumPy数组发生改变，对应的Tensor也有相同的变化。"
 441 |    ]
 442 |   },
 443 |   {
 444 |    "cell_type": "code",
 445 |    "execution_count": 16,
 446 |    "metadata": {},
 447 |    "outputs": [
 448 |     {
 449 |      "name": "stdout",
 450 |      "output_type": "stream",
 451 |      "text": [
 452 |       "[[2 2]\n",
 453 |       " [2 2]]\n",
 454 |       "tensor([[2, 2],\n",
 455 |       "        [2, 2]], dtype=torch.int32)\n"
 456 |      ]
 457 |     }
 458 |    ],
 459 |    "source": [
 460 |     "np.add(a, 1, out=a)\n",
 461 |     "print(a)\n",
 462 |     "print(b)"
 463 |    ]
 464 |   },
 465 |   {
 466 |    "cell_type": "markdown",
 467 |    "metadata": {},
 468 |    "source": [
 469 |     "### CUDA Tensor"
 470 |    ]
 471 |   },
 472 |   {
 473 |    "cell_type": "code",
 474 |    "execution_count": 17,
 475 |    "metadata": {},
 476 |    "outputs": [],
 477 |    "source": [
 478 |     "a = torch.ones(2,2)\n",
 479 |     "# 检查是否可以使用GPU\n",
 480 |     "if torch.cuda.is_available():\n",
 481 |     "    a_cuda = a.cuda()\n",
 482 |     "    print(a_cuda)"
 483 |    ]
 484 |   },
 485 |   {
 486 |    "cell_type": "markdown",
 487 |    "metadata": {},
 488 |    "source": [
 489 |     "因为我们安装的是CPU版本的PyTorch，所以这里不会执行if语句。如果安装了GPU，a_cuda的打印结果如下：\n",
 490 |     "\n",
 491 |     "```\n",
 492 |     "tensor([[1., 1.],\n",
 493 |     "        [1., 1.]], device='cuda:0')\n",
 494 |     "```"
 495 |    ]
 496 |   },
 497 |   {
 498 |    "cell_type": "markdown",
 499 |    "metadata": {},
 500 |    "source": [
 501 |     "## 自动求导 autograd"
 502 |    ]
 503 |   },
 504 |   {
 505 |    "cell_type": "markdown",
 506 |    "metadata": {},
 507 |    "source": [
 508 |     "定义 Tensor x，设置参数`tensor.requries_grad=True`"
 509 |    ]
 510 |   },
 511 |   {
 512 |    "cell_type": "code",
 513 |    "execution_count": 18,
 514 |    "metadata": {},
 515 |    "outputs": [
 516 |     {
 517 |      "name": "stdout",
 518 |      "output_type": "stream",
 519 |      "text": [
 520 |       "True\n",
 521 |       "True\n"
 522 |      ]
 523 |     }
 524 |    ],
 525 |    "source": [
 526 |     "x = torch.ones(2, 2, requires_grad=True)\n",
 527 |     "y = torch.ones(2, 2, requires_grad=True)\n",
 528 |     "print(x.requires_grad)\n",
 529 |     "print(y.requires_grad)"
 530 |    ]
 531 |   },
 532 |   {
 533 |    "cell_type": "markdown",
 534 |    "metadata": {},
 535 |    "source": [
 536 |     "### 当输出是标量时"
 537 |    ]
 538 |   },
 539 |   {
 540 |    "cell_type": "markdown",
 541 |    "metadata": {},
 542 |    "source": [
 543 |     "定义输出$z=\\frac14\\sum_ix_i+y_i$"
 544 |    ]
 545 |   },
 546 |   {
 547 |    "cell_type": "code",
 548 |    "execution_count": 19,
 549 |    "metadata": {},
 550 |    "outputs": [
 551 |     {
 552 |      "name": "stdout",
 553 |      "output_type": "stream",
 554 |      "text": [
 555 |       "tensor(2., grad_fn=<MeanBackward1>)\n"
 556 |      ]
 557 |     }
 558 |    ],
 559 |    "source": [
 560 |     "z = x + y\n",
 561 |     "z = z.mean()\n",
 562 |     "print(z)"
 563 |    ]
 564 |   },
 565 |   {
 566 |    "cell_type": "markdown",
 567 |    "metadata": {},
 568 |    "source": [
 569 |     "反向传播"
 570 |    ]
 571 |   },
 572 |   {
 573 |    "cell_type": "code",
 574 |    "execution_count": 20,
 575 |    "metadata": {},
 576 |    "outputs": [],
 577 |    "source": [
 578 |     "z.backward()"
 579 |    ]
 580 |   },
 581 |   {
 582 |    "cell_type": "markdown",
 583 |    "metadata": {},
 584 |    "source": [
 585 |     "计算$\\frac{\\partial z}{\\partial x}$和$\\frac{\\partial z}{\\partial y}$"
 586 |    ]
 587 |   },
 588 |   {
 589 |    "cell_type": "code",
 590 |    "execution_count": 21,
 591 |    "metadata": {},
 592 |    "outputs": [
 593 |     {
 594 |      "name": "stdout",
 595 |      "output_type": "stream",
 596 |      "text": [
 597 |       "tensor([[0.2500, 0.2500],\n",
 598 |       "        [0.2500, 0.2500]])\n"
 599 |      ]
 600 |     }
 601 |    ],
 602 |    "source": [
 603 |     "print(x.grad)"
 604 |    ]
 605 |   },
 606 |   {
 607 |    "cell_type": "code",
 608 |    "execution_count": 22,
 609 |    "metadata": {},
 610 |    "outputs": [
 611 |     {
 612 |      "name": "stdout",
 613 |      "output_type": "stream",
 614 |      "text": [
 615 |       "tensor([[0.2500, 0.2500],\n",
 616 |       "        [0.2500, 0.2500]])\n"
 617 |      ]
 618 |     }
 619 |    ],
 620 |    "source": [
 621 |     "print(y.grad)"
 622 |    ]
 623 |   },
 624 |   {
 625 |    "cell_type": "markdown",
 626 |    "metadata": {},
 627 |    "source": [
 628 |     "### 当输出是多维张量时"
 629 |    ]
 630 |   },
 631 |   {
 632 |    "cell_type": "markdown",
 633 |    "metadata": {},
 634 |    "source": [
 635 |     "定义输出 $z=2x+3y$"
 636 |    ]
 637 |   },
 638 |   {
 639 |    "cell_type": "code",
 640 |    "execution_count": 23,
 641 |    "metadata": {},
 642 |    "outputs": [
 643 |     {
 644 |      "name": "stdout",
 645 |      "output_type": "stream",
 646 |      "text": [
 647 |       "tensor([[5., 5.],\n",
 648 |       "        [5., 5.]], grad_fn=<AddBackward0>)\n"
 649 |      ]
 650 |     }
 651 |    ],
 652 |    "source": [
 653 |     "x = torch.ones(2, 2, requires_grad=True)\n",
 654 |     "y = torch.ones(2, 2, requires_grad=True)\n",
 655 |     "z = 2 * x + 3 * y\n",
 656 |     "print(z)"
 657 |    ]
 658 |   },
 659 |   {
 660 |    "cell_type": "markdown",
 661 |    "metadata": {},
 662 |    "source": [
 663 |     "反向传播"
 664 |    ]
 665 |   },
 666 |   {
 667 |    "cell_type": "code",
 668 |    "execution_count": 24,
 669 |    "metadata": {},
 670 |    "outputs": [],
 671 |    "source": [
 672 |     "z.backward(torch.ones_like(z))"
 673 |    ]
 674 |   },
 675 |   {
 676 |    "cell_type": "markdown",
 677 |    "metadata": {},
 678 |    "source": [
 679 |     "计算$\\frac{\\partial z}{\\partial x}$和$\\frac{\\partial z}{\\partial y}$"
 680 |    ]
 681 |   },
 682 |   {
 683 |    "cell_type": "code",
 684 |    "execution_count": 25,
 685 |    "metadata": {},
 686 |    "outputs": [
 687 |     {
 688 |      "name": "stdout",
 689 |      "output_type": "stream",
 690 |      "text": [
 691 |       "tensor([[2., 2.],\n",
 692 |       "        [2., 2.]])\n"
 693 |      ]
 694 |     }
 695 |    ],
 696 |    "source": [
 697 |     "print(x.grad)"
 698 |    ]
 699 |   },
 700 |   {
 701 |    "cell_type": "code",
 702 |    "execution_count": 26,
 703 |    "metadata": {},
 704 |    "outputs": [
 705 |     {
 706 |      "name": "stdout",
 707 |      "output_type": "stream",
 708 |      "text": [
 709 |       "tensor([[3., 3.],\n",
 710 |       "        [3., 3.]])\n"
 711 |      ]
 712 |     }
 713 |    ],
 714 |    "source": [
 715 |     "print(y.grad)"
 716 |    ]
 717 |   },
 718 |   {
 719 |    "cell_type": "markdown",
 720 |    "metadata": {},
 721 |    "source": [
 722 |     "### 禁止自动求导"
 723 |    ]
 724 |   },
 725 |   {
 726 |    "cell_type": "code",
 727 |    "execution_count": 27,
 728 |    "metadata": {},
 729 |    "outputs": [
 730 |     {
 731 |      "name": "stdout",
 732 |      "output_type": "stream",
 733 |      "text": [
 734 |       "True\n",
 735 |       "True\n",
 736 |       "False\n"
 737 |      ]
 738 |     }
 739 |    ],
 740 |    "source": [
 741 |     "print(x.requires_grad)\n",
 742 |     "print((2 * x).requires_grad)\n",
 743 |     "\n",
 744 |     "with torch.no_grad():\n",
 745 |     "    print((2 * x).requires_grad)"
 746 |    ]
 747 |   },
 748 |   {
 749 |    "cell_type": "markdown",
 750 |    "metadata": {},
 751 |    "source": [
 752 |     "## 神经网络包 nn 和优化器 optim"
 753 |    ]
 754 |   },
 755 |   {
 756 |    "cell_type": "markdown",
 757 |    "metadata": {},
 758 |    "source": [
 759 |     "### torch.nn"
 760 |    ]
 761 |   },
 762 |   {
 763 |    "cell_type": "code",
 764 |    "execution_count": 28,
 765 |    "metadata": {},
 766 |    "outputs": [],
 767 |    "source": [
 768 |     "import torch.nn as nn"
 769 |    ]
 770 |   },
 771 |   {
 772 |    "cell_type": "code",
 773 |    "execution_count": 29,
 774 |    "metadata": {},
 775 |    "outputs": [],
 776 |    "source": [
 777 |     "class net_name(nn.Module):\n",
 778 |     "    def __init__(self):\n",
 779 |     "        super(net_name, self).__init__()\n",
 780 |     "        self.fc = nn.Linear(1, 1)\n",
 781 |     "        # 其它层\n",
 782 |     "    \n",
 783 |     "    def forward(self, x):\n",
 784 |     "        out = self.fc(x)\n",
 785 |     "        return out"
 786 |    ]
 787 |   },
 788 |   {
 789 |    "cell_type": "markdown",
 790 |    "metadata": {},
 791 |    "source": [
 792 |     "新建一个该模型的对象"
 793 |    ]
 794 |   },
 795 |   {
 796 |    "cell_type": "code",
 797 |    "execution_count": 30,
 798 |    "metadata": {},
 799 |    "outputs": [],
 800 |    "source": [
 801 |     "net = net_name()"
 802 |    ]
 803 |   },
 804 |   {
 805 |    "cell_type": "markdown",
 806 |    "metadata": {},
 807 |    "source": [
 808 |     "### torch.optim"
 809 |    ]
 810 |   },
 811 |   {
 812 |    "cell_type": "code",
 813 |    "execution_count": 31,
 814 |    "metadata": {},
 815 |    "outputs": [],
 816 |    "source": [
 817 |     "import torch.optim as optim"
 818 |    ]
 819 |   },
 820 |   {
 821 |    "cell_type": "markdown",
 822 |    "metadata": {},
 823 |    "source": [
 824 |     "计算预测值与真实值的均方误差"
 825 |    ]
 826 |   },
 827 |   {
 828 |    "cell_type": "markdown",
 829 |    "metadata": {},
 830 |    "source": [
 831 |     "```\n",
 832 |     "criterion = nn.MSELoss()\n",
 833 |     "loss = criterion(output, target)\n",
 834 |     "```"
 835 |    ]
 836 |   },
 837 |   {
 838 |    "cell_type": "markdown",
 839 |    "metadata": {},
 840 |    "source": [
 841 |     "使用随机梯度下降（SGD)优化"
 842 |    ]
 843 |   },
 844 |   {
 845 |    "cell_type": "code",
 846 |    "execution_count": null,
 847 |    "metadata": {},
 848 |    "outputs": [],
 849 |    "source": []
 850 |   },
 851 |   {
 852 |    "cell_type": "markdown",
 853 |    "metadata": {},
 854 |    "source": [
 855 |     "```\n",
 856 |     "optimizer = optim.SGD(net.parameters(), lr=0.01)\n",
 857 |     "```"
 858 |    ]
 859 |   },
 860 |   {
 861 |    "cell_type": "markdown",
 862 |    "metadata": {},
 863 |    "source": [
 864 |     "单次迭代对应的代码为"
 865 |    ]
 866 |   },
 867 |   {
 868 |    "cell_type": "markdown",
 869 |    "metadata": {},
 870 |    "source": [
 871 |     "```\n",
 872 |     "optimizer.zero_grad()    # 梯度清零\n",
 873 |     "output = net(input)\n",
 874 |     "loss = criterion(output, target)\n",
 875 |     "loss.backward()\n",
 876 |     "optimizer.step()    # 完成更新\n",
 877 |     "\n",
 878 |     "```"
 879 |    ]
 880 |   },
 881 |   {
 882 |    "cell_type": "markdown",
 883 |    "metadata": {},
 884 |    "source": [
 885 |     "## PyTorch 线性回归"
 886 |    ]
 887 |   },
 888 |   {
 889 |    "cell_type": "markdown",
 890 |    "metadata": {},
 891 |    "source": [
 892 |     "### 创建数据集"
 893 |    ]
 894 |   },
 895 |   {
 896 |    "cell_type": "code",
 897 |    "execution_count": 32,
 898 |    "metadata": {},
 899 |    "outputs": [],
 900 |    "source": [
 901 |     "# y=3x+10，后面加上torch.randn()函数制造噪音\n",
 902 |     "x = torch.unsqueeze(torch.linspace(-1, 1, 50), dim=1)\n",
 903 |     "y = 3 * x + 10 + 0.5 * torch.randn(x.size())"
 904 |    ]
 905 |   },
 906 |   {
 907 |    "cell_type": "markdown",
 908 |    "metadata": {},
 909 |    "source": [
 910 |     "显示数据分布"
 911 |    ]
 912 |   },
 913 |   {
 914 |    "cell_type": "code",
 915 |    "execution_count": 33,
 916 |    "metadata": {},
 917 |    "outputs": [
 918 |     {
 919 |      "data": {
 920 |       "text/plain": [
 921 |        "<Figure size 1000x600 with 1 Axes>"
 922 |       ]
 923 |      },
 924 |      "metadata": {},
 925 |      "output_type": "display_data"
 926 |     }
 927 |    ],
 928 |    "source": [
 929 |     "import matplotlib.pyplot as plt\n",
 930 |     "plt.rcParams['font.sans-serif']=['SimHei']\n",
 931 |     "plt.rcParams['axes.unicode_minus']=False\n",
 932 |     "plt.rcParams['figure.figsize'] = (10.0, 6.0) # set default size of plots\n",
 933 |     "\n",
 934 |     "plt.scatter(x.numpy(), y.numpy())\n",
 935 |     "plt.show()"
 936 |    ]
 937 |   },
 938 |   {
 939 |    "cell_type": "markdown",
 940 |    "metadata": {},
 941 |    "source": [
 942 |     "### 定义模型"
 943 |    ]
 944 |   },
 945 |   {
 946 |    "cell_type": "code",
 947 |    "execution_count": 34,
 948 |    "metadata": {},
 949 |    "outputs": [],
 950 |    "source": [
 951 |     "class LinearRegression(nn.Module):\n",
 952 |     "    def __init__(self):\n",
 953 |     "        super(LinearRegression, self).__init__()\n",
 954 |     "        self.fc = nn.Linear(1, 1)\n",
 955 |     "    \n",
 956 |     "    def forward(self, x):\n",
 957 |     "        out = self.fc(x)\n",
 958 |     "        return out"
 959 |    ]
 960 |   },
 961 |   {
 962 |    "cell_type": "code",
 963 |    "execution_count": 35,
 964 |    "metadata": {},
 965 |    "outputs": [],
 966 |    "source": [
 967 |     "model = LinearRegression()"
 968 |    ]
 969 |   },
 970 |   {
 971 |    "cell_type": "markdown",
 972 |    "metadata": {},
 973 |    "source": [
 974 |     "### 定义 loss 和优化函数"
 975 |    ]
 976 |   },
 977 |   {
 978 |    "cell_type": "code",
 979 |    "execution_count": 36,
 980 |    "metadata": {},
 981 |    "outputs": [],
 982 |    "source": [
 983 |     "# 定义loss和优化函数\n",
 984 |     "criterion = nn.MSELoss()\n",
 985 |     "optimizer = optim.SGD(model.parameters(), lr=5e-3)"
 986 |    ]
 987 |   },
 988 |   {
 989 |    "cell_type": "markdown",
 990 |    "metadata": {},
 991 |    "source": [
 992 |     "### 训练"
 993 |    ]
 994 |   },
 995 |   {
 996 |    "cell_type": "code",
 997 |    "execution_count": 37,
 998 |    "metadata": {},
 999 |    "outputs": [
1000 |     {
1001 |      "name": "stdout",
1002 |      "output_type": "stream",
1003 |      "text": [
1004 |       "Epoch[20/1000], loss: 58.123623\n",
1005 |       "Epoch[40/1000], loss: 39.301388\n",
1006 |       "Epoch[60/1000], loss: 26.664478\n",
1007 |       "Epoch[80/1000], loss: 18.171267\n",
1008 |       "Epoch[100/1000], loss: 12.455207\n",
1009 |       "Epoch[120/1000], loss: 8.601423\n",
1010 |       "Epoch[140/1000], loss: 5.997334\n",
1011 |       "Epoch[160/1000], loss: 4.232640\n",
1012 |       "Epoch[180/1000], loss: 3.032410\n",
1013 |       "Epoch[200/1000], loss: 2.212354\n",
1014 |       "Epoch[220/1000], loss: 1.648835\n",
1015 |       "Epoch[240/1000], loss: 1.258876\n",
1016 |       "Epoch[260/1000], loss: 0.986709\n",
1017 |       "Epoch[280/1000], loss: 0.794804\n",
1018 |       "Epoch[300/1000], loss: 0.657871\n",
1019 |       "Epoch[320/1000], loss: 0.558824\n",
1020 |       "Epoch[340/1000], loss: 0.486085\n",
1021 |       "Epoch[360/1000], loss: 0.431788\n",
1022 |       "Epoch[380/1000], loss: 0.390558\n",
1023 |       "Epoch[400/1000], loss: 0.358707\n",
1024 |       "Epoch[420/1000], loss: 0.333685\n",
1025 |       "Epoch[440/1000], loss: 0.313713\n",
1026 |       "Epoch[460/1000], loss: 0.297540\n",
1027 |       "Epoch[480/1000], loss: 0.284271\n",
1028 |       "Epoch[500/1000], loss: 0.273265\n",
1029 |       "Epoch[520/1000], loss: 0.264049\n",
1030 |       "Epoch[540/1000], loss: 0.256270\n",
1031 |       "Epoch[560/1000], loss: 0.249662\n",
1032 |       "Epoch[580/1000], loss: 0.244020\n",
1033 |       "Epoch[600/1000], loss: 0.239182\n",
1034 |       "Epoch[620/1000], loss: 0.235021\n",
1035 |       "Epoch[640/1000], loss: 0.231432\n",
1036 |       "Epoch[660/1000], loss: 0.228331\n",
1037 |       "Epoch[680/1000], loss: 0.225646\n",
1038 |       "Epoch[700/1000], loss: 0.223320\n",
1039 |       "Epoch[720/1000], loss: 0.221302\n",
1040 |       "Epoch[740/1000], loss: 0.219550\n",
1041 |       "Epoch[760/1000], loss: 0.218029\n",
1042 |       "Epoch[780/1000], loss: 0.216707\n",
1043 |       "Epoch[800/1000], loss: 0.215558\n",
1044 |       "Epoch[820/1000], loss: 0.214558\n",
1045 |       "Epoch[840/1000], loss: 0.213690\n",
1046 |       "Epoch[860/1000], loss: 0.212934\n",
1047 |       "Epoch[880/1000], loss: 0.212276\n",
1048 |       "Epoch[900/1000], loss: 0.211704\n",
1049 |       "Epoch[920/1000], loss: 0.211207\n",
1050 |       "Epoch[940/1000], loss: 0.210774\n",
1051 |       "Epoch[960/1000], loss: 0.210397\n",
1052 |       "Epoch[980/1000], loss: 0.210070\n",
1053 |       "Epoch[1000/1000], loss: 0.209784\n"
1054 |      ]
1055 |     }
1056 |    ],
1057 |    "source": [
1058 |     "num_epochs = 1000    # 遍历整个训练集的次数\n",
1059 |     "for epoch in range(num_epochs):\n",
1060 |     "    # forward\n",
1061 |     "    out = model(x)    #前向传播\n",
1062 |     "    loss = criterion(out, y)    #计算loss\n",
1063 |     "    # backward\n",
1064 |     "    optimizer.zero_grad()    #梯度归零\n",
1065 |     "    loss.backward()          #反向传播\n",
1066 |     "    optimizer.step()         #更新参数\n",
1067 |     " \n",
1068 |     "    if (epoch+1) % 20 == 0:\n",
1069 |     "        print('Epoch[{}/{}], loss: {:.6f}'.format(epoch+1, num_epochs, loss.detach().numpy()))"
1070 |    ]
1071 |   },
1072 |   {
1073 |    "cell_type": "markdown",
1074 |    "metadata": {},
1075 |    "source": [
1076 |     "### 模型测试"
1077 |    ]
1078 |   },
1079 |   {
1080 |    "cell_type": "code",
1081 |    "execution_count": 38,
1082 |    "metadata": {},
1083 |    "outputs": [
1084 |     {
1085 |      "data": {
1086 |       "image/png": "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\n",
1087 |       "text/plain": [
1088 |        "<Figure size 432x288 with 1 Axes>"
1089 |       ]
1090 |      },
1091 |      "metadata": {
1092 |       "needs_background": "light"
1093 |      },
1094 |      "output_type": "display_data"
1095 |     }
1096 |    ],
1097 |    "source": [
1098 |     "model.eval()\n",
1099 |     "y_hat = model(x)\n",
1100 |     "plt.scatter(x.numpy(), y.numpy(), label='原始数据')\n",
1101 |     "plt.plot(x.numpy(), y_hat.detach().numpy(), c='r', label='拟合直线')\n",
1102 |     "# 显示图例\n",
1103 |     "plt.legend() \n",
1104 |     "plt.show()"
1105 |    ]
1106 |   },
1107 |   {
1108 |    "cell_type": "markdown",
1109 |    "metadata": {},
1110 |    "source": [
1111 |     "查看参数"
1112 |    ]
1113 |   },
1114 |   {
1115 |    "cell_type": "code",
1116 |    "execution_count": 39,
1117 |    "metadata": {},
1118 |    "outputs": [
1119 |     {
1120 |      "data": {
1121 |       "text/plain": [
1122 |        "[('fc.weight', Parameter containing:\n",
1123 |        "  tensor([[2.7996]], requires_grad=True)), ('fc.bias', Parameter containing:\n",
1124 |        "  tensor([9.9626], requires_grad=True))]"
1125 |       ]
1126 |      },
1127 |      "execution_count": 39,
1128 |      "metadata": {},
1129 |      "output_type": "execute_result"
1130 |     }
1131 |    ],
1132 |    "source": [
1133 |     "list(model.named_parameters())"
1134 |    ]
1135 |   },
1136 |   {
1137 |    "cell_type": "code",
1138 |    "execution_count": null,
1139 |    "metadata": {},
1140 |    "outputs": [],
1141 |    "source": []
1142 |   }
1143 |  ],
1144 |  "metadata": {
1145 |   "kernelspec": {
1146 |    "display_name": "Python 3",
1147 |    "language": "python",
1148 |    "name": "python3"
1149 |   },
1150 |   "language_info": {
1151 |    "codemirror_mode": {
1152 |     "name": "ipython",
1153 |     "version": 3
1154 |    },
1155 |    "file_extension": ".py",
1156 |    "mimetype": "text/x-python",
1157 |    "name": "python",
1158 |    "nbconvert_exporter": "python",
1159 |    "pygments_lexer": "ipython3",
1160 |    "version": "3.5.6"
1161 |   }
1162 |  },
1163 |  "nbformat": 4,
1164 |  "nbformat_minor": 2
1165 | }
1166 | 


--------------------------------------------------------------------------------
/03_tensorflow_tutorial.ipynb:
--------------------------------------------------------------------------------
  1 | {
  2 |  "cells": [
  3 |   {
  4 |    "cell_type": "markdown",
  5 |    "metadata": {},
  6 |    "source": [
  7 |     "# 第 3 章：TensorFlow"
  8 |    ]
  9 |   },
 10 |   {
 11 |    "cell_type": "code",
 12 |    "execution_count": 1,
 13 |    "metadata": {},
 14 |    "outputs": [
 15 |     {
 16 |      "name": "stdout",
 17 |      "output_type": "stream",
 18 |      "text": [
 19 |       "1.10.0\n"
 20 |      ]
 21 |     }
 22 |    ],
 23 |    "source": [
 24 |     "# 导入 tensorflow\n",
 25 |     "import tensorflow as tf\n",
 26 |     "\n",
 27 |     "# 查看 tensorflow 版本\n",
 28 |     "print(tf.__version__)"
 29 |    ]
 30 |   },
 31 |   {
 32 |    "cell_type": "markdown",
 33 |    "metadata": {},
 34 |    "source": [
 35 |     "## 张量 Tensor"
 36 |    ]
 37 |   },
 38 |   {
 39 |    "cell_type": "markdown",
 40 |    "metadata": {},
 41 |    "source": [
 42 |     "### 创建 Tensor"
 43 |    ]
 44 |   },
 45 |   {
 46 |    "cell_type": "code",
 47 |    "execution_count": 2,
 48 |    "metadata": {},
 49 |    "outputs": [],
 50 |    "source": [
 51 |     "# 0 阶 Tensor\n",
 52 |     "c0 = tf.constant(2, name='c0')\n",
 53 |     "# 1 阶 Tensor\n",
 54 |     "c1 = tf.constant([1, 2, 3], name='c1')\n",
 55 |     "# 2 阶 Tensor\n",
 56 |     "c2 = tf.constant([[1, 2], [3, 4]], name='c2')"
 57 |    ]
 58 |   },
 59 |   {
 60 |    "cell_type": "code",
 61 |    "execution_count": 3,
 62 |    "metadata": {},
 63 |    "outputs": [
 64 |     {
 65 |      "name": "stdout",
 66 |      "output_type": "stream",
 67 |      "text": [
 68 |       "Tensor(\"c0:0\", shape=(), dtype=int32)\n",
 69 |       "Tensor(\"c1:0\", shape=(3,), dtype=int32)\n",
 70 |       "Tensor(\"c2:0\", shape=(2, 2), dtype=int32)\n"
 71 |      ]
 72 |     }
 73 |    ],
 74 |    "source": [
 75 |     "print(c0)\n",
 76 |     "print(c1)\n",
 77 |     "print(c2)"
 78 |    ]
 79 |   },
 80 |   {
 81 |    "cell_type": "markdown",
 82 |    "metadata": {},
 83 |    "source": [
 84 |     "### Tensor数学运算"
 85 |    ]
 86 |   },
 87 |   {
 88 |    "cell_type": "code",
 89 |    "execution_count": 4,
 90 |    "metadata": {},
 91 |    "outputs": [
 92 |     {
 93 |      "name": "stdout",
 94 |      "output_type": "stream",
 95 |      "text": [
 96 |       "Tensor(\"add:0\", shape=(2, 2), dtype=int32)\n",
 97 |       "Tensor(\"sub:0\", shape=(2, 2), dtype=int32)\n",
 98 |       "Tensor(\"multiply:0\", shape=(2, 2), dtype=int32)\n",
 99 |       "Tensor(\"matmul:0\", shape=(2, 2), dtype=int32)\n"
100 |      ]
101 |     }
102 |    ],
103 |    "source": [
104 |     "a = tf.constant([[1, 2], [3, 4]], name='a')\n",
105 |     "b = tf.constant([[5, 6], [7, 8]], name='b')\n",
106 |     "# 加法\n",
107 |     "out = tf.add(a, b, name='add')\n",
108 |     "print(out)\n",
109 |     "# 减法\n",
110 |     "out = tf.subtract(a, b, name='sub')\n",
111 |     "print(out)\n",
112 |     "# 对应元素相乘\n",
113 |     "out = tf.multiply(a, b, name='multiply')\n",
114 |     "print(out)\n",
115 |     "# 矩阵相乘\n",
116 |     "out = tf.matmul(a, b, name='matmul')\n",
117 |     "print(out)"
118 |    ]
119 |   },
120 |   {
121 |    "cell_type": "markdown",
122 |    "metadata": {},
123 |    "source": [
124 |     "## 数据流图（Data Flow Graph）"
125 |    ]
126 |   },
127 |   {
128 |    "cell_type": "markdown",
129 |    "metadata": {},
130 |    "source": [
131 |     "![](./datasets/ch03/3-3.png)"
132 |    ]
133 |   },
134 |   {
135 |    "cell_type": "code",
136 |    "execution_count": 5,
137 |    "metadata": {},
138 |    "outputs": [],
139 |    "source": [
140 |     "c = tf.constant(2, name='const')\n",
141 |     "\n",
142 |     "a = tf.Variable(3, name='a')\n",
143 |     "b = tf.Variable(4, name='b')\n",
144 |     "\n",
145 |     "f = a*a + a*b +c"
146 |    ]
147 |   },
148 |   {
149 |    "cell_type": "markdown",
150 |    "metadata": {},
151 |    "source": [
152 |     "## 会话（Session）"
153 |    ]
154 |   },
155 |   {
156 |    "cell_type": "markdown",
157 |    "metadata": {},
158 |    "source": [
159 |     "### 会话机制 1 "
160 |    ]
161 |   },
162 |   {
163 |    "cell_type": "code",
164 |    "execution_count": 6,
165 |    "metadata": {},
166 |    "outputs": [
167 |     {
168 |      "name": "stdout",
169 |      "output_type": "stream",
170 |      "text": [
171 |       "23\n"
172 |      ]
173 |     }
174 |    ],
175 |    "source": [
176 |     "sess = tf.Session()\n",
177 |     "sess.run(a.initializer)\n",
178 |     "sess.run(b.initializer)\n",
179 |     "result = sess.run(f)\n",
180 |     "print(result)\n",
181 |     "\n",
182 |     "sess.close()"
183 |    ]
184 |   },
185 |   {
186 |    "cell_type": "markdown",
187 |    "metadata": {},
188 |    "source": [
189 |     "### 会话机制 2"
190 |    ]
191 |   },
192 |   {
193 |    "cell_type": "code",
194 |    "execution_count": 7,
195 |    "metadata": {},
196 |    "outputs": [
197 |     {
198 |      "name": "stdout",
199 |      "output_type": "stream",
200 |      "text": [
201 |       "23\n"
202 |      ]
203 |     }
204 |    ],
205 |    "source": [
206 |     "with tf.Session() as sess:\n",
207 |     "    a.initializer.run()\n",
208 |     "    b.initializer.run()\n",
209 |     "    result = f.eval()\n",
210 |     "print(result)"
211 |    ]
212 |   },
213 |   {
214 |    "cell_type": "markdown",
215 |    "metadata": {},
216 |    "source": [
217 |     "### 会话机制 3"
218 |    ]
219 |   },
220 |   {
221 |    "cell_type": "code",
222 |    "execution_count": 8,
223 |    "metadata": {},
224 |    "outputs": [],
225 |    "source": [
226 |     "init = tf.global_variables_initializer()    # 定义全局初始化节点\n",
227 |     "\n",
228 |     "with tf.Session() as sess:\n",
229 |     "    init.run()    # 初始化所有变量\n",
230 |     "    result = f.eval()"
231 |    ]
232 |   },
233 |   {
234 |    "cell_type": "markdown",
235 |    "metadata": {},
236 |    "source": [
237 |     "### 占位符 tf.placeholder()"
238 |    ]
239 |   },
240 |   {
241 |    "cell_type": "code",
242 |    "execution_count": 9,
243 |    "metadata": {},
244 |    "outputs": [
245 |     {
246 |      "name": "stdout",
247 |      "output_type": "stream",
248 |      "text": [
249 |       "Situation 1:\n",
250 |       " [12.]\n",
251 |       "Situation 2:\n",
252 |       " [[2. 4.]\n",
253 |       " [6. 8.]]\n"
254 |      ]
255 |     }
256 |    ],
257 |    "source": [
258 |     "# 在 Tensorflow 中需要定义 placeholder 的 type，一般为 float32\n",
259 |     "a = tf.placeholder(tf.float32, name='a')\n",
260 |     "b = tf.placeholder(tf.float32, name='b')\n",
261 |     "f = tf.multiply(a, b)\n",
262 |     "\n",
263 |     "with tf.Session() as sess:\n",
264 |     "    print('Situation 1:\\n', sess.run(f, feed_dict={a: [3.], b: [4.]}))\n",
265 |     "    print('Situation 2:\\n',sess.run(f, feed_dict={a:[[1.,2.],[3.,4.]], b: [2.]}))"
266 |    ]
267 |   },
268 |   {
269 |    "cell_type": "markdown",
270 |    "metadata": {},
271 |    "source": [
272 |     "## TensorFlow 线性回归"
273 |    ]
274 |   },
275 |   {
276 |    "cell_type": "code",
277 |    "execution_count": 10,
278 |    "metadata": {},
279 |    "outputs": [],
280 |    "source": [
281 |     "import numpy as np\n",
282 |     "import matplotlib.pyplot as plt"
283 |    ]
284 |   },
285 |   {
286 |    "cell_type": "markdown",
287 |    "metadata": {},
288 |    "source": [
289 |     "### 创建数据集"
290 |    ]
291 |   },
292 |   {
293 |    "cell_type": "code",
294 |    "execution_count": 11,
295 |    "metadata": {},
296 |    "outputs": [],
297 |    "source": [
298 |     "x_train = np.linspace(-1, 1, 50)\n",
299 |     "y_train = 3*x_train + 10 + 0.5 * np.random.randn(x_train.shape[0])"
300 |    ]
301 |   },
302 |   {
303 |    "cell_type": "markdown",
304 |    "metadata": {},
305 |    "source": [
306 |     "显示数据分布"
307 |    ]
308 |   },
309 |   {
310 |    "cell_type": "code",
311 |    "execution_count": 12,
312 |    "metadata": {},
313 |    "outputs": [
314 |     {
315 |      "data": {
316 |       "image/png": "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\n",
317 |       "text/plain": [
318 |        "<Figure size 720x432 with 1 Axes>"
319 |       ]
320 |      },
321 |      "metadata": {
322 |       "needs_background": "light"
323 |      },
324 |      "output_type": "display_data"
325 |     }
326 |    ],
327 |    "source": [
328 |     "import matplotlib.pyplot as plt\n",
329 |     "plt.rcParams['font.sans-serif']=['SimHei']\n",
330 |     "plt.rcParams['axes.unicode_minus']=False\n",
331 |     "plt.rcParams['figure.figsize'] = (10.0, 6.0) # set default size of plots\n",
332 |     "\n",
333 |     "plt.scatter(x_train, y_train)\n",
334 |     "plt.show()"
335 |    ]
336 |   },
337 |   {
338 |    "cell_type": "markdown",
339 |    "metadata": {},
340 |    "source": [
341 |     "### 创建输入节点 x 和 y，用于输入数据"
342 |    ]
343 |   },
344 |   {
345 |    "cell_type": "code",
346 |    "execution_count": 13,
347 |    "metadata": {},
348 |    "outputs": [],
349 |    "source": [
350 |     "x = tf.placeholder(tf.float32, name='x')\n",
351 |     "y = tf.placeholder(tf.float32, name='y')"
352 |    ]
353 |   },
354 |   {
355 |    "cell_type": "markdown",
356 |    "metadata": {},
357 |    "source": [
358 |     "### 创建变量节点 w1 和 w0，并初始化变量"
359 |    ]
360 |   },
361 |   {
362 |    "cell_type": "code",
363 |    "execution_count": 14,
364 |    "metadata": {},
365 |    "outputs": [],
366 |    "source": [
367 |     "w1 = tf.Variable(tf.random_normal([1]), name='w1')\n",
368 |     "w0 = tf.Variable(tf.zeros([1]), name='w0')"
369 |    ]
370 |   },
371 |   {
372 |    "cell_type": "markdown",
373 |    "metadata": {},
374 |    "source": [
375 |     "### 创建线性模型"
376 |    ]
377 |   },
378 |   {
379 |    "cell_type": "code",
380 |    "execution_count": 15,
381 |    "metadata": {},
382 |    "outputs": [],
383 |    "source": [
384 |     "y_hat = w0 + w1 * x"
385 |    ]
386 |   },
387 |   {
388 |    "cell_type": "markdown",
389 |    "metadata": {},
390 |    "source": [
391 |     "### 创建损失模型"
392 |    ]
393 |   },
394 |   {
395 |    "cell_type": "code",
396 |    "execution_count": 16,
397 |    "metadata": {},
398 |    "outputs": [],
399 |    "source": [
400 |     "loss = tf.reduce_mean(tf.square(y_hat - y))"
401 |    ]
402 |   },
403 |   {
404 |    "cell_type": "markdown",
405 |    "metadata": {},
406 |    "source": [
407 |     "### 创建一个梯度下降优化器"
408 |    ]
409 |   },
410 |   {
411 |    "cell_type": "code",
412 |    "execution_count": 17,
413 |    "metadata": {},
414 |    "outputs": [],
415 |    "source": [
416 |     "optimizer = tf.train.GradientDescentOptimizer(0.01)    # 学习率设为 0.01\n",
417 |     "train = optimizer.minimize(loss)"
418 |    ]
419 |   },
420 |   {
421 |    "cell_type": "markdown",
422 |    "metadata": {},
423 |    "source": [
424 |     "### 创建会话 Session 用来计算模型"
425 |    ]
426 |   },
427 |   {
428 |    "cell_type": "code",
429 |    "execution_count": 18,
430 |    "metadata": {},
431 |    "outputs": [],
432 |    "source": [
433 |     "sess = tf.Session()"
434 |    ]
435 |   },
436 |   {
437 |    "cell_type": "markdown",
438 |    "metadata": {},
439 |    "source": [
440 |     "### 初始化变量"
441 |    ]
442 |   },
443 |   {
444 |    "cell_type": "code",
445 |    "execution_count": 19,
446 |    "metadata": {},
447 |    "outputs": [
448 |     {
449 |      "name": "stdout",
450 |      "output_type": "stream",
451 |      "text": [
452 |       "w1 = [-0.4311598] w0 = [0.]\n"
453 |      ]
454 |     }
455 |    ],
456 |    "source": [
457 |     "init = tf.global_variables_initializer()\n",
458 |     "sess.run(init)\n",
459 |     "print (\"w1 =\", sess.run(w1), \"w0 =\", sess.run(w0))  # 打印初始化的w1和w0"
460 |    ]
461 |   },
462 |   {
463 |    "cell_type": "markdown",
464 |    "metadata": {},
465 |    "source": [
466 |     "### 训练"
467 |    ]
468 |   },
469 |   {
470 |    "cell_type": "code",
471 |    "execution_count": 20,
472 |    "metadata": {},
473 |    "outputs": [
474 |     {
475 |      "name": "stdout",
476 |      "output_type": "stream",
477 |      "text": [
478 |       "Iteration[20/500], loss: 47.436157\n",
479 |       "Iteration[40/500], loss: 22.252466\n",
480 |       "Iteration[60/500], loss: 10.788024\n",
481 |       "Iteration[80/500], loss: 5.496624\n",
482 |       "Iteration[100/500], loss: 3.000713\n",
483 |       "Iteration[120/500], loss: 1.784191\n",
484 |       "Iteration[140/500], loss: 1.163201\n",
485 |       "Iteration[160/500], loss: 0.826786\n",
486 |       "Iteration[180/500], loss: 0.631707\n",
487 |       "Iteration[200/500], loss: 0.510594\n",
488 |       "Iteration[220/500], loss: 0.430753\n",
489 |       "Iteration[240/500], loss: 0.375594\n",
490 |       "Iteration[260/500], loss: 0.336194\n",
491 |       "Iteration[280/500], loss: 0.307420\n",
492 |       "Iteration[300/500], loss: 0.286107\n",
493 |       "Iteration[320/500], loss: 0.270183\n",
494 |       "Iteration[340/500], loss: 0.258223\n",
495 |       "Iteration[360/500], loss: 0.249212\n",
496 |       "Iteration[380/500], loss: 0.242410\n",
497 |       "Iteration[400/500], loss: 0.237269\n",
498 |       "Iteration[420/500], loss: 0.233382\n",
499 |       "Iteration[440/500], loss: 0.230442\n",
500 |       "Iteration[460/500], loss: 0.228216\n",
501 |       "Iteration[480/500], loss: 0.226533\n",
502 |       "Iteration[500/500], loss: 0.225258\n"
503 |      ]
504 |     }
505 |    ],
506 |    "source": [
507 |     "num_iter = 500\n",
508 |     "for i in range(num_iter):\n",
509 |     "    sess.run(train, {x: x_train, y: y_train})\n",
510 |     "    if (i+1) % 20 == 0:\n",
511 |     "        print('Iteration[{}/{}], loss: {:.6f}'.format(i+1,num_iter,sess.run(loss,{x:x_train,y:y_train})))"
512 |    ]
513 |   },
514 |   {
515 |    "cell_type": "markdown",
516 |    "metadata": {},
517 |    "source": [
518 |     "打印 w1"
519 |    ]
520 |   },
521 |   {
522 |    "cell_type": "code",
523 |    "execution_count": 21,
524 |    "metadata": {},
525 |    "outputs": [
526 |     {
527 |      "data": {
528 |       "text/plain": [
529 |        "array([2.938043], dtype=float32)"
530 |       ]
531 |      },
532 |      "execution_count": 21,
533 |      "metadata": {},
534 |      "output_type": "execute_result"
535 |     }
536 |    ],
537 |    "source": [
538 |     "sess.run(w1)"
539 |    ]
540 |   },
541 |   {
542 |    "cell_type": "markdown",
543 |    "metadata": {},
544 |    "source": [
545 |     "打印 w0"
546 |    ]
547 |   },
548 |   {
549 |    "cell_type": "code",
550 |    "execution_count": 22,
551 |    "metadata": {},
552 |    "outputs": [
553 |     {
554 |      "data": {
555 |       "text/plain": [
556 |        "array([9.940149], dtype=float32)"
557 |       ]
558 |      },
559 |      "execution_count": 22,
560 |      "metadata": {},
561 |      "output_type": "execute_result"
562 |     }
563 |    ],
564 |    "source": [
565 |     "sess.run(w0)"
566 |    ]
567 |   },
568 |   {
569 |    "cell_type": "markdown",
570 |    "metadata": {},
571 |    "source": [
572 |     "### 模型测试"
573 |    ]
574 |   },
575 |   {
576 |    "cell_type": "code",
577 |    "execution_count": 23,
578 |    "metadata": {},
579 |    "outputs": [
580 |     {
581 |      "data": {
582 |       "image/png": "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\n",
583 |       "text/plain": [
584 |        "<Figure size 720x432 with 1 Axes>"
585 |       ]
586 |      },
587 |      "metadata": {
588 |       "needs_background": "light"
589 |      },
590 |      "output_type": "display_data"
591 |     }
592 |    ],
593 |    "source": [
594 |     "y_hat = sess.run(y_hat, {x: x_train, y: y_train})    # 模型的预测输出\n",
595 |     "plt.scatter(x_train, y_train, label='原始数据')\n",
596 |     "plt.plot(x_train, y_hat, c='r', label='拟合直线')\n",
597 |     "# 显示图例\n",
598 |     "plt.legend() \n",
599 |     "plt.show()"
600 |    ]
601 |   },
602 |   {
603 |    "cell_type": "markdown",
604 |    "metadata": {},
605 |    "source": [
606 |     "### 关闭会话& 图复位"
607 |    ]
608 |   },
609 |   {
610 |    "cell_type": "code",
611 |    "execution_count": 24,
612 |    "metadata": {},
613 |    "outputs": [],
614 |    "source": [
615 |     "sess.close()    # 关闭会话 Session\n",
616 |     "tf.reset_default_graph()    # 图复位"
617 |    ]
618 |   },
619 |   {
620 |    "cell_type": "markdown",
621 |    "metadata": {},
622 |    "source": [
623 |     "## TensorBoard"
624 |    ]
625 |   },
626 |   {
627 |    "cell_type": "markdown",
628 |    "metadata": {},
629 |    "source": [
630 |     "### 创建日志存放目录"
631 |    ]
632 |   },
633 |   {
634 |    "cell_type": "code",
635 |    "execution_count": 25,
636 |    "metadata": {},
637 |    "outputs": [],
638 |    "source": [
639 |     "from datetime import  datetime\n",
640 |     "\n",
641 |     "now = datetime.utcnow().strftime('%Y%m%d%H%M%S')\n",
642 |     "root_logdir = 'tf_logs'\n",
643 |     "logdir = '{}/ch03/run-{}'.format(root_logdir, now)"
644 |    ]
645 |   },
646 |   {
647 |    "cell_type": "markdown",
648 |    "metadata": {},
649 |    "source": [
650 |     "### 构建计算图"
651 |    ]
652 |   },
653 |   {
654 |    "cell_type": "code",
655 |    "execution_count": 26,
656 |    "metadata": {},
657 |    "outputs": [],
658 |    "source": [
659 |     "x = tf.placeholder(tf.float32, name='x')\n",
660 |     "y = tf.placeholder(tf.float32, name='y')\n",
661 |     "w1 = tf.Variable(tf.random_normal([1]), name='w1')\n",
662 |     "w0 = tf.Variable(tf.zeros([1]), name='w0')\n",
663 |     "y_hat = w0 + w1 * x\n",
664 |     "loss = tf.reduce_mean(tf.square(y_hat - y))\n",
665 |     "optimizer = tf.train.GradientDescentOptimizer(0.01)    # 学习率设为 0.01\n",
666 |     "train = optimizer.minimize(loss)"
667 |    ]
668 |   },
669 |   {
670 |    "cell_type": "markdown",
671 |    "metadata": {},
672 |    "source": [
673 |     "### 设置 TensorBoard"
674 |    ]
675 |   },
676 |   {
677 |    "cell_type": "code",
678 |    "execution_count": 27,
679 |    "metadata": {},
680 |    "outputs": [],
681 |    "source": [
682 |     "# 给损失模型的输出添加scalar，用来观察loss的收敛曲线\n",
683 |     "loss_summary = tf.summary.scalar('loss', loss)\n",
684 |     "# 模型运行产生的所有数据保存到文件夹供 TensorBoard 使用\n",
685 |     "file_writer = tf.summary.FileWriter(logdir, tf.get_default_graph())"
686 |    ]
687 |   },
688 |   {
689 |    "cell_type": "markdown",
690 |    "metadata": {},
691 |    "source": [
692 |     "### 会话 Session"
693 |    ]
694 |   },
695 |   {
696 |    "cell_type": "code",
697 |    "execution_count": 28,
698 |    "metadata": {},
699 |    "outputs": [
700 |     {
701 |      "name": "stdout",
702 |      "output_type": "stream",
703 |      "text": [
704 |       "Iteration[20/500], loss: 49.139008\n",
705 |       "Iteration[40/500], loss: 23.541363\n",
706 |       "Iteration[60/500], loss: 11.763593\n",
707 |       "Iteration[80/500], loss: 6.235038\n",
708 |       "Iteration[100/500], loss: 3.559620\n",
709 |       "Iteration[120/500], loss: 2.207230\n",
710 |       "Iteration[140/500], loss: 1.483401\n",
711 |       "Iteration[160/500], loss: 1.069147\n",
712 |       "Iteration[180/500], loss: 0.815151\n",
713 |       "Iteration[200/500], loss: 0.649443\n",
714 |       "Iteration[220/500], loss: 0.535848\n",
715 |       "Iteration[240/500], loss: 0.455141\n",
716 |       "Iteration[260/500], loss: 0.396404\n",
717 |       "Iteration[280/500], loss: 0.352993\n",
718 |       "Iteration[300/500], loss: 0.320601\n",
719 |       "Iteration[320/500], loss: 0.296292\n",
720 |       "Iteration[340/500], loss: 0.277985\n",
721 |       "Iteration[360/500], loss: 0.264170\n",
722 |       "Iteration[380/500], loss: 0.253731\n",
723 |       "Iteration[400/500], loss: 0.245839\n",
724 |       "Iteration[420/500], loss: 0.239868\n",
725 |       "Iteration[440/500], loss: 0.235351\n",
726 |       "Iteration[460/500], loss: 0.231933\n",
727 |       "Iteration[480/500], loss: 0.229345\n",
728 |       "Iteration[500/500], loss: 0.227387\n"
729 |      ]
730 |     }
731 |    ],
732 |    "source": [
733 |     "init = tf.global_variables_initializer()\n",
734 |     "sess = tf.Session()\n",
735 |     "sess.run(init)\n",
736 |     "\n",
737 |     "num_iter = 500\n",
738 |     "for i in range(num_iter):\n",
739 |     "    # 训练时传入loss_summary\n",
740 |     "    summary, _ = sess.run([loss_summary, train], {x: x_train, y: y_train})\n",
741 |     "    # 收集每次训练产生的数据\n",
742 |     "    file_writer.add_summary(summary, i)\n",
743 |     "    if (i+1) % 20 == 0:\n",
744 |     "        print('Iteration[{}/{}], loss: {:.6f}'.format(i+1,num_iter,sess.run(loss,{x:x_train,y:y_train})))"
745 |    ]
746 |   },
747 |   {
748 |    "cell_type": "code",
749 |    "execution_count": null,
750 |    "metadata": {},
751 |    "outputs": [],
752 |    "source": []
753 |   }
754 |  ],
755 |  "metadata": {
756 |   "kernelspec": {
757 |    "display_name": "Python 3",
758 |    "language": "python",
759 |    "name": "python3"
760 |   },
761 |   "language_info": {
762 |    "codemirror_mode": {
763 |     "name": "ipython",
764 |     "version": 3
765 |    },
766 |    "file_extension": ".py",
767 |    "mimetype": "text/x-python",
768 |    "name": "python",
769 |    "nbconvert_exporter": "python",
770 |    "pygments_lexer": "ipython3",
771 |    "version": "3.5.6"
772 |   }
773 |  },
774 |  "nbformat": 4,
775 |  "nbformat_minor": 2
776 | }
777 | 


--------------------------------------------------------------------------------
/08_cnn_pytorch.ipynb:
--------------------------------------------------------------------------------
  1 | {
  2 |  "cells": [
  3 |   {
  4 |    "cell_type": "markdown",
  5 |    "metadata": {},
  6 |    "source": [
  7 |     "# 第 8 章：卷积神经网络 — PyTorch"
  8 |    ]
  9 |   },
 10 |   {
 11 |    "cell_type": "markdown",
 12 |    "metadata": {},
 13 |    "source": [
 14 |     "## 导入 MNIST 数据集"
 15 |    ]
 16 |   },
 17 |   {
 18 |    "cell_type": "code",
 19 |    "execution_count": 1,
 20 |    "metadata": {},
 21 |    "outputs": [],
 22 |    "source": [
 23 |     "import torch\n",
 24 |     "import torchvision\n",
 25 |     "import torchvision.transforms as transforms\n",
 26 |     "import torch.nn as nn\n",
 27 |     "import torch.nn.functional as F\n",
 28 |     "import torch.optim as optim\n",
 29 |     "import matplotlib.pyplot as plt\n",
 30 |     "import numpy as np\n",
 31 |     "\n",
 32 |     "transform = transforms.Compose(\n",
 33 |     "    [transforms.ToTensor()])\n",
 34 |     "\n",
 35 |     "# 训练集\n",
 36 |     "trainset = torchvision.datasets.MNIST(root='./datasets/ch08/pytorch',     # 选择数据的根目录\n",
 37 |     "                                      train=True,\n",
 38 |     "                                      download=True,    # 不从网络上download图片\n",
 39 |     "                                      transform=transform)\n",
 40 |     "trainloader = torch.utils.data.DataLoader(trainset, batch_size=4,\n",
 41 |     "                                         shuffle=True, num_workers=2)\n",
 42 |     "# 测试集\n",
 43 |     "testset = torchvision.datasets.MNIST(root='./datasets/ch08/pytorch',     # 选择数据的根目录\n",
 44 |     "                                     train=False,\n",
 45 |     "                                     download=True,    # 不从网络上download图片\n",
 46 |     "                                     transform=transform)\n",
 47 |     "testloader = torch.utils.data.DataLoader(testset, batch_size=4,\n",
 48 |     "                                        shuffle=False, num_workers=2)"
 49 |    ]
 50 |   },
 51 |   {
 52 |    "cell_type": "code",
 53 |    "execution_count": 2,
 54 |    "metadata": {},
 55 |    "outputs": [
 56 |     {
 57 |      "name": "stdout",
 58 |      "output_type": "stream",
 59 |      "text": [
 60 |       "Dataset MNIST\n",
 61 |       "    Number of datapoints: 60000\n",
 62 |       "    Split: train\n",
 63 |       "    Root Location: ./datasets/ch08/pytorch\n",
 64 |       "    Transforms (if any): Compose(\n",
 65 |       "                             ToTensor()\n",
 66 |       "                         )\n",
 67 |       "    Target Transforms (if any): None\n"
 68 |      ]
 69 |     }
 70 |    ],
 71 |    "source": [
 72 |     "print(trainset)"
 73 |    ]
 74 |   },
 75 |   {
 76 |    "cell_type": "code",
 77 |    "execution_count": 3,
 78 |    "metadata": {},
 79 |    "outputs": [
 80 |     {
 81 |      "name": "stdout",
 82 |      "output_type": "stream",
 83 |      "text": [
 84 |       "Dataset MNIST\n",
 85 |       "    Number of datapoints: 10000\n",
 86 |       "    Split: test\n",
 87 |       "    Root Location: ./datasets/ch08/pytorch\n",
 88 |       "    Transforms (if any): Compose(\n",
 89 |       "                             ToTensor()\n",
 90 |       "                         )\n",
 91 |       "    Target Transforms (if any): None\n"
 92 |      ]
 93 |     }
 94 |    ],
 95 |    "source": [
 96 |     "print(testset)"
 97 |    ]
 98 |   },
 99 |   {
100 |    "cell_type": "markdown",
101 |    "metadata": {},
102 |    "source": [
103 |     "接下来展示一些训练样本图像"
104 |    ]
105 |   },
106 |   {
107 |    "cell_type": "code",
108 |    "execution_count": 4,
109 |    "metadata": {},
110 |    "outputs": [
111 |     {
112 |      "data": {
113 |       "image/png": "iVBORw0KGgoAAAANSUhEUgAAAXYAAAB6CAYAAACr63iqAAAABHNCSVQICAgIfAhkiAAAAAlwSFlzAAALEgAACxIB0t1+/AAAADl0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uIDMuMC4yLCBodHRwOi8vbWF0cGxvdGxpYi5vcmcvOIA7rQAAElRJREFUeJzt3XmQVNXZx/HvIyiJWwBxYYugEgV3VIpXLU3hhiuJimI0olJFueDrVlHEJS5vVPSNr0tpFEFFYwQkLqh5X0FwX1AQNSogGBGQQeKOS6ng8/7R9545I91MT09Pz/Sd36eK6qdPd9977tz2ePrcc59j7o6IiGTHOs1dARERKS817CIiGaOGXUQkY9Swi4hkjBp2EZGMUcMuIpIxathFRDKmUQ27mQ00s/lmttDMRparUiIiUjor9QYlM2sDvAscACwFXgWOc/d3ylc9ERFpqLaN+Gw/YKG7/wvAzCYAg4CCDbuZ6TZXEZGG+9jdNy32zY0ZiukKLImeL03K6jCz4WY2y8xmNWJfIiKt2QcNeXNjeuyWp2yNHrm7jwHGgHrsIiKV0Jge+1Kge/S8G7CscdUREZHGakzD/irQy8x6mtl6wBBgSnmqJSIipSp5KMbdV5nZCOAJoA1wp7u/XbaaiYhISUqe7ljSzjTGLiJSitnuvnuxb9adpyIiGaOGXUQkY9Swi4hkjBp2EZGMUcMuIpIxathFRDJGDbuISMY0JleMSKNsueWWIb788ssBmDhxYigzq01H1K9fvxA/99xzACxatCiUvffee01VTWkhRowYEeKbb74ZgEL34UyePDnEn376KQBff/11KJs3b17ez02dOhWADz5oUM6tFkc9dhGRjNGdp9Lk1l9//RCPHz8+xAMGDAhxhw4dGrzdtHcFcNJJJ4V4+fLlDd6WtHwvv/xyiPfYY48m2ccZZ5wBwG233dYk228E3XkqItKaqWEXEckYXTyVslpnndq+whFHHAHAn/70p1DWu3fvEK9atSrE7777LlD3otdjjz0W4h49eoT4oosuAuDAAw8MZVdffXWITz755JLrL63bddddB8DSpUtDWfw9rBbqsYuIZIwadhGRjNFQjJRV166165k/+OCDa7wezx8+55xzQvzEE0+sdbvxjIhtttkGgCuuuKLkerZEgwcPDvHYsWNDvPHGG4f4hRdeAGD48OGh7J133qlA7VqHdAbXsGHDQtlTTz0V4ngufEumHruISMaoYRcRyZhWfYPSQQcdBMCoUaNC2U477RTiZ599NsQjR44EYO7cuRWqXXVq06ZNiI855pg1Xn/44YdD/O2335a0j3QoZtasWaHszTffDPE+++xT0nYrKf47pbOG/vCHP4SyOJ3CkiVLQty9e3cApkypXTf+/PPPD3FNTU2IV65cWcYaN7/DDz88xKNHjwbgV7/6VSj74YcfQvz++++HeKONNgKgS5cuJe23Z8+eIV68eHFJ2yiD8t6gZGZ3mtkKM3srKutoZtPMbEHy2PDbBkVEpEnU22M3s32Ar4B73H2HpOxa4FN3v8bMRgId3P2CenfWAnrsl112WYgPOeQQAB5//PFQNmPGjBD/7ne/C3Ha+7z00ktD2S233NJU1ZS1yNdjnzBhQohPPfXUitepoeK59uPGjQNg9erVoeziiy8OcXx7+9/+9jcADj744Lzbveuuu0IcXwDMqrPPPjvEX3zxRYjjv8O2224LwGGHHRbK0l/rAPvtt99a95HJHru7Pwt8+pPiQUCa9GM88JuiqyciIk2q1OmOm7t7DYC715jZZoXeaGbDgeGFXhcRkfJq8nns7j4GGAOVHYqJ51PfcMMNIe7UqVOI06GWhQsX5t1GmvcbarMSPvTQQ6HsjjvuCPH333/fyBpLsdLMfvH87tmzZzdXdYr2i1/8IsTxUEsqHlYoNMw3Z84coPBQzKabbtqYKlad+L/tQubPn1/nEaB9+/Yhrm8ophqVOt3xIzPrDJA8rihflUREpDFKbdinAEOTeCjwSHmqIyIijVXvUIyZ3Q/8GuhkZkuBPwLXAJPMbBiwGBhceAvN46abbgpxPL81ngv71VdfFb29dHmtzTffPJQNGjQoxA888EBJ9ZSGS+dtf/bZZ6Es/pndUsWzK+L4888/B+qmESgkvb09vvdCmlY8RHbuuec2Y02KV2/D7u7HFXgpewNTIiIZoJQCIiIZk7nsjrvssgtQu8gD1L3tuCHDL7E08f5HH30UyuJb1zUU07QuuKD2/rcdd9wRqHsjUpz+oaXq06dP3vIbb7wRgO+++67ebcQ32eRTzDakYdJFYKqJeuwiIhmTuR77BhtsANRdqT5OCJRP27a1f4Z4ubUjjzwyxL169QJgiy22CGUvvfRS4yoraxXP+z7vvPNCnC6/98wzz1S8To2x22675S2PUyPUp0OHNdMyxUsM/vnPf254xVqJNNc61P73XIx//OMfTVGdJqUeu4hIxqhhFxHJmMwNxSxbtgyoe5EzvvU/Xkasc+fOAAwYMCCUvfHGGyG++eabQ/z6668DdefHd+vWrVzVlkSaOxtg0qRJIY5vlU+HZRYsWFC5ipVZfJEzvTBfqvjCcbyEoOSkWVzj+ehZTCMQU49dRCRj1LCLiGRM5oZi0hkwJ5xwQiiLf4IdddRRIX711VeBunOD49k0+cQzMeLFEq699toSayyx+++/P8QHHHBAiJ9//vkQx0Nk1eStt8IiZLRr1y7E6UIQ8TBgPFNr6NChIf7tb3+7xnarcZ51U9t///1DPHnyZKDu37w+8fftyy+/LF/FKkQ9dhGRjFHDLiKSMZkbiknNmzcvxOVcAzNe+X3PPfcs23ZbsjijZToUEM8kKjQ7aOrUqUDtjCKAKVOmhHi99dYLcbpGZXyDWLyG5emnnx7iOFtnNZk+fXqI02yhUDvTKh5miW/OGjhwYAVqV/3izK333ntviBsyBJMumHPnnXeGsjT7ZjVRj11EJGMy22OXhtt6661DfNVVV4V45513DnGcUK0+/fv3X6PsyiuvDHG/fv1CnK4av2jRolDWt2/fEFdjr+mn4hXuTzvttBBfeumlABx77LGhLJ7nPmPGjBBvv/32QN3UFq3ZfffdF+L4gml8P0RDpN+zRx6pXTsoTWEB8OOPP5a03UpTj11EJGPUsIuIZIyGYlqp+KfqK6+8AkCXLl1CWTxf//bbb1/j8/GQSfzeNLsmwDXXXAPAccfVLsJ1ySWX5K1Pur04x30Whl8KifP3P/744wB07NgxlK1evTrENTU1Ib777rsBOPHEE5u4hi3PZpttFuL0+AcPrl2Vs02bNmXbxyeffBLK4mHJ+IJ+aty4cSGOl2tsTvX22M2su5k9ZWZzzextMzsrKe9oZtPMbEHyuGY+URERqbhihmJWAee5e2+gP3CGmfUBRgLT3b0XMD15LiIizayYxaxrgJokXmlmc4GuwCDg18nbxgNPAxfk2URmffjhh81dhZKNHj06xOkt7ffcc08oO+mkk4re1rrrrhvieAZNPMumPunfstSlC6vZN998U+dR8hs7dmyIDz300Irtd9SoUWt9PZ7hFN9jEc/YiWeDVUKDxtjNrAewKzAT2Dxp9HH3GjPbrMBnhgPDG1dNEREpVtENu5ltCPwdONvdvzSzoj7n7mOAMck2vJRKiohI8Ypq2M1sXXKN+n3u/mBS/JGZdU56652BFU1VyZbqtddea+4qlCzOnJiaNm1a0Z/v3r17iOMZGvl+cqZZNAE23HDDEPfu3TvEPXv2BGCvvfYKZelsEamVLqoR/8333Xff5qpORcWzYlqSHj165C2PZ4Ols5mWLFlSgRoVNyvGgHHAXHe/PnppCpDmEx0KPPLTz4qISOUV02PfC/g98E8zS7M5jQKuASaZ2TBgMTC4wOelSsRJlOKV2eNkZ2nu+rjHv9VWW4V4zpw5IX700UeBujnW0yRLAH/9619DnN42/+STT5Z+AK1AmtM9/jtut912IR40aFCI49vipfLSSQkA7du3ByrXYy9mVszzQKEB9WwvHCgiUoWUUkBEJGOUUqCVilezT+ebH3300aFs7733DnGcaiD18ccfhzhNHQBw/fXX531PPvGFUilOmv7h6aefDmVxDvsjjzwyxFkbikmXvQTYY489mnRfy5YtC3GnTp1CHK8h0JKpxy4ikjFq2EVEMkZDMa3UpEmTQrxq1aqiPzd37lyg7tJh9Q25SPnNmjUrxPFQTJ8+fUKcLgkXL9pRzUaMGBHidPbUrbfeGsrats3fnJ111lkAfPvtt0XvK17OsVevXiGOs5fmc8opp4Q430IzlaIeu4hIxqhhFxHJGHOvXPqWLOSKiX/2Hn/88SEeOnRovreLNIlNNtkkxDNnzgxxfLNYesOZUjNkwmx3373YN6vHLiKSMbp42gi//OUvm7sK0krFS7eNGTMmxPE9BUOGDAFg+fLloWz27NkVqJ00N/XYRUQyRg27iEjGaCimgVauXBnivn37hjjOM94al3eT5jNx4sQQx1k304v7L730UijTUEzroB67iEjGqGEXEckYzWNvhPhn7ZlnnhniF198sTmqIyLZpXnsIiKtmRp2EZGMqXcoxsx+BjwLtCM3i2ayu//RzHoCE4COwGvA7939+8Jbyt5QjIhIhZR9KOY7YIC77wzsAgw0s/7AaOB/3L0X8BkwrJTaiohIedXbsHtOOjF73eSfAwOAyUn5eOA3TVJDERFpkKLG2M2sjZm9DqwApgHvAZ+7e7pCw1Kga9NUUUREGqKoht3dV7v7LkA3oB/QO9/b8n3WzIab2Swzm5XvdRERKa8GzYpx98+Bp4H+QHszS1MSdAOWFfjMGHffvSED/yIiUrp6G3Yz29TM2ifxz4H9gbnAU8DRyduGAo80VSVFRKR4xSQB6wyMN7M25P5HMMndHzOzd4AJZvZfwBxgXBPWU0REilTplAL/Br4GsrqsfSd0bNVIx1adWtOxbenumxb74Yo27ABmNiur4+06tuqkY6tOOrbClFJARCRj1LCLiGRMczTsY+p/S9XSsVUnHVt10rEVUPExdhERaVoaihERyRg17CIiGVPRht3MBprZfDNbaGYjK7nvcjOz7mb2lJnNNbO3zeyspLyjmU0zswXJY4fmrmspksRvc8zsseR5TzObmRzXRDNbr7nrWAoza29mk81sXnLu/iND5+yc5Lv4lpndb2Y/q9bzZmZ3mtkKM3srKst7niznpqRdedPM+jZfzetX4NiuS76Tb5rZQ+nd/slrFybHNt/MDipmHxVr2JM7V28BDgb6AMeZWZ9K7b8JrALOc/fe5HLnnJEcz0hgepKnfnryvBqdRS51RCor+fdvBP7P3bcDdiZ3jFV/zsysK/CfwO7uvgPQBhhC9Z63u4GBPykrdJ4OBnol/4YDf6lQHUt1N2se2zRgB3ffCXgXuBAgaVOGANsnn7k1aUvXqpI99n7AQnf/V7LS0gRgUAX3X1buXuPuryXxSnINRFdyxzQ+eVtV5qk3s27AocDY5LmRgfz7ZrYxsA9J+gt3/z5JbFf15yzRFvh5kpxvfaCGKj1v7v4s8OlPigudp0HAPcnaES+TS1DYuTI1bbh8x+buU6M06C+TS6wIuWOb4O7fufv7wEJybelaVbJh7wosiZ5nJoe7mfUAdgVmApu7ew3kGn9gs+arWcluAM4Hfkyeb0I28u9vBfwbuCsZZhprZhuQgXPm7h8C/w0sJtegfwHMJhvnLVXoPGWtbTkF+N8kLunYKtmwW56yqp9raWYbAn8Hznb3L5u7Po1lZocBK9x9dlyc563VeO7aAn2Bv7j7ruTyFlXdsEs+yXjzIKAn0AXYgNwQxU9V43mrT1a+n5jZReSGee9Li/K8rd5jq2TDvhToHj0vmMO9WpjZuuQa9fvc/cGk+KP0Z2DyuKK56leivYAjzGwRueGyAeR68EXl32/hlgJL3X1m8nwyuYa+2s8Z5NJpv+/u/3b3H4AHgT3JxnlLFTpPmWhbzGwocBhwvNfeYFTSsVWyYX8V6JVcpV+P3AWBKRXcf1kl487jgLnufn300hRy+emhCvPUu/uF7t7N3XuQO0cz3P14MpB/392XA0vMbNukaD/gHar8nCUWA/3NbP3ku5keW9Wft0ih8zQFODGZHdMf+CIdsqkWZjYQuAA4wt2/iV6aAgwxs3Zm1pPcBeJX6t2gu1fsH3AIuSu+7wEXVXLfTXAse5P7SfQm8Hry7xBy49HTgQXJY8fmrmsjjvHXwGNJvFXyhVoIPAC0a+76lXhMuwCzkvP2MNAhK+cMuByYB7wF3Au0q9bzBtxP7lrBD+R6rcMKnSdywxW3JO3KP8nNDGr2Y2jgsS0kN5aetiW3Re+/KDm2+cDBxexDKQVERDJGd56KiGSMGnYRkYxRwy4ikjFq2EVEMkYNu4hIxqhhFxHJGDXsIiIZ8/8aIlzCxxtQlAAAAABJRU5ErkJggg==\n",
114 |       "text/plain": [
115 |        "<Figure size 432x288 with 1 Axes>"
116 |       ]
117 |      },
118 |      "metadata": {
119 |       "needs_background": "light"
120 |      },
121 |      "output_type": "display_data"
122 |     },
123 |     {
124 |      "name": "stdout",
125 |      "output_type": "stream",
126 |      "text": [
127 |       "          9           2           9           2\n"
128 |      ]
129 |     }
130 |    ],
131 |    "source": [
132 |     "def imshow(img):\n",
133 |     "    npimg = img.numpy()\n",
134 |     "    plt.imshow(np.transpose(npimg, (1, 2, 0)))\n",
135 |     "\n",
136 |     "# 选择一个 batch 的图片\n",
137 |     "dataiter = iter(trainloader)\n",
138 |     "images, labels = dataiter.next()\n",
139 |     "\n",
140 |     "# 显示图片\n",
141 |     "imshow(torchvision.utils.make_grid(images))\n",
142 |     "plt.show()\n",
143 |     "# 打印 labels\n",
144 |     "print(' '.join('%11s' % labels[j].numpy() for j in range(4)))"
145 |    ]
146 |   },
147 |   {
148 |    "cell_type": "markdown",
149 |    "metadata": {},
150 |    "source": [
151 |     "## 定义卷积神经网络"
152 |    ]
153 |   },
154 |   {
155 |    "cell_type": "code",
156 |    "execution_count": 5,
157 |    "metadata": {},
158 |    "outputs": [],
159 |    "source": [
160 |     "class Net(nn.Module):\n",
161 |     "    def __init__(self):\n",
162 |     "        super(Net, self).__init__()\n",
163 |     "        self.conv1 = nn.Conv2d(1, 6, 5)        # 1个输入图片通道，6个输出通道，5x5 卷积核\n",
164 |     "        self.pool1 = nn.MaxPool2d(2, 2)        # max pooling，2x2\n",
165 |     "        self.conv2 = nn.Conv2d(6, 16, 5)       # 6个输入图片通道，16个输出通道，5x5 卷积核\n",
166 |     "        self.pool2 = nn.MaxPool2d(2,2)         # max pooling，2x2\n",
167 |     "        self.fc1 = nn.Linear(16 * 4 * 4, 120)  # 拉伸成一维向量，全连接层\n",
168 |     "        self.fc2 = nn.Linear(120, 84)          # 全连接层 \n",
169 |     "        self.fc3 = nn.Linear(84, 10)           # 全连接层，输出层 softmax，10个数字\n",
170 |     "    \n",
171 |     "    def forward(self, x):\n",
172 |     "        x = F.relu(self.conv1(x))\n",
173 |     "        x = self.pool1(x)\n",
174 |     "        x = F.relu(self.conv2(x))\n",
175 |     "        x = self.pool2(x)\n",
176 |     "        x = x.view(-1, 16 * 4 * 4)    # 拉伸成一维向量\n",
177 |     "        x = F.relu(self.fc1(x))\n",
178 |     "        x = F.relu(self.fc2(x))\n",
179 |     "        x = self.fc3(x)\n",
180 |     "        return x"
181 |    ]
182 |   },
183 |   {
184 |    "cell_type": "code",
185 |    "execution_count": 6,
186 |    "metadata": {},
187 |    "outputs": [
188 |     {
189 |      "name": "stdout",
190 |      "output_type": "stream",
191 |      "text": [
192 |       "Net(\n",
193 |       "  (conv1): Conv2d(1, 6, kernel_size=(5, 5), stride=(1, 1))\n",
194 |       "  (pool1): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)\n",
195 |       "  (conv2): Conv2d(6, 16, kernel_size=(5, 5), stride=(1, 1))\n",
196 |       "  (pool2): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)\n",
197 |       "  (fc1): Linear(in_features=256, out_features=120, bias=True)\n",
198 |       "  (fc2): Linear(in_features=120, out_features=84, bias=True)\n",
199 |       "  (fc3): Linear(in_features=84, out_features=10, bias=True)\n",
200 |       ")\n"
201 |      ]
202 |     }
203 |    ],
204 |    "source": [
205 |     "net = Net()\n",
206 |     "print(net) "
207 |    ]
208 |   },
209 |   {
210 |    "cell_type": "markdown",
211 |    "metadata": {},
212 |    "source": [
213 |     "## 定义损失函数与优化算法"
214 |    ]
215 |   },
216 |   {
217 |    "cell_type": "code",
218 |    "execution_count": 7,
219 |    "metadata": {},
220 |    "outputs": [],
221 |    "source": [
222 |     "criterion = nn.CrossEntropyLoss()    # 交叉熵损失\n",
223 |     "optimizer = optim.Adam(net.parameters(), lr=0.0001)    # Adam 优化算法"
224 |    ]
225 |   },
226 |   {
227 |    "cell_type": "markdown",
228 |    "metadata": {},
229 |    "source": [
230 |     "## 训练网络"
231 |    ]
232 |   },
233 |   {
234 |    "cell_type": "code",
235 |    "execution_count": 8,
236 |    "metadata": {},
237 |    "outputs": [
238 |     {
239 |      "name": "stdout",
240 |      "output_type": "stream",
241 |      "text": [
242 |       "[epoch: 1, mini-batch:  2000] loss: 1.085\n",
243 |       "[epoch: 1, mini-batch:  4000] loss: 0.397\n",
244 |       "[epoch: 1, mini-batch:  6000] loss: 0.305\n",
245 |       "[epoch: 1, mini-batch:  8000] loss: 0.231\n",
246 |       "[epoch: 1, mini-batch: 10000] loss: 0.191\n",
247 |       "[epoch: 1, mini-batch: 12000] loss: 0.174\n",
248 |       "[epoch: 1, mini-batch: 14000] loss: 0.152\n",
249 |       "[epoch: 2, mini-batch:  2000] loss: 0.133\n",
250 |       "[epoch: 2, mini-batch:  4000] loss: 0.119\n",
251 |       "[epoch: 2, mini-batch:  6000] loss: 0.125\n",
252 |       "[epoch: 2, mini-batch:  8000] loss: 0.102\n",
253 |       "[epoch: 2, mini-batch: 10000] loss: 0.110\n",
254 |       "[epoch: 2, mini-batch: 12000] loss: 0.094\n",
255 |       "[epoch: 2, mini-batch: 14000] loss: 0.097\n",
256 |       "[epoch: 3, mini-batch:  2000] loss: 0.084\n",
257 |       "[epoch: 3, mini-batch:  4000] loss: 0.087\n",
258 |       "[epoch: 3, mini-batch:  6000] loss: 0.078\n",
259 |       "[epoch: 3, mini-batch:  8000] loss: 0.070\n",
260 |       "[epoch: 3, mini-batch: 10000] loss: 0.086\n",
261 |       "[epoch: 3, mini-batch: 12000] loss: 0.073\n",
262 |       "[epoch: 3, mini-batch: 14000] loss: 0.079\n",
263 |       "[epoch: 4, mini-batch:  2000] loss: 0.059\n",
264 |       "[epoch: 4, mini-batch:  4000] loss: 0.070\n",
265 |       "[epoch: 4, mini-batch:  6000] loss: 0.070\n",
266 |       "[epoch: 4, mini-batch:  8000] loss: 0.055\n",
267 |       "[epoch: 4, mini-batch: 10000] loss: 0.057\n",
268 |       "[epoch: 4, mini-batch: 12000] loss: 0.066\n",
269 |       "[epoch: 4, mini-batch: 14000] loss: 0.061\n",
270 |       "[epoch: 5, mini-batch:  2000] loss: 0.052\n",
271 |       "[epoch: 5, mini-batch:  4000] loss: 0.053\n",
272 |       "[epoch: 5, mini-batch:  6000] loss: 0.052\n",
273 |       "[epoch: 5, mini-batch:  8000] loss: 0.059\n",
274 |       "[epoch: 5, mini-batch: 10000] loss: 0.047\n",
275 |       "[epoch: 5, mini-batch: 12000] loss: 0.049\n",
276 |       "[epoch: 5, mini-batch: 14000] loss: 0.047\n"
277 |      ]
278 |     }
279 |    ],
280 |    "source": [
281 |     "num_epoches = 5    # 设置 epoch 数目\n",
282 |     "cost = []     # 损失函数累加\n",
283 |     "\n",
284 |     "for epoch in range(num_epoches):    \n",
285 |     "    \n",
286 |     "    running_loss = 0.0\n",
287 |     "    for i, data in enumerate(trainloader):\n",
288 |     "        # 输入样本和标签\n",
289 |     "        inputs, labels = data\n",
290 |     "        \n",
291 |     "        # 每次训练梯度清零\n",
292 |     "        optimizer.zero_grad()\n",
293 |     "        \n",
294 |     "        # 正向传播、反向传播和优化过程\n",
295 |     "        outputs = net(inputs)\n",
296 |     "        loss = criterion(outputs, labels)\n",
297 |     "        loss.backward()\n",
298 |     "        optimizer.step()\n",
299 |     "        \n",
300 |     "        # 打印训练情况\n",
301 |     "        running_loss += loss.item()\n",
302 |     "        if (i+1) % 2000 == 0:    # 每隔2000 mini-batches，打印一次\n",
303 |     "            print('[epoch: %d, mini-batch: %5d] loss: %.3f' % \n",
304 |     "                 (epoch + 1, i + 1, running_loss / 2000))\n",
305 |     "            cost.append(running_loss / 2000)\n",
306 |     "            running_loss = 0.0"
307 |    ]
308 |   },
309 |   {
310 |    "cell_type": "code",
311 |    "execution_count": 9,
312 |    "metadata": {},
313 |    "outputs": [
314 |     {
315 |      "data": {
316 |       "image/png": "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\n",
317 |       "text/plain": [
318 |        "<Figure size 432x288 with 1 Axes>"
319 |       ]
320 |      },
321 |      "metadata": {
322 |       "needs_background": "light"
323 |      },
324 |      "output_type": "display_data"
325 |     }
326 |    ],
327 |    "source": [
328 |     "plt.plot(cost)\n",
329 |     "plt.xlabel('mini-batches(per 2000)')\n",
330 |     "plt.ylabel('running_loss')\n",
331 |     "plt.show()"
332 |    ]
333 |   },
334 |   {
335 |    "cell_type": "markdown",
336 |    "metadata": {},
337 |    "source": [
338 |     "## 测试数据"
339 |    ]
340 |   },
341 |   {
342 |    "cell_type": "code",
343 |    "execution_count": 10,
344 |    "metadata": {},
345 |    "outputs": [
346 |     {
347 |      "name": "stdout",
348 |      "output_type": "stream",
349 |      "text": [
350 |       "Accuracy on the 60000 train images: 98.883 %\n"
351 |      ]
352 |     }
353 |    ],
354 |    "source": [
355 |     "correct = 0\n",
356 |     "total = 0\n",
357 |     "with torch.no_grad():\n",
358 |     "    for data in trainloader:\n",
359 |     "        images, labels = data\n",
360 |     "        outputs = net(images)\n",
361 |     "        _, predicted = torch.max(outputs.data, 1)\n",
362 |     "        total += labels.size(0)\n",
363 |     "        correct += (predicted == labels).sum().item()\n",
364 |     "\n",
365 |     "print('Accuracy on the 60000 train images: %.3f %%' % \n",
366 |     "     (100 * correct / total))"
367 |    ]
368 |   },
369 |   {
370 |    "cell_type": "code",
371 |    "execution_count": 11,
372 |    "metadata": {},
373 |    "outputs": [
374 |     {
375 |      "name": "stdout",
376 |      "output_type": "stream",
377 |      "text": [
378 |       "Accuracy on the 10000 test images: 98.580 %\n"
379 |      ]
380 |     }
381 |    ],
382 |    "source": [
383 |     "correct = 0\n",
384 |     "total = 0\n",
385 |     "with torch.no_grad():\n",
386 |     "    for data in testloader:\n",
387 |     "        images, labels = data\n",
388 |     "        outputs = net(images)\n",
389 |     "        _, predicted = torch.max(outputs.data, 1)\n",
390 |     "        total += labels.size(0)\n",
391 |     "        correct += (predicted == labels).sum().item()\n",
392 |     "\n",
393 |     "print('Accuracy on the 10000 test images: %.3f %%' % \n",
394 |     "     (100 * correct / total))"
395 |    ]
396 |   },
397 |   {
398 |    "cell_type": "code",
399 |    "execution_count": null,
400 |    "metadata": {},
401 |    "outputs": [],
402 |    "source": []
403 |   }
404 |  ],
405 |  "metadata": {
406 |   "kernelspec": {
407 |    "display_name": "Python 3",
408 |    "language": "python",
409 |    "name": "python3"
410 |   },
411 |   "language_info": {
412 |    "codemirror_mode": {
413 |     "name": "ipython",
414 |     "version": 3
415 |    },
416 |    "file_extension": ".py",
417 |    "mimetype": "text/x-python",
418 |    "name": "python",
419 |    "nbconvert_exporter": "python",
420 |    "pygments_lexer": "ipython3",
421 |    "version": "3.5.6"
422 |   }
423 |  },
424 |  "nbformat": 4,
425 |  "nbformat_minor": 2
426 | }
427 | 


--------------------------------------------------------------------------------
/08_cnn_tensorflow.ipynb:
--------------------------------------------------------------------------------
  1 | {
  2 |  "cells": [
  3 |   {
  4 |    "cell_type": "markdown",
  5 |    "metadata": {},
  6 |    "source": [
  7 |     "# 第 8 章：卷积神经网络 — TensorFlow"
  8 |    ]
  9 |   },
 10 |   {
 11 |    "cell_type": "code",
 12 |    "execution_count": 1,
 13 |    "metadata": {},
 14 |    "outputs": [],
 15 |    "source": [
 16 |     "from tensorflow.examples.tutorials.mnist import input_data\n",
 17 |     "import tensorflow as tf\n",
 18 |     "import matplotlib.pyplot as plt"
 19 |    ]
 20 |   },
 21 |   {
 22 |    "cell_type": "markdown",
 23 |    "metadata": {},
 24 |    "source": [
 25 |     "## 载入数据集"
 26 |    ]
 27 |   },
 28 |   {
 29 |    "cell_type": "code",
 30 |    "execution_count": 2,
 31 |    "metadata": {},
 32 |    "outputs": [
 33 |     {
 34 |      "name": "stdout",
 35 |      "output_type": "stream",
 36 |      "text": [
 37 |       "WARNING:tensorflow:From <ipython-input-2-3f5d8b810895>:1: read_data_sets (from tensorflow.contrib.learn.python.learn.datasets.mnist) is deprecated and will be removed in a future version.\n",
 38 |       "Instructions for updating:\n",
 39 |       "Please use alternatives such as official/mnist/dataset.py from tensorflow/models.\n",
 40 |       "WARNING:tensorflow:From C:\\ProgramData\\Anaconda3\\envs\\tensorflow\\lib\\site-packages\\tensorflow\\contrib\\learn\\python\\learn\\datasets\\mnist.py:260: maybe_download (from tensorflow.contrib.learn.python.learn.datasets.base) is deprecated and will be removed in a future version.\n",
 41 |       "Instructions for updating:\n",
 42 |       "Please write your own downloading logic.\n",
 43 |       "WARNING:tensorflow:From C:\\ProgramData\\Anaconda3\\envs\\tensorflow\\lib\\site-packages\\tensorflow\\contrib\\learn\\python\\learn\\datasets\\mnist.py:262: extract_images (from tensorflow.contrib.learn.python.learn.datasets.mnist) is deprecated and will be removed in a future version.\n",
 44 |       "Instructions for updating:\n",
 45 |       "Please use tf.data to implement this functionality.\n",
 46 |       "Extracting ./datasets/ch08/tensorflow/MNIST\\train-images-idx3-ubyte.gz\n",
 47 |       "WARNING:tensorflow:From C:\\ProgramData\\Anaconda3\\envs\\tensorflow\\lib\\site-packages\\tensorflow\\contrib\\learn\\python\\learn\\datasets\\mnist.py:267: extract_labels (from tensorflow.contrib.learn.python.learn.datasets.mnist) is deprecated and will be removed in a future version.\n",
 48 |       "Instructions for updating:\n",
 49 |       "Please use tf.data to implement this functionality.\n",
 50 |       "Extracting ./datasets/ch08/tensorflow/MNIST\\train-labels-idx1-ubyte.gz\n",
 51 |       "WARNING:tensorflow:From C:\\ProgramData\\Anaconda3\\envs\\tensorflow\\lib\\site-packages\\tensorflow\\contrib\\learn\\python\\learn\\datasets\\mnist.py:110: dense_to_one_hot (from tensorflow.contrib.learn.python.learn.datasets.mnist) is deprecated and will be removed in a future version.\n",
 52 |       "Instructions for updating:\n",
 53 |       "Please use tf.one_hot on tensors.\n",
 54 |       "Extracting ./datasets/ch08/tensorflow/MNIST\\t10k-images-idx3-ubyte.gz\n",
 55 |       "Extracting ./datasets/ch08/tensorflow/MNIST\\t10k-labels-idx1-ubyte.gz\n",
 56 |       "WARNING:tensorflow:From C:\\ProgramData\\Anaconda3\\envs\\tensorflow\\lib\\site-packages\\tensorflow\\contrib\\learn\\python\\learn\\datasets\\mnist.py:290: DataSet.__init__ (from tensorflow.contrib.learn.python.learn.datasets.mnist) is deprecated and will be removed in a future version.\n",
 57 |       "Instructions for updating:\n",
 58 |       "Please use alternatives such as official/mnist/dataset.py from tensorflow/models.\n"
 59 |      ]
 60 |     }
 61 |    ],
 62 |    "source": [
 63 |     "mnist = input_data.read_data_sets('./datasets/ch08/tensorflow/MNIST',one_hot=True)"
 64 |    ]
 65 |   },
 66 |   {
 67 |    "cell_type": "code",
 68 |    "execution_count": 3,
 69 |    "metadata": {},
 70 |    "outputs": [
 71 |     {
 72 |      "name": "stdout",
 73 |      "output_type": "stream",
 74 |      "text": [
 75 |       "(55000, 784)\n",
 76 |       "(55000, 10)\n"
 77 |      ]
 78 |     }
 79 |    ],
 80 |    "source": [
 81 |     "print(mnist.train.images.shape)\n",
 82 |     "print(mnist.train.labels.shape)"
 83 |    ]
 84 |   },
 85 |   {
 86 |    "cell_type": "code",
 87 |    "execution_count": 4,
 88 |    "metadata": {},
 89 |    "outputs": [
 90 |     {
 91 |      "name": "stdout",
 92 |      "output_type": "stream",
 93 |      "text": [
 94 |       "(5000, 784)\n",
 95 |       "(5000, 10)\n"
 96 |      ]
 97 |     }
 98 |    ],
 99 |    "source": [
100 |     "print(mnist.validation.images.shape)\n",
101 |     "print(mnist.validation.labels.shape)"
102 |    ]
103 |   },
104 |   {
105 |    "cell_type": "code",
106 |    "execution_count": 5,
107 |    "metadata": {},
108 |    "outputs": [
109 |     {
110 |      "name": "stdout",
111 |      "output_type": "stream",
112 |      "text": [
113 |       "(10000, 784)\n",
114 |       "(10000, 10)\n"
115 |      ]
116 |     }
117 |    ],
118 |    "source": [
119 |     "print(mnist.test.images.shape)\n",
120 |     "print(mnist.test.labels.shape)"
121 |    ]
122 |   },
123 |   {
124 |    "cell_type": "markdown",
125 |    "metadata": {},
126 |    "source": [
127 |     "## 定义创建模型需要的函数"
128 |    ]
129 |   },
130 |   {
131 |    "cell_type": "code",
132 |    "execution_count": 6,
133 |    "metadata": {},
134 |    "outputs": [],
135 |    "source": [
136 |     "# input 代表输入，filter 代表卷积核\n",
137 |     "# 卷积层\n",
138 |     "def conv2d(x, filter):\n",
139 |     "    return tf.nn.conv2d(x, \n",
140 |     "                        filter, \n",
141 |     "                        strides=[1,1,1,1], \n",
142 |     "                        padding='SAME')\n",
143 |     "\n",
144 |     "# 池化层\n",
145 |     "def max_pool(x):\n",
146 |     "    return tf.nn.max_pool(x, \n",
147 |     "                          ksize=[1,2,2,1], \n",
148 |     "                          strides=[1,2,2,1], \n",
149 |     "                          padding='SAME')\n",
150 |     "\n",
151 |     "# 初始化卷积核或者是权重数组的值\n",
152 |     "def weight_variable(shape):\n",
153 |     "    initial = tf.truncated_normal(shape, stddev=0.1)\n",
154 |     "    return tf.Variable(initial)\n",
155 |     "\n",
156 |     "# 初始化bias的值\n",
157 |     "def bias_variable(shape):\n",
158 |     "    initial = tf.constant(0.1, shape=shape)\n",
159 |     "    return tf.Variable(initial)"
160 |    ]
161 |   },
162 |   {
163 |    "cell_type": "markdown",
164 |    "metadata": {},
165 |    "source": [
166 |     "## 定义占位符 placeholder"
167 |    ]
168 |   },
169 |   {
170 |    "cell_type": "code",
171 |    "execution_count": 7,
172 |    "metadata": {},
173 |    "outputs": [],
174 |    "source": [
175 |     "# None　代表图片数量未知\n",
176 |     "x = tf.placeholder(tf.float32, [None,784])/255\n",
177 |     "# 将input 重新调整结构，适用于CNN的特征提取\n",
178 |     "x_image = tf.reshape(x, [-1,28,28,1])\n",
179 |     "\n",
180 |     "# y是最终预测的结果\n",
181 |     "y = tf.placeholder(tf.float32, [None,10])"
182 |    ]
183 |   },
184 |   {
185 |    "cell_type": "markdown",
186 |    "metadata": {},
187 |    "source": [
188 |     "## 定义 CNN 模型"
189 |    ]
190 |   },
191 |   {
192 |    "cell_type": "markdown",
193 |    "metadata": {},
194 |    "source": [
195 |     "### CONV1"
196 |    ]
197 |   },
198 |   {
199 |    "cell_type": "code",
200 |    "execution_count": 8,
201 |    "metadata": {},
202 |    "outputs": [],
203 |    "source": [
204 |     "W_conv1 = weight_variable([5, 5, 1, 32])  # 卷积核尺寸：5x5, 输入通道：1, 输出通道：32\n",
205 |     "b_conv1 = bias_variable([32])\n",
206 |     "h_conv1 = tf.nn.relu(conv2d(x_image, W_conv1) + b_conv1)  # 输出尺寸：28x28x32\n",
207 |     "h_pool1 = max_pool(h_conv1)     # 输出尺寸：14x14x32"
208 |    ]
209 |   },
210 |   {
211 |    "cell_type": "markdown",
212 |    "metadata": {},
213 |    "source": [
214 |     "### CONV2"
215 |    ]
216 |   },
217 |   {
218 |    "cell_type": "code",
219 |    "execution_count": 9,
220 |    "metadata": {},
221 |    "outputs": [],
222 |    "source": [
223 |     "W_conv2 = weight_variable([5, 5, 32, 64])  # 卷积核尺寸：5x5, 输入通道：32, 输出通道：64\n",
224 |     "b_conv2 = bias_variable([64])\n",
225 |     "h_conv2 = tf.nn.relu(conv2d(h_pool1, W_conv2) + b_conv2)  # 输出尺寸：14x14x64\n",
226 |     "h_pool2 = max_pool(h_conv2)     # 输出尺寸：7x7x64"
227 |    ]
228 |   },
229 |   {
230 |    "cell_type": "markdown",
231 |    "metadata": {},
232 |    "source": [
233 |     "### FC1"
234 |    ]
235 |   },
236 |   {
237 |    "cell_type": "code",
238 |    "execution_count": 10,
239 |    "metadata": {},
240 |    "outputs": [],
241 |    "source": [
242 |     "W_fc1 = weight_variable([7*7*64, 1024])\n",
243 |     "b_fc1 = bias_variable([1024])\n",
244 |     "h_pool2_flat = tf.reshape(h_pool2, [-1, 7*7*64])  # 展开为一维向量\n",
245 |     "h_fc1 = tf.nn.relu(tf.matmul(h_pool2_flat, W_fc1) + b_fc1)"
246 |    ]
247 |   },
248 |   {
249 |    "cell_type": "markdown",
250 |    "metadata": {},
251 |    "source": [
252 |     "### FC2"
253 |    ]
254 |   },
255 |   {
256 |    "cell_type": "code",
257 |    "execution_count": 11,
258 |    "metadata": {},
259 |    "outputs": [],
260 |    "source": [
261 |     "W_fc2 = weight_variable([1024, 10])\n",
262 |     "b_fc2 = bias_variable([10])\n",
263 |     "prediction = tf.nn.softmax(tf.matmul(h_fc1, W_fc2) + b_fc2)"
264 |    ]
265 |   },
266 |   {
267 |    "cell_type": "markdown",
268 |    "metadata": {},
269 |    "source": [
270 |     "## 定义损失函数与优化算法"
271 |    ]
272 |   },
273 |   {
274 |    "cell_type": "code",
275 |    "execution_count": 12,
276 |    "metadata": {},
277 |    "outputs": [],
278 |    "source": [
279 |     "cross_entropy = tf.reduce_mean(-tf.reduce_sum(y * tf.log(prediction),reduction_indices=[1]))   \n",
280 |     "train_step = tf.train.AdamOptimizer(1e-4).minimize(cross_entropy)"
281 |    ]
282 |   },
283 |   {
284 |    "cell_type": "code",
285 |    "execution_count": 13,
286 |    "metadata": {},
287 |    "outputs": [],
288 |    "source": [
289 |     "#判断预测标签和实际标签是否匹配\n",
290 |     "correct_prediction = tf.equal(tf.argmax(prediction,1), tf.argmax(y,1))\n",
291 |     "accuracy = tf.reduce_mean(tf.cast(correct_prediction,\"float\"))"
292 |    ]
293 |   },
294 |   {
295 |    "cell_type": "markdown",
296 |    "metadata": {},
297 |    "source": [
298 |     "## 训练并验证准确率"
299 |    ]
300 |   },
301 |   {
302 |    "cell_type": "code",
303 |    "execution_count": 14,
304 |    "metadata": {},
305 |    "outputs": [
306 |     {
307 |      "name": "stdout",
308 |      "output_type": "stream",
309 |      "text": [
310 |       "train accuracy 0.810\n",
311 |       "train accuracy 0.890\n",
312 |       "train accuracy 0.880\n",
313 |       "train accuracy 0.960\n",
314 |       "train accuracy 0.950\n",
315 |       "train accuracy 0.940\n",
316 |       "train accuracy 0.950\n",
317 |       "train accuracy 0.940\n",
318 |       "train accuracy 0.940\n",
319 |       "train accuracy 0.960\n",
320 |       "train accuracy 0.980\n",
321 |       "train accuracy 0.950\n",
322 |       "train accuracy 0.940\n",
323 |       "train accuracy 0.980\n",
324 |       "train accuracy 0.980\n",
325 |       "train accuracy 0.980\n",
326 |       "train accuracy 1.000\n",
327 |       "train accuracy 0.990\n",
328 |       "train accuracy 0.990\n",
329 |       "train accuracy 0.990\n",
330 |       "test accuracy 0.977\n"
331 |      ]
332 |     }
333 |    ],
334 |    "source": [
335 |     "sess = tf.Session()\n",
336 |     "init = tf.global_variables_initializer()\n",
337 |     "sess.run(init)\n",
338 |     "\n",
339 |     "cost = []\n",
340 |     "\n",
341 |     "for i in range(1000):\n",
342 |     "    batch_x, batch_y = mnist.train.next_batch(100)\n",
343 |     "    sess.run(train_step, feed_dict={x: batch_x, y: batch_y})\n",
344 |     "    if (i+1) % 50 == 0:\n",
345 |     "        print(\"train accuracy %.3f\" % accuracy.eval(session = sess,\n",
346 |     "                feed_dict = {x:batch_x, y:batch_y}))\n",
347 |     "print(\"test accuracy %.3f\" % accuracy.eval(session = sess,\n",
348 |     "    feed_dict = {x:mnist.test.images, y:mnist.test.labels}))"
349 |    ]
350 |   },
351 |   {
352 |    "cell_type": "code",
353 |    "execution_count": null,
354 |    "metadata": {},
355 |    "outputs": [],
356 |    "source": []
357 |   }
358 |  ],
359 |  "metadata": {
360 |   "kernelspec": {
361 |    "display_name": "Python 3",
362 |    "language": "python",
363 |    "name": "python3"
364 |   },
365 |   "language_info": {
366 |    "codemirror_mode": {
367 |     "name": "ipython",
368 |     "version": 3
369 |    },
370 |    "file_extension": ".py",
371 |    "mimetype": "text/x-python",
372 |    "name": "python",
373 |    "nbconvert_exporter": "python",
374 |    "pygments_lexer": "ipython3",
375 |    "version": "3.5.6"
376 |   }
377 |  },
378 |  "nbformat": 4,
379 |  "nbformat_minor": 2
380 | }
381 | 


--------------------------------------------------------------------------------
/09_rnn_pytorch.ipynb:
--------------------------------------------------------------------------------
  1 | {
  2 |  "cells": [
  3 |   {
  4 |    "cell_type": "markdown",
  5 |    "metadata": {},
  6 |    "source": [
  7 |     "# 第 9 章：循环神经网络 — PyTorch"
  8 |    ]
  9 |   },
 10 |   {
 11 |    "cell_type": "markdown",
 12 |    "metadata": {},
 13 |    "source": [
 14 |     "## 导入 MNIST 数据集"
 15 |    ]
 16 |   },
 17 |   {
 18 |    "cell_type": "code",
 19 |    "execution_count": 1,
 20 |    "metadata": {},
 21 |    "outputs": [],
 22 |    "source": [
 23 |     "import torch\n",
 24 |     "import torchvision\n",
 25 |     "import torchvision.transforms as transforms\n",
 26 |     "import torch.nn as nn\n",
 27 |     "import torch.nn.functional as F\n",
 28 |     "import torch.optim as optim\n",
 29 |     "import matplotlib.pyplot as plt\n",
 30 |     "import numpy as np\n",
 31 |     "\n",
 32 |     "transform = transforms.Compose(\n",
 33 |     "    [transforms.ToTensor()])\n",
 34 |     "\n",
 35 |     "# 训练集\n",
 36 |     "trainset = torchvision.datasets.MNIST(root='./datasets/ch08/pytorch',     # 选择数据的根目录\n",
 37 |     "                                      train=True,\n",
 38 |     "                                      download=True,    # 不从网络上download图片\n",
 39 |     "                                      transform=transform)\n",
 40 |     "trainloader = torch.utils.data.DataLoader(trainset, batch_size=4,\n",
 41 |     "                                         shuffle=True, num_workers=2)\n",
 42 |     "# 测试集\n",
 43 |     "testset = torchvision.datasets.MNIST(root='./datasets/ch08/pytorch',     # 选择数据的根目录\n",
 44 |     "                                     train=False,\n",
 45 |     "                                     download=True,    # 不从网络上download图片\n",
 46 |     "                                     transform=transform)\n",
 47 |     "testloader = torch.utils.data.DataLoader(testset, batch_size=4,\n",
 48 |     "                                        shuffle=False, num_workers=2)"
 49 |    ]
 50 |   },
 51 |   {
 52 |    "cell_type": "code",
 53 |    "execution_count": 2,
 54 |    "metadata": {},
 55 |    "outputs": [
 56 |     {
 57 |      "name": "stdout",
 58 |      "output_type": "stream",
 59 |      "text": [
 60 |       "Dataset MNIST\n",
 61 |       "    Number of datapoints: 60000\n",
 62 |       "    Split: train\n",
 63 |       "    Root Location: ./datasets/ch08/pytorch\n",
 64 |       "    Transforms (if any): Compose(\n",
 65 |       "                             ToTensor()\n",
 66 |       "                         )\n",
 67 |       "    Target Transforms (if any): None\n"
 68 |      ]
 69 |     }
 70 |    ],
 71 |    "source": [
 72 |     "print(trainset)"
 73 |    ]
 74 |   },
 75 |   {
 76 |    "cell_type": "code",
 77 |    "execution_count": 3,
 78 |    "metadata": {},
 79 |    "outputs": [
 80 |     {
 81 |      "name": "stdout",
 82 |      "output_type": "stream",
 83 |      "text": [
 84 |       "Dataset MNIST\n",
 85 |       "    Number of datapoints: 10000\n",
 86 |       "    Split: test\n",
 87 |       "    Root Location: ./datasets/ch08/pytorch\n",
 88 |       "    Transforms (if any): Compose(\n",
 89 |       "                             ToTensor()\n",
 90 |       "                         )\n",
 91 |       "    Target Transforms (if any): None\n"
 92 |      ]
 93 |     }
 94 |    ],
 95 |    "source": [
 96 |     "print(testset)"
 97 |    ]
 98 |   },
 99 |   {
100 |    "cell_type": "markdown",
101 |    "metadata": {},
102 |    "source": [
103 |     "接下来展示一些训练样本图像"
104 |    ]
105 |   },
106 |   {
107 |    "cell_type": "code",
108 |    "execution_count": 4,
109 |    "metadata": {},
110 |    "outputs": [
111 |     {
112 |      "data": {
113 |       "image/png": "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\n",
114 |       "text/plain": [
115 |        "<Figure size 432x288 with 1 Axes>"
116 |       ]
117 |      },
118 |      "metadata": {
119 |       "needs_background": "light"
120 |      },
121 |      "output_type": "display_data"
122 |     },
123 |     {
124 |      "name": "stdout",
125 |      "output_type": "stream",
126 |      "text": [
127 |       "          0           7           9           6\n"
128 |      ]
129 |     }
130 |    ],
131 |    "source": [
132 |     "def imshow(img):\n",
133 |     "    npimg = img.numpy()\n",
134 |     "    plt.imshow(np.transpose(npimg, (1, 2, 0)))\n",
135 |     "\n",
136 |     "# 选择一个 batch 的图片\n",
137 |     "dataiter = iter(trainloader)\n",
138 |     "images, labels = dataiter.next()\n",
139 |     "\n",
140 |     "# 显示图片\n",
141 |     "imshow(torchvision.utils.make_grid(images))\n",
142 |     "plt.show()\n",
143 |     "# 打印 labels\n",
144 |     "print(' '.join('%11s' % labels[j].numpy() for j in range(4)))"
145 |    ]
146 |   },
147 |   {
148 |    "cell_type": "markdown",
149 |    "metadata": {},
150 |    "source": [
151 |     "## 定义循环神经网络"
152 |    ]
153 |   },
154 |   {
155 |    "cell_type": "code",
156 |    "execution_count": 5,
157 |    "metadata": {},
158 |    "outputs": [],
159 |    "source": [
160 |     "class Net(nn.Module):\n",
161 |     "    def __init__(self):\n",
162 |     "        super(Net, self).__init__()\n",
163 |     "        self.rnn = nn.LSTM(          # 使用 LSTM 结构\n",
164 |     "            input_size = 28,         # 输入每个元素的维度，即图片每行包含 28 个像素点\n",
165 |     "            hidden_size = 84,        # 隐藏层神经元设置为 84 个\n",
166 |     "            num_layers=2,            # 隐藏层数目，两层\n",
167 |     "            batch_first=True,        # 是否将 batch 放在维度的第一位，(batch, time_step, input_size)\n",
168 |     "        )\n",
169 |     "        self.out = nn.Linear(84, 10) # 输出层，包含 10 个神经元，对应 0～9 数字\n",
170 |     "\n",
171 |     "    def forward(self, x):\n",
172 |     "        r_out, (h_n, h_c) = self.rnn(x, None)   \n",
173 |     "        # 选择图片的最后一行作为 RNN 输出\n",
174 |     "        out = self.out(r_out[:, -1, :])\n",
175 |     "        return out"
176 |    ]
177 |   },
178 |   {
179 |    "cell_type": "code",
180 |    "execution_count": 6,
181 |    "metadata": {},
182 |    "outputs": [
183 |     {
184 |      "name": "stdout",
185 |      "output_type": "stream",
186 |      "text": [
187 |       "Net(\n",
188 |       "  (rnn): LSTM(28, 84, num_layers=2, batch_first=True)\n",
189 |       "  (out): Linear(in_features=84, out_features=10, bias=True)\n",
190 |       ")\n"
191 |      ]
192 |     }
193 |    ],
194 |    "source": [
195 |     "net = Net()\n",
196 |     "print(net) "
197 |    ]
198 |   },
199 |   {
200 |    "cell_type": "markdown",
201 |    "metadata": {},
202 |    "source": [
203 |     "## 定义损失函数与优化算法"
204 |    ]
205 |   },
206 |   {
207 |    "cell_type": "code",
208 |    "execution_count": 7,
209 |    "metadata": {},
210 |    "outputs": [],
211 |    "source": [
212 |     "criterion = nn.CrossEntropyLoss()\n",
213 |     "optimizer = optim.Adam(net.parameters(), lr=0.0001)"
214 |    ]
215 |   },
216 |   {
217 |    "cell_type": "markdown",
218 |    "metadata": {},
219 |    "source": [
220 |     "## 训练网络"
221 |    ]
222 |   },
223 |   {
224 |    "cell_type": "code",
225 |    "execution_count": 8,
226 |    "metadata": {},
227 |    "outputs": [
228 |     {
229 |      "name": "stdout",
230 |      "output_type": "stream",
231 |      "text": [
232 |       "[epoch: 1, mini-batch:  2000] loss: 1.592\n",
233 |       "[epoch: 1, mini-batch:  4000] loss: 0.789\n",
234 |       "[epoch: 1, mini-batch:  6000] loss: 0.564\n",
235 |       "[epoch: 1, mini-batch:  8000] loss: 0.429\n",
236 |       "[epoch: 1, mini-batch: 10000] loss: 0.357\n",
237 |       "[epoch: 1, mini-batch: 12000] loss: 0.294\n",
238 |       "[epoch: 1, mini-batch: 14000] loss: 0.290\n",
239 |       "[epoch: 2, mini-batch:  2000] loss: 0.236\n",
240 |       "[epoch: 2, mini-batch:  4000] loss: 0.213\n",
241 |       "[epoch: 2, mini-batch:  6000] loss: 0.196\n",
242 |       "[epoch: 2, mini-batch:  8000] loss: 0.195\n",
243 |       "[epoch: 2, mini-batch: 10000] loss: 0.173\n",
244 |       "[epoch: 2, mini-batch: 12000] loss: 0.161\n",
245 |       "[epoch: 2, mini-batch: 14000] loss: 0.161\n",
246 |       "[epoch: 3, mini-batch:  2000] loss: 0.140\n",
247 |       "[epoch: 3, mini-batch:  4000] loss: 0.124\n",
248 |       "[epoch: 3, mini-batch:  6000] loss: 0.141\n",
249 |       "[epoch: 3, mini-batch:  8000] loss: 0.133\n",
250 |       "[epoch: 3, mini-batch: 10000] loss: 0.121\n",
251 |       "[epoch: 3, mini-batch: 12000] loss: 0.129\n",
252 |       "[epoch: 3, mini-batch: 14000] loss: 0.107\n",
253 |       "[epoch: 4, mini-batch:  2000] loss: 0.110\n",
254 |       "[epoch: 4, mini-batch:  4000] loss: 0.099\n",
255 |       "[epoch: 4, mini-batch:  6000] loss: 0.101\n",
256 |       "[epoch: 4, mini-batch:  8000] loss: 0.101\n",
257 |       "[epoch: 4, mini-batch: 10000] loss: 0.094\n",
258 |       "[epoch: 4, mini-batch: 12000] loss: 0.087\n",
259 |       "[epoch: 4, mini-batch: 14000] loss: 0.098\n",
260 |       "[epoch: 5, mini-batch:  2000] loss: 0.082\n",
261 |       "[epoch: 5, mini-batch:  4000] loss: 0.094\n",
262 |       "[epoch: 5, mini-batch:  6000] loss: 0.082\n",
263 |       "[epoch: 5, mini-batch:  8000] loss: 0.082\n",
264 |       "[epoch: 5, mini-batch: 10000] loss: 0.081\n",
265 |       "[epoch: 5, mini-batch: 12000] loss: 0.078\n",
266 |       "[epoch: 5, mini-batch: 14000] loss: 0.078\n"
267 |      ]
268 |     }
269 |    ],
270 |    "source": [
271 |     "num_epoches = 5    # 设置 epoch 数目\n",
272 |     "cost = []     # 损失函数累加\n",
273 |     "\n",
274 |     "for epoch in range(num_epoches):    \n",
275 |     "\n",
276 |     "    running_loss = 0.0\n",
277 |     "    for i, data in enumerate(trainloader, 0):\n",
278 |     "        # 输入样本和标签\n",
279 |     "        inputs, labels = data\n",
280 |     "        inputs = inputs.view(-1, 28, 28)  # 设置 RNN 输入维度为 (batch, time_step, input_size)\n",
281 |     "\n",
282 |     "        # 每次训练梯度清零\n",
283 |     "        optimizer.zero_grad()\n",
284 |     "\n",
285 |     "        # 正向传播、反向传播和优化过程\n",
286 |     "        outputs = net(inputs)\n",
287 |     "        loss = criterion(outputs, labels)\n",
288 |     "        loss.backward()\n",
289 |     "        optimizer.step()\n",
290 |     "\n",
291 |     "        # 打印训练情况\n",
292 |     "        running_loss += loss.item()\n",
293 |     "        if i % 2000 == 1999:    # 每隔2000 mini-batches，打印一次\n",
294 |     "            print('[epoch: %d, mini-batch: %5d] loss: %.3f' % \n",
295 |     "                 (epoch + 1, i + 1, running_loss / 2000))\n",
296 |     "            cost.append(running_loss / 2000)\n",
297 |     "            running_loss = 0.0"
298 |    ]
299 |   },
300 |   {
301 |    "cell_type": "code",
302 |    "execution_count": 9,
303 |    "metadata": {},
304 |    "outputs": [
305 |     {
306 |      "data": {
307 |       "image/png": "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\n",
308 |       "text/plain": [
309 |        "<Figure size 432x288 with 1 Axes>"
310 |       ]
311 |      },
312 |      "metadata": {
313 |       "needs_background": "light"
314 |      },
315 |      "output_type": "display_data"
316 |     }
317 |    ],
318 |    "source": [
319 |     "plt.plot(cost)\n",
320 |     "plt.xlabel('mini-batches(per 2000)')\n",
321 |     "plt.ylabel('running_loss')\n",
322 |     "plt.show()"
323 |    ]
324 |   },
325 |   {
326 |    "cell_type": "markdown",
327 |    "metadata": {},
328 |    "source": [
329 |     "## 测试数据"
330 |    ]
331 |   },
332 |   {
333 |    "cell_type": "code",
334 |    "execution_count": 10,
335 |    "metadata": {},
336 |    "outputs": [
337 |     {
338 |      "name": "stdout",
339 |      "output_type": "stream",
340 |      "text": [
341 |       "Accuracy of the network on the 60000 test images: 96.355 %\n"
342 |      ]
343 |     }
344 |    ],
345 |    "source": [
346 |     "correct = 0\n",
347 |     "total = 0\n",
348 |     "with torch.no_grad():\n",
349 |     "    for data in trainloader:\n",
350 |     "        images, labels = data\n",
351 |     "        images = images.view(-1, 28, 28)\n",
352 |     "        outputs = net(images)\n",
353 |     "        _, predicted = torch.max(outputs.data, 1)\n",
354 |     "        total += labels.size(0)\n",
355 |     "        correct += (predicted == labels).sum().item()\n",
356 |     "\n",
357 |     "print('Accuracy of the network on the 60000 test images: %.3f %%' % \n",
358 |     "     (100 * correct / total))"
359 |    ]
360 |   },
361 |   {
362 |    "cell_type": "code",
363 |    "execution_count": 11,
364 |    "metadata": {},
365 |    "outputs": [
366 |     {
367 |      "name": "stdout",
368 |      "output_type": "stream",
369 |      "text": [
370 |       "Accuracy of the network on the 10000 test images: 95.790 %\n"
371 |      ]
372 |     }
373 |    ],
374 |    "source": [
375 |     "correct = 0\n",
376 |     "total = 0\n",
377 |     "with torch.no_grad():\n",
378 |     "    for data in testloader:\n",
379 |     "        images, labels = data\n",
380 |     "        images = images.view(-1, 28, 28)\n",
381 |     "        outputs = net(images)\n",
382 |     "        _, predicted = torch.max(outputs.data, 1)\n",
383 |     "        total += labels.size(0)\n",
384 |     "        correct += (predicted == labels).sum().item()\n",
385 |     "\n",
386 |     "print('Accuracy of the network on the 10000 test images: %.3f %%' % \n",
387 |     "     (100 * correct / total))"
388 |    ]
389 |   },
390 |   {
391 |    "cell_type": "code",
392 |    "execution_count": null,
393 |    "metadata": {},
394 |    "outputs": [],
395 |    "source": []
396 |   }
397 |  ],
398 |  "metadata": {
399 |   "kernelspec": {
400 |    "display_name": "Python 3",
401 |    "language": "python",
402 |    "name": "python3"
403 |   },
404 |   "language_info": {
405 |    "codemirror_mode": {
406 |     "name": "ipython",
407 |     "version": 3
408 |    },
409 |    "file_extension": ".py",
410 |    "mimetype": "text/x-python",
411 |    "name": "python",
412 |    "nbconvert_exporter": "python",
413 |    "pygments_lexer": "ipython3",
414 |    "version": "3.7.2"
415 |   }
416 |  },
417 |  "nbformat": 4,
418 |  "nbformat_minor": 2
419 | }
420 | 


--------------------------------------------------------------------------------
/09_rnn_tensorflow.ipynb:
--------------------------------------------------------------------------------
  1 | {
  2 |  "cells": [
  3 |   {
  4 |    "cell_type": "markdown",
  5 |    "metadata": {},
  6 |    "source": [
  7 |     "# 循环神经网络—TensorFlow"
  8 |    ]
  9 |   },
 10 |   {
 11 |    "cell_type": "code",
 12 |    "execution_count": 1,
 13 |    "metadata": {},
 14 |    "outputs": [],
 15 |    "source": [
 16 |     "from tensorflow.examples.tutorials.mnist import input_data\n",
 17 |     "import tensorflow as tf\n",
 18 |     "import matplotlib.pyplot as plt"
 19 |    ]
 20 |   },
 21 |   {
 22 |    "cell_type": "markdown",
 23 |    "metadata": {},
 24 |    "source": [
 25 |     "## 导入 MNIST 数据集"
 26 |    ]
 27 |   },
 28 |   {
 29 |    "cell_type": "code",
 30 |    "execution_count": 2,
 31 |    "metadata": {},
 32 |    "outputs": [
 33 |     {
 34 |      "name": "stdout",
 35 |      "output_type": "stream",
 36 |      "text": [
 37 |       "WARNING:tensorflow:From <ipython-input-2-3f5d8b810895>:1: read_data_sets (from tensorflow.contrib.learn.python.learn.datasets.mnist) is deprecated and will be removed in a future version.\n",
 38 |       "Instructions for updating:\n",
 39 |       "Please use alternatives such as official/mnist/dataset.py from tensorflow/models.\n",
 40 |       "WARNING:tensorflow:From C:\\ProgramData\\Anaconda3\\envs\\tensorflow\\lib\\site-packages\\tensorflow\\contrib\\learn\\python\\learn\\datasets\\mnist.py:260: maybe_download (from tensorflow.contrib.learn.python.learn.datasets.base) is deprecated and will be removed in a future version.\n",
 41 |       "Instructions for updating:\n",
 42 |       "Please write your own downloading logic.\n",
 43 |       "WARNING:tensorflow:From C:\\ProgramData\\Anaconda3\\envs\\tensorflow\\lib\\site-packages\\tensorflow\\contrib\\learn\\python\\learn\\datasets\\mnist.py:262: extract_images (from tensorflow.contrib.learn.python.learn.datasets.mnist) is deprecated and will be removed in a future version.\n",
 44 |       "Instructions for updating:\n",
 45 |       "Please use tf.data to implement this functionality.\n",
 46 |       "Extracting ./datasets/ch08/tensorflow/MNIST\\train-images-idx3-ubyte.gz\n",
 47 |       "WARNING:tensorflow:From C:\\ProgramData\\Anaconda3\\envs\\tensorflow\\lib\\site-packages\\tensorflow\\contrib\\learn\\python\\learn\\datasets\\mnist.py:267: extract_labels (from tensorflow.contrib.learn.python.learn.datasets.mnist) is deprecated and will be removed in a future version.\n",
 48 |       "Instructions for updating:\n",
 49 |       "Please use tf.data to implement this functionality.\n",
 50 |       "Extracting ./datasets/ch08/tensorflow/MNIST\\train-labels-idx1-ubyte.gz\n",
 51 |       "WARNING:tensorflow:From C:\\ProgramData\\Anaconda3\\envs\\tensorflow\\lib\\site-packages\\tensorflow\\contrib\\learn\\python\\learn\\datasets\\mnist.py:110: dense_to_one_hot (from tensorflow.contrib.learn.python.learn.datasets.mnist) is deprecated and will be removed in a future version.\n",
 52 |       "Instructions for updating:\n",
 53 |       "Please use tf.one_hot on tensors.\n",
 54 |       "Extracting ./datasets/ch08/tensorflow/MNIST\\t10k-images-idx3-ubyte.gz\n",
 55 |       "Extracting ./datasets/ch08/tensorflow/MNIST\\t10k-labels-idx1-ubyte.gz\n",
 56 |       "WARNING:tensorflow:From C:\\ProgramData\\Anaconda3\\envs\\tensorflow\\lib\\site-packages\\tensorflow\\contrib\\learn\\python\\learn\\datasets\\mnist.py:290: DataSet.__init__ (from tensorflow.contrib.learn.python.learn.datasets.mnist) is deprecated and will be removed in a future version.\n",
 57 |       "Instructions for updating:\n",
 58 |       "Please use alternatives such as official/mnist/dataset.py from tensorflow/models.\n"
 59 |      ]
 60 |     }
 61 |    ],
 62 |    "source": [
 63 |     "mnist = input_data.read_data_sets('./datasets/ch08/tensorflow/MNIST',one_hot=True)"
 64 |    ]
 65 |   },
 66 |   {
 67 |    "cell_type": "code",
 68 |    "execution_count": 3,
 69 |    "metadata": {},
 70 |    "outputs": [
 71 |     {
 72 |      "name": "stdout",
 73 |      "output_type": "stream",
 74 |      "text": [
 75 |       "(55000, 784)\n",
 76 |       "(55000, 10)\n"
 77 |      ]
 78 |     }
 79 |    ],
 80 |    "source": [
 81 |     "print(mnist.train.images.shape)\n",
 82 |     "print(mnist.train.labels.shape)"
 83 |    ]
 84 |   },
 85 |   {
 86 |    "cell_type": "code",
 87 |    "execution_count": 4,
 88 |    "metadata": {},
 89 |    "outputs": [
 90 |     {
 91 |      "name": "stdout",
 92 |      "output_type": "stream",
 93 |      "text": [
 94 |       "(5000, 784)\n",
 95 |       "(5000, 10)\n"
 96 |      ]
 97 |     }
 98 |    ],
 99 |    "source": [
100 |     "print(mnist.validation.images.shape)\n",
101 |     "print(mnist.validation.labels.shape)"
102 |    ]
103 |   },
104 |   {
105 |    "cell_type": "code",
106 |    "execution_count": 5,
107 |    "metadata": {},
108 |    "outputs": [
109 |     {
110 |      "name": "stdout",
111 |      "output_type": "stream",
112 |      "text": [
113 |       "(10000, 784)\n",
114 |       "(10000, 10)\n"
115 |      ]
116 |     }
117 |    ],
118 |    "source": [
119 |     "print(mnist.test.images.shape)\n",
120 |     "print(mnist.test.labels.shape)"
121 |    ]
122 |   },
123 |   {
124 |    "cell_type": "markdown",
125 |    "metadata": {},
126 |    "source": [
127 |     "## 定义占位符 placeholder"
128 |    ]
129 |   },
130 |   {
131 |    "cell_type": "code",
132 |    "execution_count": 6,
133 |    "metadata": {},
134 |    "outputs": [],
135 |    "source": [
136 |     "# 参数设置\n",
137 |     "batch_size = 100        # BATCH 的大小，相当于一次处理100个image\n",
138 |     "time_step = 28          # 一个LSTM中，输入序列的长度Tx，image有28行\n",
139 |     "input_size = 28         # 单个x向量长度，image有28列\n",
140 |     "lr = 0.001              # 学习率\n",
141 |     "num_units = 100         # 隐藏层多少个LTSM单元\n",
142 |     "iterations =1000        # 迭代训练次数\n",
143 |     "classes =10             # 输出大小，0-9十个数字的概率\n",
144 |     "\n",
145 |     "# 定义 placeholders\n",
146 |     "# 维度是[batch_size，time_step * input_size]\n",
147 |     "x = tf.placeholder(tf.float32, [None, time_step * input_size]) \n",
148 |     "# 输入的是二维数据，将其还原为三维，维度是[batch_size, time_step, input_size]\n",
149 |     "x_image = tf.reshape(x, [-1, time_step, input_size])                   \n",
150 |     "y = tf.placeholder(tf.int32, [None, classes])                     "
151 |    ]
152 |   },
153 |   {
154 |    "cell_type": "markdown",
155 |    "metadata": {},
156 |    "source": [
157 |     "## 定义RNN（LSTM）结构"
158 |    ]
159 |   },
160 |   {
161 |    "cell_type": "code",
162 |    "execution_count": 7,
163 |    "metadata": {},
164 |    "outputs": [],
165 |    "source": [
166 |     "rnn_cell = tf.contrib.rnn.BasicLSTMCell(num_units=num_units) \n",
167 |     "outputs,final_state = tf.nn.dynamic_rnn(\n",
168 |     "    cell=rnn_cell,              # 选择传入的cell\n",
169 |     "    inputs=x_image,             # 传入的数据\n",
170 |     "    initial_state=None,         # 初始状态\n",
171 |     "    dtype=tf.float32,           # 数据类型\n",
172 |     "    time_major=False,           # False: (batch, time_step, input); True: (time_step, batch, input)，这里根据x_image结构选择False\n",
173 |     ")\n",
174 |     "output = tf.layers.dense(inputs=outputs[:, -1, :], units=classes)    "
175 |    ]
176 |   },
177 |   {
178 |    "cell_type": "markdown",
179 |    "metadata": {},
180 |    "source": [
181 |     "## 定义损失函数与优化算法"
182 |    ]
183 |   },
184 |   {
185 |    "cell_type": "code",
186 |    "execution_count": 8,
187 |    "metadata": {},
188 |    "outputs": [],
189 |    "source": [
190 |     "cross_entropy = tf.losses.softmax_cross_entropy(onehot_labels=y, logits=output)      # 计算loss\n",
191 |     "train_step = tf.train.AdamOptimizer(lr).minimize(cross_entropy)      #选择优化方法"
192 |    ]
193 |   },
194 |   {
195 |    "cell_type": "code",
196 |    "execution_count": 9,
197 |    "metadata": {},
198 |    "outputs": [],
199 |    "source": [
200 |     "#判断预测标签和实际标签是否匹配\n",
201 |     "correct_prediction = tf.equal(tf.argmax(y, axis=1),tf.argmax(output, axis=1))\n",
202 |     "accuracy = tf.reduce_mean(tf.cast(correct_prediction,'float'))  #计算正确率"
203 |    ]
204 |   },
205 |   {
206 |    "cell_type": "markdown",
207 |    "metadata": {},
208 |    "source": [
209 |     "## 训练并验证准确率"
210 |    ]
211 |   },
212 |   {
213 |    "cell_type": "code",
214 |    "execution_count": 10,
215 |    "metadata": {},
216 |    "outputs": [
217 |     {
218 |      "name": "stdout",
219 |      "output_type": "stream",
220 |      "text": [
221 |       "train accuracy 0.560\n",
222 |       "train accuracy 0.740\n",
223 |       "train accuracy 0.900\n",
224 |       "train accuracy 0.810\n",
225 |       "train accuracy 0.940\n",
226 |       "train accuracy 0.890\n",
227 |       "train accuracy 0.880\n",
228 |       "train accuracy 0.910\n",
229 |       "train accuracy 0.910\n",
230 |       "train accuracy 0.930\n",
231 |       "train accuracy 0.870\n",
232 |       "train accuracy 0.960\n",
233 |       "train accuracy 0.960\n",
234 |       "train accuracy 0.940\n",
235 |       "train accuracy 0.930\n",
236 |       "train accuracy 0.960\n",
237 |       "train accuracy 0.940\n",
238 |       "train accuracy 0.980\n",
239 |       "train accuracy 0.920\n",
240 |       "train accuracy 0.980\n",
241 |       "test accuracy 0.958\n"
242 |      ]
243 |     }
244 |    ],
245 |    "source": [
246 |     "sess = tf.Session()\n",
247 |     "init = tf.global_variables_initializer()\n",
248 |     "sess.run(init)\n",
249 |     "\n",
250 |     "for i in range(iterations):\n",
251 |     "    batch_x, batch_y = mnist.train.next_batch(batch_size)\n",
252 |     "    sess.run(train_step, feed_dict={x: batch_x, y: batch_y})\n",
253 |     "    if (i+1) % 50 == 0:\n",
254 |     "        print(\"train accuracy %.3f\" % accuracy.eval(session = sess,\n",
255 |     "                feed_dict = {x:batch_x, y:batch_y}))\n",
256 |     "print(\"test accuracy %.3f\" % accuracy.eval(session = sess,\n",
257 |     "    feed_dict = {x:mnist.test.images, y:mnist.test.labels}))"
258 |    ]
259 |   },
260 |   {
261 |    "cell_type": "code",
262 |    "execution_count": null,
263 |    "metadata": {},
264 |    "outputs": [],
265 |    "source": []
266 |   }
267 |  ],
268 |  "metadata": {
269 |   "kernelspec": {
270 |    "display_name": "Python 3",
271 |    "language": "python",
272 |    "name": "python3"
273 |   },
274 |   "language_info": {
275 |    "codemirror_mode": {
276 |     "name": "ipython",
277 |     "version": 3
278 |    },
279 |    "file_extension": ".py",
280 |    "mimetype": "text/x-python",
281 |    "name": "python",
282 |    "nbconvert_exporter": "python",
283 |    "pygments_lexer": "ipython3",
284 |    "version": "3.5.6"
285 |   }
286 |  },
287 |  "nbformat": 4,
288 |  "nbformat_minor": 2
289 | }
290 | 


--------------------------------------------------------------------------------
/LICENSE:
--------------------------------------------------------------------------------
  1 |                                  Apache License
  2 |                            Version 2.0, January 2004
  3 |                         http://www.apache.org/licenses/
  4 | 
  5 |    TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
  6 | 
  7 |    1. Definitions.
  8 | 
  9 |       "License" shall mean the terms and conditions for use, reproduction,
 10 |       and distribution as defined by Sections 1 through 9 of this document.
 11 | 
 12 |       "Licensor" shall mean the copyright owner or entity authorized by
 13 |       the copyright owner that is granting the License.
 14 | 
 15 |       "Legal Entity" shall mean the union of the acting entity and all
 16 |       other entities that control, are controlled by, or are under common
 17 |       control with that entity. For the purposes of this definition,
 18 |       "control" means (i) the power, direct or indirect, to cause the
 19 |       direction or management of such entity, whether by contract or
 20 |       otherwise, or (ii) ownership of fifty percent (50%) or more of the
 21 |       outstanding shares, or (iii) beneficial ownership of such entity.
 22 | 
 23 |       "You" (or "Your") shall mean an individual or Legal Entity
 24 |       exercising permissions granted by this License.
 25 | 
 26 |       "Source" form shall mean the preferred form for making modifications,
 27 |       including but not limited to software source code, documentation
 28 |       source, and configuration files.
 29 | 
 30 |       "Object" form shall mean any form resulting from mechanical
 31 |       transformation or translation of a Source form, including but
 32 |       not limited to compiled object code, generated documentation,
 33 |       and conversions to other media types.
 34 | 
 35 |       "Work" shall mean the work of authorship, whether in Source or
 36 |       Object form, made available under the License, as indicated by a
 37 |       copyright notice that is included in or attached to the work
 38 |       (an example is provided in the Appendix below).
 39 | 
 40 |       "Derivative Works" shall mean any work, whether in Source or Object
 41 |       form, that is based on (or derived from) the Work and for which the
 42 |       editorial revisions, annotations, elaborations, or other modifications
 43 |       represent, as a whole, an original work of authorship. For the purposes
 44 |       of this License, Derivative Works shall not include works that remain
 45 |       separable from, or merely link (or bind by name) to the interfaces of,
 46 |       the Work and Derivative Works thereof.
 47 | 
 48 |       "Contribution" shall mean any work of authorship, including
 49 |       the original version of the Work and any modifications or additions
 50 |       to that Work or Derivative Works thereof, that is intentionally
 51 |       submitted to Licensor for inclusion in the Work by the copyright owner
 52 |       or by an individual or Legal Entity authorized to submit on behalf of
 53 |       the copyright owner. For the purposes of this definition, "submitted"
 54 |       means any form of electronic, verbal, or written communication sent
 55 |       to the Licensor or its representatives, including but not limited to
 56 |       communication on electronic mailing lists, source code control systems,
 57 |       and issue tracking systems that are managed by, or on behalf of, the
 58 |       Licensor for the purpose of discussing and improving the Work, but
 59 |       excluding communication that is conspicuously marked or otherwise
 60 |       designated in writing by the copyright owner as "Not a Contribution."
 61 | 
 62 |       "Contributor" shall mean Licensor and any individual or Legal Entity
 63 |       on behalf of whom a Contribution has been received by Licensor and
 64 |       subsequently incorporated within the Work.
 65 | 
 66 |    2. Grant of Copyright License. Subject to the terms and conditions of
 67 |       this License, each Contributor hereby grants to You a perpetual,
 68 |       worldwide, non-exclusive, no-charge, royalty-free, irrevocable
 69 |       copyright license to reproduce, prepare Derivative Works of,
 70 |       publicly display, publicly perform, sublicense, and distribute the
 71 |       Work and such Derivative Works in Source or Object form.
 72 | 
 73 |    3. Grant of Patent License. Subject to the terms and conditions of
 74 |       this License, each Contributor hereby grants to You a perpetual,
 75 |       worldwide, non-exclusive, no-charge, royalty-free, irrevocable
 76 |       (except as stated in this section) patent license to make, have made,
 77 |       use, offer to sell, sell, import, and otherwise transfer the Work,
 78 |       where such license applies only to those patent claims licensable
 79 |       by such Contributor that are necessarily infringed by their
 80 |       Contribution(s) alone or by combination of their Contribution(s)
 81 |       with the Work to which such Contribution(s) was submitted. If You
 82 |       institute patent litigation against any entity (including a
 83 |       cross-claim or counterclaim in a lawsuit) alleging that the Work
 84 |       or a Contribution incorporated within the Work constitutes direct
 85 |       or contributory patent infringement, then any patent licenses
 86 |       granted to You under this License for that Work shall terminate
 87 |       as of the date such litigation is filed.
 88 | 
 89 |    4. Redistribution. You may reproduce and distribute copies of the
 90 |       Work or Derivative Works thereof in any medium, with or without
 91 |       modifications, and in Source or Object form, provided that You
 92 |       meet the following conditions:
 93 | 
 94 |       (a) You must give any other recipients of the Work or
 95 |           Derivative Works a copy of this License; and
 96 | 
 97 |       (b) You must cause any modified files to carry prominent notices
 98 |           stating that You changed the files; and
 99 | 
100 |       (c) You must retain, in the Source form of any Derivative Works
101 |           that You distribute, all copyright, patent, trademark, and
102 |           attribution notices from the Source form of the Work,
103 |           excluding those notices that do not pertain to any part of
104 |           the Derivative Works; and
105 | 
106 |       (d) If the Work includes a "NOTICE" text file as part of its
107 |           distribution, then any Derivative Works that You distribute must
108 |           include a readable copy of the attribution notices contained
109 |           within such NOTICE file, excluding those notices that do not
110 |           pertain to any part of the Derivative Works, in at least one
111 |           of the following places: within a NOTICE text file distributed
112 |           as part of the Derivative Works; within the Source form or
113 |           documentation, if provided along with the Derivative Works; or,
114 |           within a display generated by the Derivative Works, if and
115 |           wherever such third-party notices normally appear. The contents
116 |           of the NOTICE file are for informational purposes only and
117 |           do not modify the License. You may add Your own attribution
118 |           notices within Derivative Works that You distribute, alongside
119 |           or as an addendum to the NOTICE text from the Work, provided
120 |           that such additional attribution notices cannot be construed
121 |           as modifying the License.
122 | 
123 |       You may add Your own copyright statement to Your modifications and
124 |       may provide additional or different license terms and conditions
125 |       for use, reproduction, or distribution of Your modifications, or
126 |       for any such Derivative Works as a whole, provided Your use,
127 |       reproduction, and distribution of the Work otherwise complies with
128 |       the conditions stated in this License.
129 | 
130 |    5. Submission of Contributions. Unless You explicitly state otherwise,
131 |       any Contribution intentionally submitted for inclusion in the Work
132 |       by You to the Licensor shall be under the terms and conditions of
133 |       this License, without any additional terms or conditions.
134 |       Notwithstanding the above, nothing herein shall supersede or modify
135 |       the terms of any separate license agreement you may have executed
136 |       with Licensor regarding such Contributions.
137 | 
138 |    6. Trademarks. This License does not grant permission to use the trade
139 |       names, trademarks, service marks, or product names of the Licensor,
140 |       except as required for reasonable and customary use in describing the
141 |       origin of the Work and reproducing the content of the NOTICE file.
142 | 
143 |    7. Disclaimer of Warranty. Unless required by applicable law or
144 |       agreed to in writing, Licensor provides the Work (and each
145 |       Contributor provides its Contributions) on an "AS IS" BASIS,
146 |       WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
147 |       implied, including, without limitation, any warranties or conditions
148 |       of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
149 |       PARTICULAR PURPOSE. You are solely responsible for determining the
150 |       appropriateness of using or redistributing the Work and assume any
151 |       risks associated with Your exercise of permissions under this License.
152 | 
153 |    8. Limitation of Liability. In no event and under no legal theory,
154 |       whether in tort (including negligence), contract, or otherwise,
155 |       unless required by applicable law (such as deliberate and grossly
156 |       negligent acts) or agreed to in writing, shall any Contributor be
157 |       liable to You for damages, including any direct, indirect, special,
158 |       incidental, or consequential damages of any character arising as a
159 |       result of this License or out of the use or inability to use the
160 |       Work (including but not limited to damages for loss of goodwill,
161 |       work stoppage, computer failure or malfunction, or any and all
162 |       other commercial damages or losses), even if such Contributor
163 |       has been advised of the possibility of such damages.
164 | 
165 |    9. Accepting Warranty or Additional Liability. While redistributing
166 |       the Work or Derivative Works thereof, You may choose to offer,
167 |       and charge a fee for, acceptance of support, warranty, indemnity,
168 |       or other liability obligations and/or rights consistent with this
169 |       License. However, in accepting such obligations, You may act only
170 |       on Your own behalf and on Your sole responsibility, not on behalf
171 |       of any other Contributor, and only if You agree to indemnify,
172 |       defend, and hold each Contributor harmless for any liability
173 |       incurred by, or claims asserted against, such Contributor by reason
174 |       of your accepting any such warranty or additional liability.
175 | 
176 |    END OF TERMS AND CONDITIONS
177 | 
178 |    APPENDIX: How to apply the Apache License to your work.
179 | 
180 |       To apply the Apache License to your work, attach the following
181 |       boilerplate notice, with the fields enclosed by brackets "[]"
182 |       replaced with your own identifying information. (Don't include
183 |       the brackets!)  The text should be enclosed in the appropriate
184 |       comment syntax for the file format. We also recommend that a
185 |       file or class name and description of purpose be included on the
186 |       same "printed page" as the copyright notice for easier
187 |       identification within third-party archives.
188 | 
189 |    Copyright [yyyy] [name of copyright owner]
190 | 
191 |    Licensed under the Apache License, Version 2.0 (the "License");
192 |    you may not use this file except in compliance with the License.
193 |    You may obtain a copy of the License at
194 | 
195 |        http://www.apache.org/licenses/LICENSE-2.0
196 | 
197 |    Unless required by applicable law or agreed to in writing, software
198 |    distributed under the License is distributed on an "AS IS" BASIS,
199 |    WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
200 |    See the License for the specific language governing permissions and
201 |    limitations under the License.
202 | 


--------------------------------------------------------------------------------
/README.md:
--------------------------------------------------------------------------------
  1 | # 《深度学习入门：基于PyTorch和TensorFlow的理论与实战》
  2 | 
  3 | **作者：红色石头**
  4 | 
  5 | **出版社：清华大学出版社**
  6 | 
  7 | 19世纪70 年代，电力的发明和应用掀起了第二次工业化高潮，从此改变了人们的生活方式，大大提高了人类的科技水平。现如今，深度学习（Deep Learning）技术也正在发挥同样的作用。纵观这近几年，深度学习发展非常迅速，发展势头一直高歌猛进。毫无疑问，深度学习技术正在影响着我们的日常生活和行为方式。
  8 | 
  9 | 
 10 | ## 深度学习怎么学
 11 | 
 12 | 深度学习怎么学？事实上，很多初学者常常会误入两大误区：一是找不到一本真正适合自己的教材或书本来学习，陷入到海量资源中手足无措；二是受制于数学理论知识，自认为数学基础不好而影响学习的主观性和积极性。
 13 | 
 14 | 这两大误区很容易让初学者陷入到一种迷茫的状态。所以，第一步就是要放弃海量资源，选择一份真正适合自己的资料，好好研读下去！第二步就是重视实践。深度学习涉及的理论知识很多，有些人可能基础不是特别扎实，就想着从最底层的知识开始学起，概率论、线性代数、凸优化理论等。但是这样做比较耗时间，而且容易打消学习的积极性。啃书本和推导公式相对来说是比较枯燥的，远不如自己搭建一个简单的神经网络模型更能激发学习积极性。当然，基础理论知识非常重要！只是说，在入门的时候，最好先从顶层框架上有个系统的认识，然后再从实践到理论，有的放矢的查缺补漏深度学习知识点。从宏观到微观，从整体到细节，更有利于快速入门！ 
 15 | 
 16 | ## 为什么写这本书
 17 | 
 18 | 在学习深度学习的几年时间里，我学过一些国内外优秀的深度学习公开课程，口碑都很好；我也看过不少大牛老师写的高质量书籍，收获颇丰；我也在学习的过程中走过一些弯路，遇到一些坑，这些也是宝贵的经验。
 19 | 
 20 | 我个人觉得，任何前沿技术，如深度学习，扎实的基础知识非常重要。而最好的基础知识的获取方式还是教材和书本。但是，反观现在一些深度学习方面的书籍，或多或少存在一些问题：
 21 | 
 22 | 1、数学理论太多，公式多，起点高，对初学者不友善，容易削弱入门学习的积极性。
 23 | 
 24 | 2、只讲深度学习框架，教你如何调包、调用库函数，不讲深度学习理论知识。容易造成一知半解，沦为“调包侠”。
 25 | 
 26 | 3、理论与实战的脱节，过于侧重理论或者过于侧重实战，二者之间没有很好的融合。
 27 | 
 28 | 基于以上这些问题，我认为写一本真正适合深度学习初学者的入门书籍非常必要。这样的书籍不仅要兼顾理论和实战，还应该将重难点知识通俗化讲解、全面细致。难度有阶梯性，照顾不同水平的读者。这样的书籍才能最大程度地让读者受益。
 29 | 
 30 | 基于这样的考量，《深度学习入门：基于PyTorch与TensorFlow的理论与实战》跟大家见面了。
 31 | 
 32 | ## 本书特色
 33 | 
 34 | 我刚刚也说了，对于初学者而言，一本好的深度学习书籍，宗旨就是让读者能够轻轻松松地掌握知识、触类旁通。本书作为一本深度学习的入门书籍，对初学者是非常友好的。本书的内容来自于我多年的知识积累和技术沉淀，也是我的一份深度学习经验总结。
 35 | 
 36 | 首先，这本书包含了Python的基本介绍。Python作为人工智能的首选语言，其重要性不言而喻。Python入门非常简单，本书将会对深度学习中所需的基本Python语法知识进行简明扼要的提炼和概括。如果有的读者之前没有接触过Python，那么本书将轻松带你入门。
 37 | 
 38 | 其次，这本书介绍了如今主流的深度学习框架PyTorch和TensorFlow。通过本书，读者可以对这两个框架的基本语法和基础知识有一个系统的学习，夯实基础。如果你之前对PyTorch和TensorFlow不了解也没有关系，这本书也可以是这两个框架的知识学习手册。
 39 | 
 40 | 然而，最重要的，这是一本关于深度学习的入门教程。我在编写该书的时候，从小白的视角出发，结合我多年的知识和经验，尽量将深度学习、神经网络的理论知识用通俗易懂的语言描绘出来。这本书能让读者真正了解、熟悉神经网络的结构和优化方法，也能帮助读者梳理一些容易被忽视的技术细节。例如最简单的梯度下降算法，它的公式来源和理论支持是什么？本书都会有详细的解释。
 41 | 
 42 | 值得注意的是，我一贯的坚持是将复杂的理论简单化，这本书会将理论以平易化的语言描述清楚，但不会深陷于数学公式之中。这本书面向的深度学习的入门者和初学者，不会涉及太多太复杂的理论知识。因为入门深度学习，前期整体上的感性认识尤为重要。轻松入门，往往是比较正确的学习路线。我在编写该书的时候，也一直在把握这个分寸和尺度。如果想要学习更深层次、更高级的深度学习知识，读者可以查阅更多的书籍、会议论文、前沿技术等。
 43 | 除此之外，深度学习更重要的是代码实践，这也是本书一直秉承的一个重点。这本书的另一个优势就是不仅讲理论知识，也配备了完整的实战项目和代码。从简单的逻辑回归，到浅层神经网络、深层神经网络，再到CNN、RNN，都会通过一个实际项目从零搭建神经网络，或者使用PyTorch、TensorFlow来构建更复杂的例如CNN、RNN模型解决问题。
 44 | 
 45 | 本书的所有代码，我都开源放在了GitHub上，地址如下：
 46 | 
 47 | [https://github.com/RedstoneWill/dl-from-scratch](https://github.com/RedstoneWill/dl-from-scratch)
 48 | 
 49 | ## 面向的读者
 50 | 
 51 | 这是一本深度学习的入门书籍，也是一本关于Python、PyTorch、TensorFlow的工具手册；这是一本深度学习的理论书籍，也是一本教你如何编写代码构建神经网络的实战手册。我希望这本书能够帮助更多想要入门深度学习的爱好者，能够帮助读者扫清学习过程中的障碍，再上新台阶。
 52 | 
 53 | 本书面向的读者包括：深度学习初学者； 对深度学习感兴趣的在校大学生； 有意向转行AI领域的IT从业人员。当然，这本书也是不错的深度学习工具手册，里面不仅有理论知识，也有示例代码。
 54 | 
 55 | 值得一提的是，如果你已经有很高的深度学习水平了，那么可能本书不太适合你，你应该更关注深度学习的前沿理论和论文。
 56 | 
 57 | ## 关于作者
 58 | 
 59 | 红色石头，北京大学硕士。专注AI领域多年，爱好写作，累计写过的AI领域的原创文章超过150篇，累计读者达20W，文章阅读量超100W。写作风格是擅长以通俗化的语言来解释机器学习、深度学习的算法理论和技术细节。
 60 | 
 61 | 个人网站：[www.redstonewill.com](www.redstonewill.com)
 62 | 
 63 | 创办了AI技术领域的微信公众号：AI 有道（ID：redstonewill），欢迎读者关注，方便第一时间获取机器学习、深度学习等有价值的干货分享和信息资源。
 64 | 
 65 | ![](./微信公众号：AI有道.jpg)
 66 | 
 67 | ## 书籍目录
 68 | 
 69 | **第1章	深度学习基础**	1
 70 | 
 71 | - 1.1 深度学习概述	1
 72 | 
 73 |     - 1.1.1 什么是深度学习	1
 74 | 
 75 |     - 1.1.2 深度学习的应用场景	3
 76 | 
 77 |     - 1.1.3 深度学习的发展动力	4
 78 | 
 79 |     - 1.1.4 深度学习的未来	5
 80 | 
 81 | - 1.2 Python入门	6
 82 | 
 83 |     - 1.2.1 Python简介	6
 84 | 
 85 |     - 1.2.2 Python的安装	7
 86 | 
 87 |     - 1.2.3 Python基础知识	8
 88 | 
 89 |     - 1.2.4 NumPy矩阵运算	15
 90 | 
 91 |     - 1.2.5 Matplitlib绘图	20
 92 | 
 93 | - 1.3 Anaconda与Jupyter Notebook	24
 94 | 
 95 |     - 1.3.1 Anaconda	25
 96 | 
 97 |     - 1.3.2 Jupyter Notebook	27
 98 | 
 99 | **第2章 PyTorch**	34
100 | 
101 | - 2.1 PyTorch简介	34
102 | 
103 |     - 2.1.1 什么是PyTorch	34
104 | 
105 |     - 2.1.2 为什么使用PyTorch	35
106 | 
107 | - 2.2 PyTorch安装	36
108 | 
109 | - 2.3 张量Tensor	39
110 | 
111 |     - 2.3.1 创建Tensor	39
112 | 
113 |     - 2.3.2 Tensor的数学运算	40
114 | 
115 |     - 2.3.3 Tensor与NumPy	41
116 | 
117 |     - 2.3.4 CUDA Tensor	42
118 | 
119 | - 2.4 自动求导 autograd	43
120 | 
121 |     - 2.4.1 返回值是标量	43
122 | 
123 |     - 2.4.2 返回值是张量	44
124 | 
125 |     - 2.4.3 禁止自动求导	45
126 | 
127 | - 2.5 神经网络包nn和优化器optim	45
128 | 
129 |     - 2.5.1 torch.nn	45
130 | 
131 |     - 2.5.2 torch.optim	46
132 | 
133 | - 2.6 PyTorch线性回归	47
134 | 
135 |     - 2.6.1 线性回归基本原理	48
136 | 
137 |     - 2.6.2 PyTorch实现	49
138 | 
139 | **第3章 TensorFlow**	53
140 | 
141 | - 3.1 TensorFlow简介	53
142 | 
143 |     - 3.1.1 什么是TensorFlow	53
144 | 
145 |     - 3.1.2 为什么使用TensorFLow	54
146 | 
147 | - 3.2 TensorFlow安装	54
148 | 
149 | - 3.3 张量Tensor	56
150 | 
151 |     - 3.3.1 创建Tensor	56
152 | 
153 |     - 3.3.2 Tensor的数学运算	57
154 | 
155 | - 3.4 数据流图	58
156 | 
157 | - 3.5 会话Session	60
158 | 
159 | - 3.6 TensorFlow线性回归	62
160 | 
161 | - 3.7 TensorBoard	66
162 | 
163 |     - 3.7.1 TensorBoard代码	66
164 | 
165 |     - 3.7.2 TensorBoard显示	67
166 | 
167 | **第4章 神经网络基础**	71
168 | 
169 | - 4.1 感知机	71
170 | 
171 |     - 4.1.1 感知机模型	71
172 | 
173 |     - 4.1.2 感知机与逻辑电路	72
174 | 
175 | - 4.2 多层感知机	77
176 | 
177 |     - 4.2.1 感知机的局限性	78
178 | 
179 |     - 4.2.2 多层感知机实现异或	79
180 | 
181 | - 4.3 逻辑回归	81
182 | 
183 |     - 4.3.1 基本原理	82
184 | 
185 |     - 4.3.2 损失函数	84
186 | 
187 |     - 4.3.3 梯度下降	87
188 | 
189 |     - 4.3.4逻辑回归的Python实现	92
190 | 
191 | **第5章 神经网络**	98
192 | 
193 | - 5.1 基本结构	98
194 | 
195 | - 5.2 前向传播	100
196 | 
197 | - 5.3 激活函数	101
198 | 
199 | - 5.4 反向传播	106
200 | 
201 | - 5.5 更新参数	108
202 | 
203 | - 5.6 初始化	108
204 | 
205 | - 5.7 神经网络的Python实现	109
206 | 
207 |     - 5.7.1 准备数据	109
208 | 
209 |     - 5.7.2 参数初始化	110
210 | 
211 |     - 5.7.3 前向传播	111
212 | 
213 |     - 5.7.4 交叉熵损失	112
214 | 
215 |     - 5.7.5 反向传播	113
216 | 
217 |     - 5.7.6 更新参数	114
218 | 
219 |     - 5.7.7 构建整个神经网络模型	115
220 | 
221 |     - 5.7.8 训练	116
222 | 
223 |     - 5.7.9 预测	116
224 | 
225 | **第6章	深层神经网络**	119
226 | 
227 | - 6.1 神经网络为什么要深	119
228 | 
229 | - 6.2 符号标记	121
230 | 
231 | - 6.3 前向传播与反向传播	122
232 | 
233 | - 6.4 多分类Softmax	125
234 | 
235 |     - 6.4.1 Softmax基本原理	126
236 | 
237 |     - 6.4.2 Softmax损失函数	127
238 | 
239 |     - 6.4.3 Softmax求导	128
240 | 
241 | - 6.5 深层神经网络的Python实现	130
242 | 
243 |     - 6.5.1 准备数据	130
244 | 
245 |     - 6.5.2 参数初始化	133
246 | 
247 |     - 6.5.3 前向传播	134
248 | 
249 |     - 6.5.4 交叉熵损失	137
250 | 
251 |     - 6.5.5 反向传播	137
252 | 
253 |     - 6.5.6 更新参数	140
254 | 
255 |     - 6.5.7 构建整个神经网络	141
256 | 
257 |     - 6.5.8 训练与预测	143
258 | 
259 | **第7章	优化神经网络**	146
260 | 
261 | - 7.1 正则化	146
262 | 
263 |     - 7.1.1 什么是过拟合	146
264 | 
265 |     - 7.1.2 L1和L2正则化	149
266 | 
267 |     - 7.1.3 Dropout正则化	153
268 | 
269 |     - 7.1.4 其它正则化技巧	157
270 | 
271 | - 7.2 梯度优化	159
272 | 
273 |     - 7.2.1 BGD、SGD、MBGD	159
274 | 
275 |     - 7.2.2 Momentum GD	163
276 | 
277 |     - 7.2.3 Nesterov Momentum	165
278 | 
279 |     - 7.2.4 AdaGrad	166
280 | 
281 |     - 7.2.5 RMSprop	167
282 | 
283 |     - 7.2.6 Adam	168
284 | 
285 |     - 7.2.7 Learning Rate Decay	169
286 | 
287 | - 7.3 网络初始化与超参数调试	170
288 | 
289 |     - 7.3.1 输入标准化	171
290 | 
291 |     - 7.3.2 权重参数初始化	173
292 | 
293 |     - 7.3.3 批归一化	176
294 | 
295 |     - 7.3.4 超参数调试	179
296 | 
297 | - 7.4 模型评估与调试	183
298 | 
299 |     - 7.4.1 模型评估	183
300 | 
301 |     - 7.4.2 训练/验证/测试集	184
302 | 
303 |     - 7.4.3 偏差与方差	187
304 | 
305 |     - 7.4.4 错误分析	187
306 | 
307 | **第8章	卷积神经网络**	192
308 | 
309 | - 8.1 为什么选择CNN	192
310 | 
311 | - 8.2 CNN基本结构	193
312 | 
313 | - 8.3 卷积层	194
314 | 
315 |     - 8.3.1 卷积	194
316 | 
317 |     - 8.3.2 边缘检测	196
318 | 
319 |     - 8.3.3 填充Padding	198
320 | 
321 |     - 8.3.4 步幅Stride	199
322 | 
323 |     - 8.3.5 CNN卷积	200
324 | 
325 |     - 8.3.6 卷积层的作用	205
326 | 
327 | - 8.4 池化层	205
328 | 
329 | - 8.5 全连接层	208
330 | 
331 | - 8.6 CNN模型	210
332 | 
333 | - 8.7 典型的CNN模型	213
334 | 
335 |     - 8.7.1 LeNet-5	213
336 | 
337 |     - 8.7.2 AlexNet	214
338 | 
339 | - 8.8 CNN的PyTorch实现	215
340 | 
341 |     - 8.8.1 准备数据	215
342 | 
343 |     - 8.8.2 定义CNN模型	219
344 | 
345 |     - 8.8.3 损失函数与梯度优化	221
346 | 
347 |     - 8.8.4 训练模型	221
348 | 
349 |     - 8.8.5 测试模型	223
350 | 
351 | - 8.9 CNN的TensorFlow实现	224
352 | 
353 |     - 8.9.1 准备数据	224
354 | 
355 |     - 8.9.2 定义CNN模型	225
356 | 
357 |     - 8.9.3 损失函数与优化算法	227
358 | 
359 |     - 8.9.4 训练并测试	228
360 | 
361 | **第9章	循环神经网络**	229
362 | 
363 | - 9.1 为什么选择RNN	229
364 | 
365 | - 9.2 RNN基本结构	230
366 | 
367 | - 9.3 模型参数	232
368 | 
369 | - 9.4 梯度消失	234
370 | 
371 | - 9.5 GRU	234
372 | 
373 | - 9.6 LSTM	236
374 | 
375 | - 9.7 多种RNN模型	237
376 | 
377 | - 9.8 RNN的PyTorch实现	241
378 | 
379 |     - 9.8.1 准备数据	242
380 | 
381 |     - 9.8.2 定义RNN模型	244
382 | 
383 |     - 9.8.3 损失函数与梯度优化	246
384 | 
385 |     - 9.8.4 训练模型	246
386 | 
387 |     - 9.8.5 测试模型	247
388 | 
389 | - 9.9 RNN的TensorFlow实现	248
390 | 
391 |     - 9.9.1 准备数据	249
392 | 
393 |     - 9.9.2 定义RNN模型	249
394 | 
395 |     - 9.9.3 损失函数与优化算法	250
396 | 
397 |     - 9.9.4 训练并测试	251
398 | 
399 | **后记**	252
400 | 
401 | 
402 | 
403 | ## 源代码目录
404 | 
405 | - ## [01_the_foundation_of_deep_learning.ipynb](https://github.com/RedstoneWill/dl-from-scratch/blob/master/01_the_foundation_of_deep_learning.ipynb)
406 | 
407 | - ## [02_pytorch_tutorial.ipynb](https://github.com/RedstoneWill/dl-from-scratch/blob/master/02_pytorch_tutorial.ipynb)
408 | 
409 | - ## [03_tensorflow_tutorial.ipynb](https://github.com/RedstoneWill/dl-from-scratch/blob/master/03_tensorflow_tutorial.ipynb)
410 | 
411 | - ## [04_neural_network_foundation.ipynb](https://github.com/RedstoneWill/dl-from-scratch/blob/master/04_neural_network_foundation.ipynb)
412 | 
413 | - ## [05_neural_network.ipynb](https://github.com/RedstoneWill/dl-from-scratch/blob/master/05_neural_network.ipynb)
414 | 
415 | - ## [06_deep_neural_network.ipynb](https://github.com/RedstoneWill/dl-from-scratch/blob/master/06_deep_neural_network.ipynb)
416 | 
417 | - ## [08_cnn_pytorch.ipynb](https://github.com/RedstoneWill/dl-from-scratch/blob/master/08_cnn_pytorch.ipynb)
418 | 
419 | - ## [08_cnn_tensorflow.ipynb](https://github.com/RedstoneWill/dl-from-scratch/blob/master/08_cnn_tensorflow.ipynb)
420 | 
421 | - ## [09_rnn_pytorch.ipynb](https://github.com/RedstoneWill/dl-from-scratch/blob/master/09_rnn_pytorch.ipynb)
422 | 
423 | - ## [09_rnn_tensorflow.ipynb](https://github.com/RedstoneWill/dl-from-scratch/blob/master/09_rnn_tensorflow.ipynb)


--------------------------------------------------------------------------------
/datasets/ch01/cat.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/RedstoneWill/dl-from-scratch/34c87663e5ec30384129c528031c3e9085877418/datasets/ch01/cat.jpg


--------------------------------------------------------------------------------
/datasets/ch03/3-3.png:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/RedstoneWill/dl-from-scratch/34c87663e5ec30384129c528031c3e9085877418/datasets/ch03/3-3.png


--------------------------------------------------------------------------------
/datasets/ch06/test.rar:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/RedstoneWill/dl-from-scratch/34c87663e5ec30384129c528031c3e9085877418/datasets/ch06/test.rar


--------------------------------------------------------------------------------
/datasets/ch06/train.rar:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/RedstoneWill/dl-from-scratch/34c87663e5ec30384129c528031c3e9085877418/datasets/ch06/train.rar


--------------------------------------------------------------------------------
/datasets/ch08/pytorch/MNIST/processed/test.pt:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/RedstoneWill/dl-from-scratch/34c87663e5ec30384129c528031c3e9085877418/datasets/ch08/pytorch/MNIST/processed/test.pt


--------------------------------------------------------------------------------
/datasets/ch08/pytorch/MNIST/processed/training.pt:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/RedstoneWill/dl-from-scratch/34c87663e5ec30384129c528031c3e9085877418/datasets/ch08/pytorch/MNIST/processed/training.pt


--------------------------------------------------------------------------------
/datasets/ch08/pytorch/MNIST/raw/t10k-images-idx3-ubyte:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/RedstoneWill/dl-from-scratch/34c87663e5ec30384129c528031c3e9085877418/datasets/ch08/pytorch/MNIST/raw/t10k-images-idx3-ubyte


--------------------------------------------------------------------------------
/datasets/ch08/pytorch/MNIST/raw/t10k-labels-idx1-ubyte:
--------------------------------------------------------------------------------
1 | 
'

			
		






			
	
	

			
				



		




	


	



				

	
	
		


		

		
	

			


	





	



	
	
	
	
		

			
	

	
			
		

	

	
				
			
	
				


	
	
	
		



	

	




		


	

	


	



	




			
	
	
	

	


	
	

	




	
		
	

	






		
		
	
			
			
	


		

	
	




			


	
	
	
	

	
			

		

	




		
	

				
		

		
				

					


		
		
	
			
		
	
	
	
			

	
	


		

		


	
		




	
		
			
		






		


		
	

	

			




			

	

		

	


	


	
		
			

	
	



	
				
	
	

	
		
	
	

	



		
		

		

		
	




	

		
	

	

	
	
		
			
					

	

	
	
		
	
			

	

			

	

	
	
		



	


			

	
			


	
	




					

		

	
	
	
			
				


		
		
	
	

	
			



	
	
		

	






	
	
			
	
	






		








		

	


	

	
					


	







		
	
			
									
	
	


		
	



	







		


	
	
				



			



	


		

					
	

		
		
										




	
		

				
	
		
			

	


		

	



		




	
				


				
	


			



	


			



	
	
	
	
				
		


	
				
					


			
			

	
	
				
			
					
		
				
			
	



	
					

	

		

	




			
	
	


				
	

				


				





		

	
	

			
					

	





	
		

	
		




		


			
	

				
	

			
		
	
	
		
		




		


	
			

	
	



	
		

			


			
	
	
		
	




		
	

			
		
	
	
		

	
	





	
	


	
	

	
	


	
				


	
	
		

	
	



		
	

				
	


			


	
	
		
		

	
	

	
			




	
	
	

	
	
	
		
		


	
	


			
		


	
	
	



	
	
	

	

				
	
	
			

		


	
		


		

	





			
	

			
		
	
	
	




		
	
			
		




				
	

		


	
	
	
	
			

	


	

	

	
	



	
	



		
		
	
	
	
	



	
		
	

			
		
	
	
	
	
	

					

	


			

	
	
	
	
				
			
	




	
	

	

	
	

				

		

	
	
	

	
	

	

	
		
				






	


		
	
	
					
	
	
	
	
	

			

	
		

	
	
	

				


		
			




	
	
		

	
		

	

	
		
	

	
			
	
	
		






	
	
	
		



			

		

		

	
	
	
	


	

		

	
	
		


			
			
	
	



	
	
	
	
	
	


		
		


	

			
	

	

		
	
	


	
	
		


			
	
	
	




	



				

	

		


	
	
	
	
		

	
	



		
	

				
	
	
				



	

				
	
	
	
		
				

	



	
	



	
		



		
	
	
	
		





	
		
		

	
	
				


	
				

	

	
		
	

	





		
	
	

	
	
	
			





				

	
	
	

		
	
		
	
			




	
	
	
	




	
	
	

	
	
	
	
	
		
		

		



	

		

	



	

		

	

	
	
		




	

	



--------------------------------------------------------------------------------
/datasets/ch08/pytorch/MNIST/raw/train-images-idx3-ubyte:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/RedstoneWill/dl-from-scratch/34c87663e5ec30384129c528031c3e9085877418/datasets/ch08/pytorch/MNIST/raw/train-images-idx3-ubyte


--------------------------------------------------------------------------------
/datasets/ch08/pytorch/MNIST/raw/train-labels-idx1-ubyte:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/RedstoneWill/dl-from-scratch/34c87663e5ec30384129c528031c3e9085877418/datasets/ch08/pytorch/MNIST/raw/train-labels-idx1-ubyte


--------------------------------------------------------------------------------
/datasets/ch08/tensorflow/MNIST/t10k-images-idx3-ubyte.gz:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/RedstoneWill/dl-from-scratch/34c87663e5ec30384129c528031c3e9085877418/datasets/ch08/tensorflow/MNIST/t10k-images-idx3-ubyte.gz


--------------------------------------------------------------------------------
/datasets/ch08/tensorflow/MNIST/t10k-labels-idx1-ubyte.gz:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/RedstoneWill/dl-from-scratch/34c87663e5ec30384129c528031c3e9085877418/datasets/ch08/tensorflow/MNIST/t10k-labels-idx1-ubyte.gz


--------------------------------------------------------------------------------
/datasets/ch08/tensorflow/MNIST/train-images-idx3-ubyte.gz:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/RedstoneWill/dl-from-scratch/34c87663e5ec30384129c528031c3e9085877418/datasets/ch08/tensorflow/MNIST/train-images-idx3-ubyte.gz


--------------------------------------------------------------------------------
/datasets/ch08/tensorflow/MNIST/train-labels-idx1-ubyte.gz:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/RedstoneWill/dl-from-scratch/34c87663e5ec30384129c528031c3e9085877418/datasets/ch08/tensorflow/MNIST/train-labels-idx1-ubyte.gz


--------------------------------------------------------------------------------
/datasets/readme.md:
--------------------------------------------------------------------------------
1 | # Datasets


--------------------------------------------------------------------------------
/tf_logs/ch03/run-20190704075350/events.out.tfevents.1562226830.AB-201810292038:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/RedstoneWill/dl-from-scratch/34c87663e5ec30384129c528031c3e9085877418/tf_logs/ch03/run-20190704075350/events.out.tfevents.1562226830.AB-201810292038


--------------------------------------------------------------------------------
/tf_logs/readme.md:
--------------------------------------------------------------------------------
1 | # TensorBoard 


--------------------------------------------------------------------------------
/微信公众号：AI有道.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/RedstoneWill/dl-from-scratch/34c87663e5ec30384129c528031c3e9085877418/微信公众号：AI有道.jpg


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