![\"Open](\"https://colab.research.google.com/assets/colab-badge.svg\"){kind=icon}
"
26 | ]
27 | },
28 | {
29 | "cell_type": "markdown",
30 | "metadata": {
31 | "id": "U91hkLpi-SBV",
32 | "colab_type": "text"
33 | },
34 | "source": [
35 | "#Imports"
36 | ]
37 | },
38 | {
39 | "cell_type": "code",
40 | "metadata": {
41 | "id": "8cjFGNq0P_re",
42 | "colab_type": "code",
43 | "colab": {}
44 | },
45 | "source": [
46 | "import numpy as np\n",
47 | "import pandas as pd\n",
48 | "import warnings\n",
49 | "warnings.filterwarnings('ignore')\n",
50 | "from sklearn.metrics import mean_squared_error\n",
51 | "\n",
52 | "from sklearn.preprocessing import StandardScaler, PowerTransformer\n",
53 | "from sklearn.compose import TransformedTargetRegressor\n",
54 | "from sklearn.pipeline import FeatureUnion, Pipeline, make_pipeline\n",
55 | "from sklearn.base import BaseEstimator, TransformerMixin\n",
56 | "\n",
57 | "from sklearn.linear_model import LinearRegression"
58 | ],
59 | "execution_count": 0,
60 | "outputs": []
61 | },
62 | {
63 | "cell_type": "markdown",
64 | "metadata": {
65 | "id": "aZ7KzRqK-KHL",
66 | "colab_type": "text"
67 | },
68 | "source": [
69 | "#Pipeline Experiment"
70 | ]
71 | },
72 | {
73 | "cell_type": "code",
74 | "metadata": {
75 | "id": "D13TnX_2_mLi",
76 | "colab_type": "code",
77 | "colab": {}
78 | },
79 | "source": [
80 | "df = pd.DataFrame(columns=['X1', 'X2', 'y'], data=[\n",
81 | " [1,16,9],\n",
82 | " [4,36,16],\n",
83 | " [1,16,9],\n",
84 | " [2,9,8],\n",
85 | " [3,36,15],\n",
86 | " [2,49,16],\n",
87 | " [4,25,14],\n",
88 | " [5,36,17]\n",
89 | "])\n",
90 | "\n",
91 | "### y = X1 + 2 * sqrt(X2)\n",
92 | "\n",
93 | "train = df.iloc[:6]\n",
94 | "test = df.iloc[6:]\n",
95 | "\n",
96 | "train_X = train.drop('y', axis=1)\n",
97 | "train_y = train.y\n",
98 | "\n",
99 | "test_X = test.drop('y', axis=1)\n",
100 | "test_y = test.y"
101 | ],
102 | "execution_count": 0,
103 | "outputs": []
104 | },
105 | {
106 | "cell_type": "code",
107 | "metadata": {
108 | "id": "Wd_emlUpBbig",
109 | "colab_type": "code",
110 | "colab": {
111 | "base_uri": "https://localhost:8080/",
112 | "height": 90
113 | },
114 | "outputId": "d33d7919-f565-4b51-d6b5-cd0dae60604d"
115 | },
116 | "source": [
117 | "# let's see if linear regression is able to predict this properly\n",
118 | "m1 = LinearRegression()\n",
119 | "fit1 = m1.fit(train_X, train_y)\n",
120 | "preds = fit1.predict(test_X)\n",
121 | "print(f\"\\n{preds}\")\n",
122 | "print(f\"RMSE: {np.sqrt(mean_squared_error(test_y, preds))}\\n\")"
123 | ],
124 | "execution_count": 29,
125 | "outputs": [
126 | {
127 | "output_type": "stream",
128 | "text": [
129 | "\n",
130 | "[13.72113586 16.93334467]\n",
131 | "RMSE: 0.20274138822160603\n",
132 | "\n"
133 | ],
134 | "name": "stdout"
135 | }
136 | ]
137 | },
138 | {
139 | "cell_type": "code",
140 | "metadata": {
141 | "id": "zQayYrzmDWE7",
142 | "colab_type": "code",
143 | "colab": {
144 | "base_uri": "https://localhost:8080/",
145 | "height": 145
146 | },
147 | "outputId": "b9b3321c-e5b9-4777-d673-8bb07def11d3"
148 | },
149 | "source": [
150 | "# what if we square-root X2 and multiply by 2?\n",
151 | "train_X.X2 = 2 * np.sqrt(train_X.X2)\n",
152 | "test_X.X2 = 2 * np.sqrt(test_X.X2)\n",
153 | "print(test_X)\n",
154 | "m2 = LinearRegression()\n",
155 | "fit2 = m2.fit(train_X, train_y)\n",
156 | "preds = fit2.predict(test_X)\n",
157 | "print(f\"\\n{preds}\")\n",
158 | "print(f\"RMSE: {np.sqrt(mean_squared_error(test_y, preds))}\\n\")"
159 | ],
160 | "execution_count": 30,
161 | "outputs": [
162 | {
163 | "output_type": "stream",
164 | "text": [
165 | " X1 X2\n",
166 | "6 4 10.0\n",
167 | "7 5 12.0\n",
168 | "\n",
169 | "[14. 17.]\n",
170 | "RMSE: 1.2560739669470201e-15\n",
171 | "\n"
172 | ],
173 | "name": "stdout"
174 | }
175 | ]
176 | },
177 | {
178 | "cell_type": "code",
179 | "metadata": {
180 | "id": "xMrbwYDDNFKu",
181 | "colab_type": "code",
182 | "colab": {}
183 | },
184 | "source": [
185 | "# a perfect prediction, because the data after transformation, fits a perfect linear trend.\n",
186 | "# let's restore the data back to original, and do this via custom transformers using pipeline.\n",
187 | "train = df.iloc[:6]\n",
188 | "test = df.iloc[6:]\n",
189 | "\n",
190 | "train_X = train.drop('y', axis=1)\n",
191 | "train_y = train.y\n",
192 | "\n",
193 | "test_X = test.drop('y', axis=1)\n",
194 | "test_y = test.y"
195 | ],
196 | "execution_count": 0,
197 | "outputs": []
198 | },
199 | {
200 | "cell_type": "code",
201 | "metadata": {
202 | "id": "SD08IkZP-Lam",
203 | "colab_type": "code",
204 | "colab": {}
205 | },
206 | "source": [
207 | "# references: \n",
208 | "# https://towardsdatascience.com/custom-transformers-and-ml-data-pipelines-with-python-20ea2a7adb65\n",
209 | "# https://machinelearningmastery.com/how-to-transform-target-variables-for-regression-with-scikit-learn/\n",
210 | "# http://zacstewart.com/2014/08/05/pipelines-of-featureunions-of-pipelines.html\n",
211 | "# https://stackoverflow.com/questions/43308042/transformer-initialize-twice-in-pipeline\n",
212 | "\n",
213 | "class ExperimentalTransformer(BaseEstimator, TransformerMixin):\n",
214 | " def __init__(self):\n",
215 | " print('\\n>>>>>>>init() called.\\n')\n",
216 | "\n",
217 | " def fit(self, X, y = None):\n",
218 | " print('\\n>>>>>>>fit() called.\\n')\n",
219 | " return self\n",
220 | "\n",
221 | " def transform(self, X, y = None):\n",
222 | " print('\\n>>>>>>>transform() called.\\n')\n",
223 | " X_ = X.copy() # creating a copy to avoid changes to original dataset\n",
224 | " X_.X2 = 2 * np.sqrt(X_.X2)\n",
225 | " return X_"
226 | ],
227 | "execution_count": 0,
228 | "outputs": []
229 | },
230 | {
231 | "cell_type": "code",
232 | "metadata": {
233 | "id": "spUiscMUPY83",
234 | "colab_type": "code",
235 | "colab": {
236 | "base_uri": "https://localhost:8080/",
237 | "height": 145
238 | },
239 | "outputId": "961ad4f6-5143-442a-8a19-bc986910479d"
240 | },
241 | "source": [
242 | "# without input transformation - to validate that we get the same results as before\n",
243 | "print(\"create pipeline 1\")\n",
244 | "pipe1 = Pipeline(steps=[\n",
245 | " ('linear_model', LinearRegression())\n",
246 | "])\n",
247 | "print(\"fit pipeline 1\")\n",
248 | "pipe1.fit(train_X, train_y)\n",
249 | "print(\"predict via pipeline 1\")\n",
250 | "preds1 = pipe1.predict(test_X)\n",
251 | "print(f\"\\n{preds1}\") # should be [13.72113586 16.93334467]\n",
252 | "print(f\"RMSE: {np.sqrt(mean_squared_error(test_y, preds1))}\\n\") "
253 | ],
254 | "execution_count": 33,
255 | "outputs": [
256 | {
257 | "output_type": "stream",
258 | "text": [
259 | "create pipeline 1\n",
260 | "fit pipeline 1\n",
261 | "predict via pipeline 1\n",
262 | "\n",
263 | "[13.72113586 16.93334467]\n",
264 | "RMSE: 0.20274138822160603\n",
265 | "\n"
266 | ],
267 | "name": "stdout"
268 | }
269 | ]
270 | },
271 | {
272 | "cell_type": "code",
273 | "metadata": {
274 | "id": "OFTnGVww_kIi",
275 | "colab_type": "code",
276 | "colab": {
277 | "base_uri": "https://localhost:8080/",
278 | "height": 363
279 | },
280 | "outputId": "a86df4fa-fa17-4a43-f87a-07167fff174c"
281 | },
282 | "source": [
283 | "# with input transformation\n",
284 | "print(\"create pipeline 2\")\n",
285 | "pipe2 = Pipeline(steps=[\n",
286 | " ('experimental_trans', ExperimentalTransformer()), # this will trigger a call to __init__\n",
287 | " ('linear_model', LinearRegression())\n",
288 | "])\n",
289 | "\n",
290 | "# an alternate, shorter syntax to do the above, without naming each step, is:\n",
291 | "#pipe2 = make_pipeline(ExperimentalTransformer(), LinearRegression())\n",
292 | "\n",
293 | "print(\"fit pipeline 2\")\n",
294 | "pipe2.fit(train_X, train_y)\n",
295 | "print(\"predict via pipeline 2\")\n",
296 | "preds2 = pipe2.predict(test_X)\n",
297 | "print(f\"\\n{preds2}\") # should be [14. 17.]\n",
298 | "print(f\"RMSE: {np.sqrt(mean_squared_error(test_y, preds2))}\\n\")"
299 | ],
300 | "execution_count": 34,
301 | "outputs": [
302 | {
303 | "output_type": "stream",
304 | "text": [
305 | "create pipeline 2\n",
306 | "\n",
307 | ">>>>>>>init() called.\n",
308 | "\n",
309 | "fit pipeline 2\n",
310 | "\n",
311 | ">>>>>>>fit() called.\n",
312 | "\n",
313 | "\n",
314 | ">>>>>>>transform() called.\n",
315 | "\n",
316 | "predict via pipeline 2\n",
317 | "\n",
318 | ">>>>>>>transform() called.\n",
319 | "\n",
320 | "\n",
321 | "[14. 17.]\n",
322 | "RMSE: 1.2560739669470201e-15\n",
323 | "\n"
324 | ],
325 | "name": "stdout"
326 | }
327 | ]
328 | },
329 | {
330 | "cell_type": "code",
331 | "metadata": {
332 | "id": "7QAxS_dkSpnC",
333 | "colab_type": "code",
334 | "colab": {}
335 | },
336 | "source": [
337 | "# we've assumed in the transform() function of our ExperimentalTransformer that the column name is X2. Let's not do so and\n",
338 | "# pass the column name via the constructor, __init__()"
339 | ],
340 | "execution_count": 0,
341 | "outputs": []
342 | },
343 | {
344 | "cell_type": "code",
345 | "metadata": {
346 | "id": "qB8bhqpdS-2m",
347 | "colab_type": "code",
348 | "colab": {}
349 | },
350 | "source": [
351 | "class ExperimentalTransformer_2(BaseEstimator, TransformerMixin):\n",
352 | " # add another additional parameter, just for fun, while we are at it\n",
353 | " def __init__(self, feature_name, additional_param = \"Himanshu\"): \n",
354 | " print('\\n>>>>>>>init() called.\\n')\n",
355 | " self.feature_name = feature_name\n",
356 | " self.additional_param = additional_param\n",
357 | "\n",
358 | " def fit(self, X, y = None):\n",
359 | " print('\\n>>>>>>>fit() called.\\n')\n",
360 | " print(f'\\nadditional param ~~~~~ {self.additional_param}\\n')\n",
361 | " return self\n",
362 | "\n",
363 | " def transform(self, X, y = None):\n",
364 | " print('\\n>>>>>>>transform() called.\\n')\n",
365 | " X_ = X.copy() # creating a copy to avoid changes to original dataset\n",
366 | " X_[self.feature_name] = 2 * np.sqrt(X_[self.feature_name])\n",
367 | " return X_"
368 | ],
369 | "execution_count": 0,
370 | "outputs": []
371 | },
372 | {
373 | "cell_type": "code",
374 | "metadata": {
375 | "id": "A_WkBgdkTOLU",
376 | "colab_type": "code",
377 | "colab": {}
378 | },
379 | "source": [
380 | "# take care to keep the parameter name exactly the same in the function argument as well as \n",
381 | "# the class' variable (feature_name). Changing that will cause problems later when we also\n",
382 | "# try to transform the target feature (y). It causes a double-call to __init__ for some reason."
383 | ],
384 | "execution_count": 0,
385 | "outputs": []
386 | },
387 | {
388 | "cell_type": "code",
389 | "metadata": {
390 | "id": "GB6WV6pUTuu-",
391 | "colab_type": "code",
392 | "colab": {
393 | "base_uri": "https://localhost:8080/",
394 | "height": 417
395 | },
396 | "outputId": "bdd1ecc7-285d-4d39-f8f3-d6101bdeeef7"
397 | },
398 | "source": [
399 | "print(\"create pipeline 2\")\n",
400 | "pipe2 = Pipeline(steps=[\n",
401 | " ('experimental_trans', ExperimentalTransformer_2('X2')),\n",
402 | " ('linear_model', LinearRegression())\n",
403 | "])\n",
404 | "print(\"fit pipeline 2\")\n",
405 | "pipe2.fit(train_X, train_y)\n",
406 | "print(\"predict via pipeline 2\")\n",
407 | "preds2 = pipe2.predict(test_X)\n",
408 | "print(f\"\\n{preds2}\") # should be [14. 17.]\n",
409 | "print(f\"RMSE: {np.sqrt(mean_squared_error(test_y, preds2))}\\n\")"
410 | ],
411 | "execution_count": 38,
412 | "outputs": [
413 | {
414 | "output_type": "stream",
415 | "text": [
416 | "create pipeline 2\n",
417 | "\n",
418 | ">>>>>>>init() called.\n",
419 | "\n",
420 | "fit pipeline 2\n",
421 | "\n",
422 | ">>>>>>>fit() called.\n",
423 | "\n",
424 | "\n",
425 | "additional param ~~~~~ Himanshu\n",
426 | "\n",
427 | "\n",
428 | ">>>>>>>transform() called.\n",
429 | "\n",
430 | "predict via pipeline 2\n",
431 | "\n",
432 | ">>>>>>>transform() called.\n",
433 | "\n",
434 | "\n",
435 | "[14. 17.]\n",
436 | "RMSE: 1.2560739669470201e-15\n",
437 | "\n"
438 | ],
439 | "name": "stdout"
440 | }
441 | ]
442 | },
443 | {
444 | "cell_type": "code",
445 | "metadata": {
446 | "id": "jBLKFfxcVhE5",
447 | "colab_type": "code",
448 | "colab": {}
449 | },
450 | "source": [
451 | "# let's take this a step further by modifying the dataframe to have target as squares of current values:"
452 | ],
453 | "execution_count": 0,
454 | "outputs": []
455 | },
456 | {
457 | "cell_type": "code",
458 | "metadata": {
459 | "id": "uqY60SOGViBG",
460 | "colab_type": "code",
461 | "colab": {}
462 | },
463 | "source": [
464 | "df = pd.DataFrame(columns=['X1', 'X2', 'y'], data=[\n",
465 | " [1,16,81],\n",
466 | " [4,36,256],\n",
467 | " [1,16,81],\n",
468 | " [2,9,64],\n",
469 | " [3,36,225],\n",
470 | " [2,49,256],\n",
471 | " [4,25,196],\n",
472 | " [5,36,289]\n",
473 | "])\n",
474 | "\n",
475 | "### sqrt(y) = X1 + 2 * sqrt(X2)\n",
476 | "\n",
477 | "train = df.iloc[:6]\n",
478 | "test = df.iloc[6:]\n",
479 | "\n",
480 | "train_X = train.drop('y', axis=1)\n",
481 | "train_y = train.y\n",
482 | "\n",
483 | "test_X = test.drop('y', axis=1)\n",
484 | "test_y = test.y"
485 | ],
486 | "execution_count": 0,
487 | "outputs": []
488 | },
489 | {
490 | "cell_type": "code",
491 | "metadata": {
492 | "id": "FSAo6KGzViSe",
493 | "colab_type": "code",
494 | "colab": {
495 | "base_uri": "https://localhost:8080/",
496 | "height": 145
497 | },
498 | "outputId": "2b16b778-bcda-4b00-a700-6e85bc162e30"
499 | },
500 | "source": [
501 | "# let's see model's performance with no input & target transformations:\n",
502 | "print(\"create pipeline 1\")\n",
503 | "pipe1 = Pipeline(steps=[\n",
504 | " ('linear_model', LinearRegression())\n",
505 | "])\n",
506 | "print(\"fit pipeline 1\")\n",
507 | "pipe1.fit(train_X, train_y)\n",
508 | "print(\"predict via pipeline 1\")\n",
509 | "preds1 = pipe1.predict(test_X)\n",
510 | "print(f\"\\n{preds1}\") \n",
511 | "print(f\"RMSE: {np.sqrt(mean_squared_error(test_y, preds1))}\\n\") "
512 | ],
513 | "execution_count": 41,
514 | "outputs": [
515 | {
516 | "output_type": "stream",
517 | "text": [
518 | "create pipeline 1\n",
519 | "fit pipeline 1\n",
520 | "predict via pipeline 1\n",
521 | "\n",
522 | "[200.34790002 279.04738423]\n",
523 | "RMSE: 7.679804528409077\n",
524 | "\n"
525 | ],
526 | "name": "stdout"
527 | }
528 | ]
529 | },
530 | {
531 | "cell_type": "code",
532 | "metadata": {
533 | "id": "i_PwqZaTVihG",
534 | "colab_type": "code",
535 | "colab": {
536 | "base_uri": "https://localhost:8080/",
537 | "height": 417
538 | },
539 | "outputId": "eb81b244-93a0-42fc-a80e-18a2dc259262"
540 | },
541 | "source": [
542 | "# with input transformation but no target transformation\n",
543 | "print(\"create pipeline 2\")\n",
544 | "pipe2 = Pipeline(steps=[\n",
545 | " ('experimental_trans', ExperimentalTransformer_2('X2')),\n",
546 | " ('linear_model', LinearRegression())\n",
547 | "])\n",
548 | "print(\"fit pipeline 2\")\n",
549 | "pipe2.fit(train_X, train_y)\n",
550 | "print(\"predict via pipeline 2\")\n",
551 | "preds2 = pipe2.predict(test_X)\n",
552 | "print(f\"\\n{preds2}\") \n",
553 | "print(f\"RMSE: {np.sqrt(mean_squared_error(test_y, preds2))}\\n\")"
554 | ],
555 | "execution_count": 42,
556 | "outputs": [
557 | {
558 | "output_type": "stream",
559 | "text": [
560 | "create pipeline 2\n",
561 | "\n",
562 | ">>>>>>>init() called.\n",
563 | "\n",
564 | "fit pipeline 2\n",
565 | "\n",
566 | ">>>>>>>fit() called.\n",
567 | "\n",
568 | "\n",
569 | "additional param ~~~~~ Himanshu\n",
570 | "\n",
571 | "\n",
572 | ">>>>>>>transform() called.\n",
573 | "\n",
574 | "predict via pipeline 2\n",
575 | "\n",
576 | ">>>>>>>transform() called.\n",
577 | "\n",
578 | "\n",
579 | "[207.42690058 280.94152047]\n",
580 | "RMSE: 9.887192456534303\n",
581 | "\n"
582 | ],
583 | "name": "stdout"
584 | }
585 | ]
586 | },
587 | {
588 | "cell_type": "code",
589 | "metadata": {
590 | "id": "9Ug-zMF0YISg",
591 | "colab_type": "code",
592 | "colab": {}
593 | },
594 | "source": [
595 | "# we'll now write a custom target transformer.\n",
596 | "# this needs 2 functions, one to transform and another to inverse-transform"
597 | ],
598 | "execution_count": 0,
599 | "outputs": []
600 | },
601 | {
602 | "cell_type": "code",
603 | "metadata": {
604 | "id": "TskWUHTcVit_",
605 | "colab_type": "code",
606 | "colab": {}
607 | },
608 | "source": [
609 | "def target_transform(target):\n",
610 | " print('\\n*****************target_transform() called.\\n')\n",
611 | " target_ = target.copy() \n",
612 | " target_ = np.sqrt(target_)\n",
613 | " return target_\n",
614 | "\n",
615 | "def inverse_target_transform(target):\n",
616 | " print('\\n*****************inverse_target_transform() called.\\n')\n",
617 | " target_ = target.copy() \n",
618 | " target_ = target_ ** 2\n",
619 | " return target_"
620 | ],
621 | "execution_count": 0,
622 | "outputs": []
623 | },
624 | {
625 | "cell_type": "code",
626 | "metadata": {
627 | "id": "OwNd-9__Ywx9",
628 | "colab_type": "code",
629 | "colab": {
630 | "base_uri": "https://localhost:8080/",
631 | "height": 799
632 | },
633 | "outputId": "da2ba50d-8f3d-405d-f137-b8abcb7d2c6d"
634 | },
635 | "source": [
636 | "# with input transformation & target transformation\n",
637 | "print(\"create pipeline 3\")\n",
638 | "# no change in input pipeline\n",
639 | "pipe3 = Pipeline(steps=[\n",
640 | " ('experimental_trans', ExperimentalTransformer_2('X2')),\n",
641 | " ('linear_model', LinearRegression())\n",
642 | "])\n",
643 | "\n",
644 | "# create a TargetTransformer \n",
645 | "model = TransformedTargetRegressor(regressor=pipe3, \n",
646 | " func=target_transform, \n",
647 | " inverse_func=inverse_target_transform)\n",
648 | "\n",
649 | "print(\"fit pipeline 3 [fit Model]\")\n",
650 | "# note the different syntax here; we fit the 'model' now, instead of 'pipe3'\n",
651 | "model.fit(train_X, train_y) \n",
652 | "print(\"predict via pipeline 3 [Model]\")\n",
653 | "preds3 = model.predict(test_X) # same here, using 'model' to predict\n",
654 | "print(f\"\\n{preds3}\") # should be [196. 289.]\n",
655 | "print(f\"RMSE: {np.sqrt(mean_squared_error(test_y, preds3))}\\n\")"
656 | ],
657 | "execution_count": 45,
658 | "outputs": [
659 | {
660 | "output_type": "stream",
661 | "text": [
662 | "create pipeline 3\n",
663 | "\n",
664 | ">>>>>>>init() called.\n",
665 | "\n",
666 | "fit pipeline 3 [fit Model]\n",
667 | "\n",
668 | "*****************target_transform() called.\n",
669 | "\n",
670 | "\n",
671 | "*****************inverse_target_transform() called.\n",
672 | "\n",
673 | "\n",
674 | "*****************target_transform() called.\n",
675 | "\n",
676 | "\n",
677 | "*****************inverse_target_transform() called.\n",
678 | "\n",
679 | "\n",
680 | "*****************target_transform() called.\n",
681 | "\n",
682 | "\n",
683 | ">>>>>>>init() called.\n",
684 | "\n",
685 | "\n",
686 | ">>>>>>>fit() called.\n",
687 | "\n",
688 | "\n",
689 | "additional param ~~~~~ Himanshu\n",
690 | "\n",
691 | "\n",
692 | ">>>>>>>transform() called.\n",
693 | "\n",
694 | "predict via pipeline 3 [Model]\n",
695 | "\n",
696 | ">>>>>>>transform() called.\n",
697 | "\n",
698 | "\n",
699 | "*****************inverse_target_transform() called.\n",
700 | "\n",
701 | "\n",
702 | "[196. 289.]\n",
703 | "RMSE: 4.0194366942304644e-14\n",
704 | "\n"
705 | ],
706 | "name": "stdout"
707 | }
708 | ]
709 | },
710 | {
711 | "cell_type": "code",
712 | "metadata": {
713 | "id": "YYQ0HystcAmr",
714 | "colab_type": "code",
715 | "colab": {}
716 | },
717 | "source": [
718 | "# perfect predictions!"
719 | ],
720 | "execution_count": 0,
721 | "outputs": []
722 | },
723 | {
724 | "cell_type": "code",
725 | "metadata": {
726 | "id": "YTM-qswpcCwf",
727 | "colab_type": "code",
728 | "colab": {}
729 | },
730 | "source": [
731 | "# we can even use in-built Transformers instead of user-defined functions. Example-\n",
732 | "# model = TransformedTargetRegressor(regressor=pipe3, transformer=PowerTransformer())\n",
733 | "# or\n",
734 | "# model = TransformedTargetRegressor(regressor=pipe3, transformer=StandardScaler())\n",
735 | "# using a built-in transformer does not require us to specify the inverse_transformer() as that is taken care of internally."
736 | ],
737 | "execution_count": 0,
738 | "outputs": []
739 | },
740 | {
741 | "cell_type": "code",
742 | "metadata": {
743 | "id": "mmEs3zaKEneX",
744 | "colab_type": "code",
745 | "colab": {}
746 | },
747 | "source": [
748 | "# in case you want to have a custom transformer inside TransformedTargetRegressor, you can do that too. The only additional \n",
749 | "# function you'll have to implement would be inverse_transform(). Here's an example:"
750 | ],
751 | "execution_count": 0,
752 | "outputs": []
753 | },
754 | {
755 | "cell_type": "code",
756 | "metadata": {
757 | "id": "yBZ-S3y9_AQz",
758 | "colab_type": "code",
759 | "colab": {}
760 | },
761 | "source": [
762 | "class CustomTargetTransformer(BaseEstimator, TransformerMixin):\n",
763 | " # no need to implement __init__ in this particular case\n",
764 | " \n",
765 | " def fit(self, target):\n",
766 | " return self\n",
767 | "\n",
768 | " def transform(self, target):\n",
769 | " print('\\n%%%%%%%%%%%%%%%custom_target_transform() called.\\n')\n",
770 | " target_ = target.copy() \n",
771 | " target_ = np.sqrt(target_)\n",
772 | " return target_\n",
773 | "\n",
774 | " # need to implement this too\n",
775 | " def inverse_transform(self, target):\n",
776 | " print('\\n%%%%%%%%%%%%%%%custom_inverse_target_transform() called.\\n')\n",
777 | " target_ = target.copy() \n",
778 | " target_ = target_ ** 2\n",
779 | " return target_"
780 | ],
781 | "execution_count": 0,
782 | "outputs": []
783 | },
784 | {
785 | "cell_type": "code",
786 | "metadata": {
787 | "id": "rsCBFs6o-gBN",
788 | "colab_type": "code",
789 | "colab": {
790 | "base_uri": "https://localhost:8080/",
791 | "height": 581
792 | },
793 | "outputId": "6ea392f9-7a0a-45d1-a7d5-05a1a0ac0d41"
794 | },
795 | "source": [
796 | "# with input transformation & target transformation\n",
797 | "print(\"create pipeline 3.1\")\n",
798 | "# no change in input pipeline\n",
799 | "pipe3_1 = Pipeline(steps=[\n",
800 | " ('experimental_trans', ExperimentalTransformer_2('X2')),\n",
801 | " ('linear_model', LinearRegression())\n",
802 | "])\n",
803 | "\n",
804 | "# create a TargetTransformer \n",
805 | "# By default, the provided functions are checked at each fit to be the inverse of each other. However, it is \n",
806 | "# possible to bypass this checking by setting check_inverse to False.\n",
807 | "model = TransformedTargetRegressor(regressor=pipe3_1, \n",
808 | " transformer=CustomTargetTransformer(),\n",
809 | " check_inverse=False) # avoid repeated calls\n",
810 | "\n",
811 | "print(\"fit pipeline 3.1 [fit Model]\")\n",
812 | "model.fit(train_X, train_y) \n",
813 | "print(\"predict via pipeline 3.1 [Model]\")\n",
814 | "preds3_1 = model.predict(test_X) \n",
815 | "print(f\"\\n{preds3_1}\") # should be [196. 289.]\n",
816 | "print(f\"RMSE: {np.sqrt(mean_squared_error(test_y, preds3_1))}\\n\")"
817 | ],
818 | "execution_count": 50,
819 | "outputs": [
820 | {
821 | "output_type": "stream",
822 | "text": [
823 | "create pipeline 3.1\n",
824 | "\n",
825 | ">>>>>>>init() called.\n",
826 | "\n",
827 | "fit pipeline 3.1 [fit Model]\n",
828 | "\n",
829 | "%%%%%%%%%%%%%%%custom_target_transform() called.\n",
830 | "\n",
831 | "\n",
832 | ">>>>>>>init() called.\n",
833 | "\n",
834 | "\n",
835 | ">>>>>>>fit() called.\n",
836 | "\n",
837 | "\n",
838 | "additional param ~~~~~ Himanshu\n",
839 | "\n",
840 | "\n",
841 | ">>>>>>>transform() called.\n",
842 | "\n",
843 | "predict via pipeline 3.1 [Model]\n",
844 | "\n",
845 | ">>>>>>>transform() called.\n",
846 | "\n",
847 | "\n",
848 | "%%%%%%%%%%%%%%%custom_inverse_target_transform() called.\n",
849 | "\n",
850 | "\n",
851 | "[196. 289.]\n",
852 | "RMSE: 4.0194366942304644e-14\n",
853 | "\n"
854 | ],
855 | "name": "stdout"
856 | }
857 | ]
858 | },
859 | {
860 | "cell_type": "code",
861 | "metadata": {
862 | "id": "Ly2iew-uEhdU",
863 | "colab_type": "code",
864 | "colab": {}
865 | },
866 | "source": [
867 | "# let's now see how to get and set parameters of the model. We'll also cache the transformer to \n",
868 | "# avoid repeated computation and make it more efficient."
869 | ],
870 | "execution_count": 0,
871 | "outputs": []
872 | },
873 | {
874 | "cell_type": "code",
875 | "metadata": {
876 | "id": "I0KB3cjDC5ho",
877 | "colab_type": "code",
878 | "colab": {
879 | "base_uri": "https://localhost:8080/",
880 | "height": 490
881 | },
882 | "outputId": "e17e937f-9ff3-485a-8730-846e2851a6da"
883 | },
884 | "source": [
885 | "# get all the params of our model\n",
886 | "model.get_params()"
887 | ],
888 | "execution_count": 52,
889 | "outputs": [
890 | {
891 | "output_type": "execute_result",
892 | "data": {
893 | "text/plain": [
894 | "{'check_inverse': False,\n",
895 | " 'func': None,\n",
896 | " 'inverse_func': None,\n",
897 | " 'regressor': Pipeline(memory=None,\n",
898 | " steps=[('experimental_trans',\n",
899 | " ExperimentalTransformer_2(additional_param='Himanshu',\n",
900 | " feature_name='X2')),\n",
901 | " ('linear_model',\n",
902 | " LinearRegression(copy_X=True, fit_intercept=True, n_jobs=None,\n",
903 | " normalize=False))],\n",
904 | " verbose=False),\n",
905 | " 'regressor__experimental_trans': ExperimentalTransformer_2(additional_param='Himanshu', feature_name='X2'),\n",
906 | " 'regressor__experimental_trans__additional_param': 'Himanshu',\n",
907 | " 'regressor__experimental_trans__feature_name': 'X2',\n",
908 | " 'regressor__linear_model': LinearRegression(copy_X=True, fit_intercept=True, n_jobs=None, normalize=False),\n",
909 | " 'regressor__linear_model__copy_X': True,\n",
910 | " 'regressor__linear_model__fit_intercept': True,\n",
911 | " 'regressor__linear_model__n_jobs': None,\n",
912 | " 'regressor__linear_model__normalize': False,\n",
913 | " 'regressor__memory': None,\n",
914 | " 'regressor__steps': [('experimental_trans',\n",
915 | " ExperimentalTransformer_2(additional_param='Himanshu', feature_name='X2')),\n",
916 | " ('linear_model',\n",
917 | " LinearRegression(copy_X=True, fit_intercept=True, n_jobs=None, normalize=False))],\n",
918 | " 'regressor__verbose': False,\n",
919 | " 'transformer': CustomTargetTransformer()}"
920 | ]
921 | },
922 | "metadata": {
923 | "tags": []
924 | },
925 | "execution_count": 52
926 | }
927 | ]
928 | },
929 | {
930 | "cell_type": "code",
931 | "metadata": {
932 | "id": "h729iuXW26OI",
933 | "colab_type": "code",
934 | "colab": {}
935 | },
936 | "source": [
937 | "from tempfile import mkdtemp\n",
938 | "from shutil import rmtree\n",
939 | "# read about caching and side effect at: https://scikit-learn.org/stable/modules/compose.html?highlight=transformedtargetregressor#pipeline-chaining-estimators"
940 | ],
941 | "execution_count": 0,
942 | "outputs": []
943 | },
944 | {
945 | "cell_type": "code",
946 | "metadata": {
947 | "id": "ev2o5jU-Cpcb",
948 | "colab_type": "code",
949 | "colab": {
950 | "base_uri": "https://localhost:8080/",
951 | "height": 635
952 | },
953 | "outputId": "374c1b5f-6618-4ca7-cd43-72b6f83c008b"
954 | },
955 | "source": [
956 | "cachedir = mkdtemp()\n",
957 | "print(\"create pipeline 4\")\n",
958 | "pipe4 = Pipeline(steps=[\n",
959 | " # incorrect column name passed\n",
960 | " ('experimental_trans', ExperimentalTransformer_2('X1')), \n",
961 | " ('linear_model', LinearRegression())\n",
962 | "], memory=cachedir)\n",
963 | "# create a TargetTransformer\n",
964 | "model = TransformedTargetRegressor(regressor=pipe4, \n",
965 | " func=target_transform, \n",
966 | " inverse_func=inverse_target_transform, \n",
967 | " check_inverse=False) \n",
968 | "# correcting the column name using set_params()\n",
969 | "model.set_params(regressor__experimental_trans__feature_name = 'X2') \n",
970 | "\n",
971 | "print(\"fit pipeline 4 [fit Model]\")\n",
972 | "model.fit(train_X, train_y) \n",
973 | "print(\"predict via pipeline 4 [Model]\")\n",
974 | "preds4 = model.predict(test_X) \n",
975 | "print(f\"\\n{preds4}\") # should be [196. 289.]\n",
976 | "print(f\"RMSE: {np.sqrt(mean_squared_error(test_y, preds4))}\\n\")\n",
977 | "\n",
978 | "# Clear the cache directory when you don't need it anymore\n",
979 | "rmtree(cachedir)"
980 | ],
981 | "execution_count": 54,
982 | "outputs": [
983 | {
984 | "output_type": "stream",
985 | "text": [
986 | "create pipeline 4\n",
987 | "\n",
988 | ">>>>>>>init() called.\n",
989 | "\n",
990 | "fit pipeline 4 [fit Model]\n",
991 | "\n",
992 | "*****************target_transform() called.\n",
993 | "\n",
994 | "\n",
995 | ">>>>>>>init() called.\n",
996 | "\n",
997 | "\n",
998 | ">>>>>>>init() called.\n",
999 | "\n",
1000 | "\n",
1001 | ">>>>>>>fit() called.\n",
1002 | "\n",
1003 | "\n",
1004 | "additional param ~~~~~ Himanshu\n",
1005 | "\n",
1006 | "\n",
1007 | ">>>>>>>transform() called.\n",
1008 | "\n",
1009 | "predict via pipeline 4 [Model]\n",
1010 | "\n",
1011 | ">>>>>>>transform() called.\n",
1012 | "\n",
1013 | "\n",
1014 | "*****************inverse_target_transform() called.\n",
1015 | "\n",
1016 | "\n",
1017 | "[196. 289.]\n",
1018 | "RMSE: 4.0194366942304644e-14\n",
1019 | "\n"
1020 | ],
1021 | "name": "stdout"
1022 | }
1023 | ]
1024 | },
1025 | {
1026 | "cell_type": "code",
1027 | "metadata": {
1028 | "id": "_5ZCrirHIADe",
1029 | "colab_type": "code",
1030 | "colab": {}
1031 | },
1032 | "source": [
1033 | "# NEXT STEPS:\n",
1034 | "\n",
1035 | "# 1. FeatureUnion and ColumnTransformer\n",
1036 | "# Some great examples:\n",
1037 | "# https://scikit-learn.org/stable/modules/compose.html#featureunion-composite-feature-spaces \n",
1038 | "# https://scikit-learn.org/stable/modules/generated/sklearn.compose.ColumnTransformer.html#sklearn.compose.ColumnTransformer\n",
1039 | "\n",
1040 | "\n",
1041 | "# 2. Using GridSearch with Pipelines\n",
1042 | "# Example: https://scikit-learn.org/stable/auto_examples/compose/plot_feature_union.html?highlight=pipeline\n"
1043 | ],
1044 | "execution_count": 0,
1045 | "outputs": []
1046 | },
1047 | {
1048 | "cell_type": "code",
1049 | "metadata": {
1050 | "id": "M7b3qRdI3EMB",
1051 | "colab_type": "code",
1052 | "colab": {}
1053 | },
1054 | "source": [
1055 | ""
1056 | ],
1057 | "execution_count": 0,
1058 | "outputs": []
1059 | }
1060 | ]
1061 | }
--------------------------------------------------------------------------------
/PipelineFoundation/README.md:
--------------------------------------------------------------------------------
1 | #iamJustAStudent - Let's study AI/ML together : http://iamjustastudent.com
2 | # Pipelines & Custom Transformers in scikit-learn: The step-by-step guide (with Python code)
3 | Understand the basics and workings of scikit-learn pipelines from the ground up, so that you can build your own..
4 |
5 | Watch the video tutorial at: https://youtu.be/mOYJCR0IDk8 6 | or
7 | Read the complete tutorial at: https://towardsdatascience.com/pipelines-custom-transformers-in-scikit-learn-the-step-by-step-guide-with-python-code-4a7d9b068156 8 | -------------------------------------------------------------------------------- /Predicted_R-squared/PredictedR2.csv: -------------------------------------------------------------------------------- 1 | X,Y 2 | 70,23 3 | 68,30 4 | 76,26 5 | 69,11 6 | 82,10 7 | 83,3 8 | 79,9 9 | 78,12 10 | 88,7 11 | 91,39 12 | 89,23 13 | 73,19 14 | -------------------------------------------------------------------------------- /Predicted_R-squared/README.md: -------------------------------------------------------------------------------- 1 | #iamJustAStudent - Let's study AI/ML together : http://iamjustastudent.com 2 | # Predicted R-squared 3 | An awesome metric you haven't heard of! 4 | 5 | ## Step by Step Guide: 6 | https://iamjustastudent.com/2020/09/07/predicted-r-squared-an-awesome-metric-you-havent-heard-of/ 7 | -------------------------------------------------------------------------------- /README.md: -------------------------------------------------------------------------------- 1 | # MachineLearning 2 | Let's study AI/ML together. 3 | -------------------------------------------------------------------------------- /SleepyOne/README.md: -------------------------------------------------------------------------------- 1 | #iamJustAStudent - Let's study AI/ML together : http://iamjustastudent.com 2 | # SleepyOne 3 | A Fun Project (Pose Detector) With Google's Teachable Machine 4 | 5 | ## Step by Step Guide: 6 | https://medium.com/@himanshuhc/a-fun-project-pose-detector-with-googles-teachable-machine-6c7c8d650be1 7 | -------------------------------------------------------------------------------- /SleepyOne/index.html: -------------------------------------------------------------------------------- 1 |
Teachable Machine Pose Model
2 |
3 |
4 |
5 | 6 | 10 | 11 | 12 | 100 | --------------------------------------------------------------------------------