/
intro.ipynb
760 lines (760 loc) · 20.6 KB
/
intro.ipynb
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
{
"cells": [
{
"cell_type": "markdown",
"metadata": {
"slideshow": {
"slide_type": "slide"
}
},
"source": [
"# Going deeper with Tensorflow\n",
"\n",
"We will start our course with [Tensorflow](https://www.tensorflow.org/) intro.\n",
"\n",
"To install tensorflow on your machine run:\n",
"\n",
"* `pip install tensorflow` -- **cpu-only** version for Linux & Mac OSX\n",
"* if you want GPU support try -- `pip install tensorflow-gpu`\n",
"* also see docs here: [TF install page](https://www.tensorflow.org/install/)"
]
},
{
"cell_type": "code",
"execution_count": 4,
"metadata": {
"collapsed": true,
"slideshow": {
"slide_type": "slide"
}
},
"outputs": [],
"source": [
"import numpy as np\n",
"import tensorflow as tf\n",
"\n",
"from sklearn.metrics import roc_auc_score, mean_squared_error\n",
"from sklearn.model_selection import train_test_split\n",
"\n",
"import matplotlib.pyplot as plt\n",
"%matplotlib inline\n",
"\n",
"gpu_options = tf.GPUOptions(allow_growth=True, per_process_gpu_memory_fraction=0.1)\n",
"s = tf.InteractiveSession(config=tf.ConfigProto(gpu_options=gpu_options))"
]
},
{
"cell_type": "markdown",
"metadata": {
"slideshow": {
"slide_type": "subslide"
}
},
"source": [
"# Lets start\n",
"\n",
"Implement simple function using numpy to calculate sum of squares from 0 to N-1\n",
"\n",
"\n",
"**Hint:**\n",
"use `numpy.arange(N)` to get array from 0 to N-1 inclusive"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"collapsed": true,
"slideshow": {
"slide_type": "fragment"
}
},
"outputs": [],
"source": [
"def sum_squares(N):\n",
" return <student.Implement_me()>"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"collapsed": true,
"slideshow": {
"slide_type": "skip"
}
},
"outputs": [],
"source": [
"%%time\n",
"sum_squares(10**8)"
]
},
{
"cell_type": "markdown",
"metadata": {
"slideshow": {
"slide_type": "slide"
}
},
"source": [
"# Tensoflow teaser\n",
"\n",
"Doing the very same thing"
]
},
{
"cell_type": "code",
"execution_count": 2,
"metadata": {
"collapsed": true
},
"outputs": [],
"source": [
"#I gonna be your function parameter\n",
"N = tf.placeholder('int64', name=\"input_to_your_function\")\n",
"\n",
"#i am a recipe on how to produce sum of squares of arange of N given N\n",
"result = tf.reduce_sum((tf.range(N)**2))"
]
},
{
"cell_type": "code",
"execution_count": 3,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"662921401752298880\n",
"CPU times: user 1.24 s, sys: 2.78 s, total: 4.02 s\n",
"Wall time: 3.19 s\n"
]
}
],
"source": [
"%%time\n",
"#example of computing the same as sum_squares\n",
"print(result.eval({N:10**8}))"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"# How does it work?\n",
"1. define placeholders where you'll send inputs;\n",
"2. make symbolic graph: a recipe for mathematical transformation of those placeholders;\n",
"3. compute outputs of your graph with particular values for each placeholder\n",
" * output.eval({placeholder:value}) \n",
" * s.run(output, {placeholder:value})\n",
"\n",
"* So far there are two main entities: \"placeholder\" and \"transformation\"\n",
"* Both can be numbers, vectors, matrices, tensors, etc.\n",
"* Both can be int32/64, floats of booleans (uint8) of various size.\n",
"\n",
"* You can define new transformations as an arbitrary operation on placeholders and other transformations\n",
" * tf.reduce_sum(tf.arange(N)\\**2) are 3 sequential transformations of placeholder N\n",
" * There's a tensorflow symbolic version for every numpy function\n",
" * `a+b, a/b, a**b, ...` behave just like in numpy\n",
" * np.mean -> tf.reduce_mean\n",
" * np.arange -> tf.range\n",
" * np.cumsum -> tf.cumsum\n",
" * If if you can't find the op you need, see the [docs](https://www.tensorflow.org/api_docs/python).\n",
" \n",
" \n",
"Still confused? We gonna fix that."
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"collapsed": true
},
"outputs": [],
"source": [
"#Default placeholder that can be arbitrary float32 scalar, vertor, matrix, etc.\n",
"arbitrary_input = tf.placeholder('float32')\n",
"\n",
"#Input vector of arbitrary length\n",
"input_vector = tf.placeholder('float32',shape=(None,))\n",
"\n",
"#Input vector that _must_ have 10 elements and integer type\n",
"fixed_vector = tf.placeholder('int32',shape=(10,))\n",
"\n",
"#Matrix of arbitrary n_rows and 15 columns (e.g. a minibatch your data table)\n",
"input_matrix = tf.placeholder('float32',shape=(None,15))\n",
"\n",
"#You can generally use None whenever you don't need a specific shape\n",
"input1 = tf.placeholder('float64',shape=(None,100,None))\n",
"input2 = tf.placeholder('int32',shape=(None,None,3,224,224))"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"collapsed": true
},
"outputs": [],
"source": [
"#elementwise multiplication\n",
"double_the_vector = input_vector*2\n",
"\n",
"#elementwise cosine\n",
"elementwise_cosine = tf.cos(input_vector)\n",
"\n",
"#difference between squared vector and vector itself\n",
"vector_squares = input_vector**2 - input_vector\n"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"**[1 point max]**"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"collapsed": true
},
"outputs": [],
"source": [
"#Practice time: create two vectors of type float32\n",
"my_vector = <student.init_float32_vector()>\n",
"my_vector2 = <student.init_one_more_such_vector()>"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"collapsed": true
},
"outputs": [],
"source": [
"#Write a transformation(recipe):\n",
"#(vec1)*(vec2) / (sin(vec1) +1)\n",
"my_transformation = <student.implementwhatwaswrittenabove()>"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"collapsed": true
},
"outputs": [],
"source": [
"print(my_transformation)\n",
"#it's okay, it's a symbolic graph"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"collapsed": true
},
"outputs": [],
"source": [
"#\n",
"dummy = np.arange(5).astype('float32')\n",
"\n",
"my_transformation.eval({my_vector:dummy,my_vector2:dummy[::-1]})"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"### Visualizing graphs\n",
"\n",
"It's often useful to visualize the computation graph when debugging or optimizing. \n",
"Interactive visualization is where tensorflow really shines as compared to other frameworks. \n",
"\n",
"There's a special instrument for that, called Tensorboard. You can launch it from console:\n",
"\n",
"```tensorboard --logdir=/tmp/tboard --port=7007```\n",
"\n",
"If you're pathologically afraid of consoles, try this:\n",
"\n",
"```os.system(\"tensorboard --logdir=/tmp/tboard --port=7007 &\"```\n",
"\n",
"_(but don't tell anyone we taught you that)_"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"collapsed": true
},
"outputs": [],
"source": [
"# launch tensorflow the ugly way, uncomment if you need that\n",
"import os\n",
"port = 7007\n",
"\n",
"#!killall tensorboard\n",
"os.system(\"tensorboard --logdir=./tboard --port=%d &\" % port)\n",
"\n",
"# show graph to tensorboard\n",
"writer = tf.summary.FileWriter(\"./tboard\", graph=tf.get_default_graph())\n",
"writer.close()"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"One basic functionality of tensorboard is drawing graphs. One you've run the cell above, go to `localhost:7007` in your browser and switch to _graphs_ tab in the topbar. \n",
"\n",
"Here's what you should see:\n",
"\n",
"<img src=\"https://www.tensorflow.org/images/graph_vis_animation.gif\" width=780>\n",
"\n",
"Tensorboard also allows you to draw graphs (e.g. learning curves), record images & audio ~~and play flash games~~. This is useful when monitoring learning progress and catching some training issues.\n",
"\n",
"One researcher said:\n",
"```\n",
"If you spent last four hours of your worktime watching as your algorithm prints numbers and draws figures, you're probably doing deep learning wrong.\n",
"```"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"You can read more on tensorboard usage [here](https://www.tensorflow.org/get_started/graph_viz)"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"# Do It Yourself\n",
"\n",
"__[2 points max]__"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"collapsed": true
},
"outputs": [],
"source": [
"# Quest #1 - implement a function that computes a mean squared error of two input vectors\n",
"# Your function has to take 2 vectors and return a single number\n",
"\n",
"<student.define_inputs_and_transformations()>\n",
"\n",
"mse =<student.define_transformation()>\n",
"\n",
"compute_mse = lambda vector1, vector2: <how to run you graph?>"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"collapsed": true
},
"outputs": [],
"source": [
"# Tests\n",
"for n in [1,5,10,10**3]:\n",
" \n",
" elems = [np.arange(n),np.arange(n,0,-1), np.zeros(n),\n",
" np.ones(n),np.random.random(n),np.random.randint(100,size=n)]\n",
" \n",
" for el in elems:\n",
" for el_2 in elems:\n",
" true_mse = np.array(mean_squared_error(el,el_2))\n",
" my_mse = compute_mse(el,el_2)\n",
" if not np.allclose(true_mse,my_mse):\n",
" print('Wrong result:')\n",
" print('mse(%s,%s)' % (el,el_2))\n",
" print(\"should be: %f, but your function returned %f\" % (true_mse,my_mse))\n",
" raise ValueError(\"Что-то не так\")\n",
"\n",
"print(\"All tests passed\") "
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"# variables\n",
"\n",
"The inputs and transformations have no value outside function call. This isn't too comfortable if you want your model to have parameters (e.g. network weights) that are always present, but can change their value over time.\n",
"\n",
"Tensorflow solves this with `tf.Variable` objects.\n",
"* You can assign variable a value at any time in your graph\n",
"* Unlike placeholders, there's no need to explicitly pass values to variables when `s.run(...)`-ing\n",
"* You can use variables the same way you use transformations \n",
" "
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"collapsed": true
},
"outputs": [],
"source": [
"#creating shared variable\n",
"shared_vector_1 = tf.Variable(initial_value=np.ones(5))"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"collapsed": true
},
"outputs": [],
"source": [
"#initialize variable(s) with initial values\n",
"s.run(tf.global_variables_initializer())\n",
"\n",
"#evaluating shared variable (outside symbolicd graph)\n",
"print(\"initial value\", s.run(shared_vector_1))\n",
"\n",
"# within symbolic graph you use them just as any other inout or transformation, not \"get value\" needed"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"collapsed": true
},
"outputs": [],
"source": [
"#setting new value\n",
"s.run(shared_vector_1.assign(np.arange(5)))\n",
"\n",
"#getting that new value\n",
"print(\"new value\", s.run(shared_vector_1))\n"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"# tf.gradients - why graphs matter\n",
"* Tensorflow can compute derivatives and gradients automatically using the computation graph\n",
"* Gradients are computed as a product of elementary derivatives via chain rule:\n",
"\n",
"$$ {\\partial f(g(x)) \\over \\partial x} = {\\partial f(g(x)) \\over \\partial g(x)}\\cdot {\\partial g(x) \\over \\partial x} $$\n",
"\n",
"It can get you the derivative of any graph as long as it knows how to differentiate elementary operations"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"collapsed": true
},
"outputs": [],
"source": [
"my_scalar = tf.placeholder('float32')\n",
"\n",
"scalar_squared = my_scalar**2\n",
"\n",
"#a derivative of scalar_squared by my_scalar\n",
"derivative = tf.gradients(scalar_squared, my_scalar)[0]"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"collapsed": true
},
"outputs": [],
"source": [
"x = np.linspace(-3,3)\n",
"x_squared, x_squared_der = s.run([scalar_squared,derivative],\n",
" {my_scalar:x})\n",
"\n",
"plt.plot(x, x_squared,label=\"x^2\")\n",
"plt.plot(x, x_squared_der, label=\"derivative\")\n",
"plt.legend();"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"# Why that rocks"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"**[2 points max]**"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"collapsed": true
},
"outputs": [],
"source": [
"my_vector = tf.placeholder('float32',[None])\n",
"\n",
"#Compute the gradient of the next weird function over my_scalar and my_vector\n",
"#warning! Trying to understand the meaning of that function may result in permanent brain damage\n",
"\n",
"weird_psychotic_function = tf.reduce_mean((my_vector+my_scalar)**(1+tf.nn.moments(my_vector,[0])[1]) + 1./ tf.atan(my_scalar))/(my_scalar**2 + 1) + 0.01*tf.sin(2*my_scalar**1.5)*(tf.reduce_sum(my_vector)* my_scalar**2)*tf.exp((my_scalar-4)**2)/(1+tf.exp((my_scalar-4)**2))*(1.-(tf.exp(-(my_scalar-4)**2))/(1+tf.exp(-(my_scalar-4)**2)))**2\n",
"\n",
"der_by_scalar = <student.compute_grad_over_scalar()>\n",
"der_by_vector = <student.compute_grad_over_vector()>"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"collapsed": true
},
"outputs": [],
"source": [
"#Plotting your derivative\n",
"scalar_space = np.linspace(1, 7, 100)\n",
"\n",
"y = [s.run(weird_psychotic_function, {my_scalar:x, my_vector:[1, 2, 3]})\n",
" for x in scalar_space]\n",
"\n",
"plt.plot(scalar_space, y, label='function')\n",
"\n",
"y_der_by_scalar = [s.run(der_by_scalar, {my_scalar:x, my_vector:[1, 2, 3]})\n",
" for x in scalar_space]\n",
"\n",
"plt.plot(scalar_space, y_der_by_scalar, label='derivative')\n",
"plt.grid()\n",
"plt.legend();"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"# Almost done - optimizers\n",
"\n",
"While you can perform gradient descent by hand with automatic grads from above, tensorflow also has some optimization methods implemented for you.\n",
"\n",
"You can try different optimizers (eg. GradientDescentOptimizer, MomentumOptimizer, AdamOptimizer) and combine their trajectories on one plot (don't forget the legend). **[1 extra point]**"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"collapsed": true
},
"outputs": [],
"source": [
"y_guess = tf.Variable(np.zeros(2,dtype='float32'))\n",
"y_true = tf.range(1,3,dtype='float32')\n",
"\n",
"loss = tf.reduce_mean((y_guess - y_true + tf.random_normal([2]))**2) \n",
"\n",
"optimizer = tf.train.MomentumOptimizer(0.01,0.9).minimize(loss,var_list=y_guess)\n",
"\n",
"#same, but more detailed:\n",
"#updates = [[tf.gradients(loss,y_guess)[0], y_guess]]\n",
"#optimizer = tf.train.MomentumOptimizer(0.01,0.9).apply_gradients(updates)"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"collapsed": true
},
"outputs": [],
"source": [
"from IPython.display import clear_output\n",
"\n",
"s.run(tf.global_variables_initializer())\n",
"\n",
"guesses = [s.run(y_guess)]\n",
"\n",
"for _ in range(100):\n",
" s.run(optimizer)\n",
" guesses.append(s.run(y_guess))\n",
" \n",
" clear_output(True)\n",
" plt.plot(*zip(*guesses),marker='.')\n",
" plt.scatter(*s.run(y_true),c='red')\n",
" plt.show()"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"# Logistic regression example\n",
"Implement the regular logistic regression training algorithm\n",
"\n",
"Tips:\n",
"* Use a shared variable for weights\n",
"* X and y are potential inputs\n",
"* Compile 2 functions:\n",
" * `train_function(X, y)` - returns error and computes weights' new values __(through updates)__\n",
" * `predict_fun(X)` - just computes probabilities (\"y\") given data\n",
" \n",
" \n",
"We shall train on a two-class MNIST dataset\n",
"* please note that target `y` are `{0,1}` and not `{-1,1}` as in some formulae"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"**[5 points max]**"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"collapsed": true
},
"outputs": [],
"source": [
"from sklearn.datasets import load_digits\n",
"mnist = load_digits(2)\n",
"\n",
"X,y = mnist.data, mnist.target\n",
"\n",
"print(\"y [shape - %s]:\" % (str(y.shape)), y[:10])\n",
"print(\"X [shape - %s]:\" % (str(X.shape)))"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"collapsed": true
},
"outputs": [],
"source": [
"print('X:\\n',X[:3,:10])\n",
"print('y:\\n',y[:10])\n",
"plt.imshow(X[0].reshape([8,8]))"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"collapsed": true
},
"outputs": [],
"source": [
"# inputs and shareds\n",
"weights = <student.code_variable()>\n",
"input_X = <student.code_placeholder()>\n",
"input_y = <student.code_placeholder()>"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"collapsed": true
},
"outputs": [],
"source": [
"predicted_y = <predicted probabilities for input_X>\n",
"loss = <logistic loss (scalar, mean over sample)>\n",
"\n",
"optimizer = <optimizer that minimizes loss>"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"collapsed": true
},
"outputs": [],
"source": [
"train_function = <compile function that takes X and y, returns log loss and updates weights>\n",
"predict_function = <compile function that takes X and computes probabilities of y>"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"collapsed": true
},
"outputs": [],
"source": [
"X_train, X_test, y_train, y_test = train_test_split(X, y)"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"collapsed": true
},
"outputs": [],
"source": [
"for i in range(5):\n",
" <run optimizer operation>\n",
" loss_i = <compute loss at iteration i>\n",
" \n",
" print(\"loss at iter %i:%.4f\" % (i, loss_i))\n",
" \n",
" print(\"train auc:\",roc_auc_score(y_train, predict_function(X_train)))\n",
" print(\"test auc:\",roc_auc_score(y_test, predict_function(X_test)))\n",
"\n",
" \n",
"print (\"resulting weights:\")\n",
"plt.imshow(weights.eval().reshape([8,8]))\n",
"plt.colorbar()"
]
}
],
"metadata": {
"kernelspec": {
"display_name": "Python 3",
"language": "python",
"name": "python3"
},
"language_info": {
"codemirror_mode": {
"name": "ipython",
"version": 3
},
"file_extension": ".py",
"mimetype": "text/x-python",
"name": "python",
"nbconvert_exporter": "python",
"pygments_lexer": "ipython3",
"version": "3.6.4"
}
},
"nbformat": 4,
"nbformat_minor": 2
}