-
Notifications
You must be signed in to change notification settings - Fork 0
/
training.py
691 lines (565 loc) · 28.7 KB
/
training.py
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
import pandas as pd
pd.options.mode.chained_assignment = None
import numpy as np
import matplotlib.pyplot as plt
import seaborn as sns
from sklearn.model_selection import train_test_split, KFold, GridSearchCV
from sklearn.preprocessing import OneHotEncoder
from sklearn.metrics import (
f1_score,
auc,
precision_score,
recall_score,
accuracy_score,
roc_curve,
roc_auc_score
)
import os
import re
import scipy.stats as st
import xgboost as xgb
import lightgbm as lgb
import warnings
warnings.filterwarnings('ignore')
import pickle
def create_dir_for_product_if_not_existent(product, plots_or_models: str = 'plots'):
"""
check if directory 'plots/{product}' exists
if true, save the fig in the directory as {filename_}.jpeg
if false, create the directory and save the fig in the newly created directory
"""
if plots_or_models == 'plots':
if os.path.isdir(f'{plots_or_models}/{product}'):
pass
# plt.savefig(fname=f'{plots_or_models}/{product}/{filename_}.jpeg')
else:
os.makedirs(f'{plots_or_models}/{product}')
# plt.savefig(fname=f'{plots_or_models}/{product}/{filename_}.jpeg')
if plots_or_models == 'models':
if os.path.isdir(f'{plots_or_models}/{product}'):
pass
# model.save_model(fname=f'{plots_or_models}/{product}/{filename_}.model')
else:
os.makedirs(f'{plots_or_models}/{product}')
# model.save_model(fname=f'{plots_or_models}/{product}/{filename_}.model')
class model_training(object):
"""
loads data from pickle by product_name.json, split the data into train and validation sets,
one-hot encode reviewer features (eye color, hair color, skin color, and skin tone),
cross one-hot encoded skin color and skin tone,
combine the crossed features, one-hot encoded features, and other features into two pd.DataFrames,
one is the training set with all features, and the other is the validation set with all features
return the training and validation sets
"""
def __init__(self, logger, file_name: str, random_seeds: int) -> None:
"""
logger: for passing logger.info
file_name: a string ends with '.json'
random_seeds: an integer for random state across the class
"""
self.logger = logger
self.data = pd.read_json(f'data_full_review_cleaned/{file_name}', lines=True)
self.all_features = [
'hair_color', 'eye_color', 'skin_tone', 'skin_type',
'skin_tone_cat', 'finish', 'coverage', 'shade_match',
'gifted', 'days_since_launch_scaled', 'month_of_purchase'
]
self.one_hot_features = [
'hair_color', 'eye_color', 'skin_tone', 'skin_type'
]
self.other_features = [
'skin_tone_cat', 'finish', 'coverage', 'shade_match',
'gifted', 'days_since_launch_scaled', 'month_of_purchase'
]
self.product_name = file_name.replace('.json', '')
self.logger.info(f"#################################")
self.logger.info(f"Starting with {self.product_name}")
self.logger.info(f"#################################")
self.random_state = random_seeds
self.X = self.data[self.all_features]
self.y = self.data['recommended']
# perhaps some logging to know if there's class imbalance issues?
### for feature engineering ###
def train_test_split(self, proportion: float = 0.3):
"""
split training and validation sets given proportion using train_test_split() from scikit-learn
"""
self.train_X, self.val_X, self.train_y, self.val_y = train_test_split(self.X, self.y, test_size=proportion,
random_state=self.random_state)
def dropping_outlier_reviewers(self, cols=None):
"""
Some reviewers can be considered outliers based on their features,
e.g., if only less than 10% of the reviewers in the dataset had gray hair, we drop these reviewers
I define outliers by (a) being less than 5% out of the whole dataset, and
(b) having less than 5 people in the validation set
This function drops these reviewers in both the training and validation sets after doing train_test_split
"""
self.logger.info(f'Before dropping outliers:')
self.logger.info(f'N in train_X is {len(self.train_X)}, N in val_X is {len(self.val_X)}')
if cols is None:
cols = ['hair_color', 'eye_color', 'skin_tone']
for c in cols:
c_proportion = self.val_X.groupby([c]).count()['coverage'] / len(self.val_X)
outliers_proportion = c_proportion[c_proportion <= 0.05].index
# if only less than 10% of some reviewers in the validation set had a certain feature
# like if only 5 people in the validation set out of 100 people in the complete dataset had red hair
# we drop these people
c_count = self.val_X.groupby([c]).count()['coverage']
outliers_count = c_count[c_count < 5].index
# if only less than 5 reviewers in the validation set had a certain feature
# we drop these reviewers
if not outliers_proportion.empty or not outliers_count.empty:
self.logger.info(f'#### Removing outliers in {c} ####')
n_dropped = 0
if not outliers_count.empty:
for i in outliers_count:
val_indices = self.val_X[self.val_X[c] == i].index
n_dropped += len(val_indices)
self.val_X.drop(index=val_indices, inplace=True)
self.val_y.drop(index=val_indices, inplace=True)
train_indices = self.train_X[self.train_X[c] == i].index
n_dropped += len(train_indices)
self.train_X.drop(index=train_indices, inplace=True)
self.train_y.drop(index=train_indices, inplace=True)
# self.train_X = self.train_X[self.train_X[c] != i]
self.logger.info(
f'{n_dropped} reviewers with value {outliers_count.to_list()} in {c} are dropped'
)
continue
elif not outliers_proportion.empty:
for i in outliers_proportion:
val_indices = self.val_X[self.val_X[c] == i].index
n_dropped += len(val_indices)
self.val_X.drop(index=val_indices, inplace=True)
self.val_y.drop(index=val_indices, inplace=True)
train_indices = self.train_X[self.train_X[c] == i].index
n_dropped += len(train_indices)
self.train_X.drop(index=train_indices, inplace=True)
self.train_y.drop(index=train_indices, inplace=True)
self.logger.info(
f'{n_dropped} reviewers with value {outliers_proportion.to_list()} in {c} are dropped'
)
self.logger.info(f'N in train_X is now {len(self.train_X)}, N in val_X is now {len(self.val_X)}')
self.logger.info(f'Distribution of the labels:'),
self.logger.info(f'train_y = 1: {sum(self.train_y)}, 0: {len(self.train_y[self.train_y ==0])}. ')
self.logger.info(f'There are {sum(self.train_y)/len(self.train_y)*100}% label 1 in train_y',)
self.logger.info(f'val_y = 1: {sum(self.val_y)}, 0: {len(self.val_y[self.val_y ==0])}.',)
self.logger.info(f'There are {sum(self.val_y)/len(self.val_y)*100}% label 1 in val_y')
def one_hot_encoding(self, col: str):
"""
one-hot encode a feature given col,
rename the columns after one-hot encoding given the categories in col,
and return training and validation sets as pd.DataFrame
"""
cols_cat = self.data.groupby([col], as_index=False).count()[col]
if re.match("^.*_color$", col):
color = re.split("_color", col)[0]
cols_cat = cols_cat.str.cat(pd.Series([color] * len(self.data)), sep='_')
# for column names after one-hot encoding
enc_rest = OneHotEncoder(sparse=False, handle_unknown='ignore')
enc_rest = enc_rest.fit(self.train_X[[col]])
train_X_transform = enc_rest.transform(self.train_X[[col]])
# saving the encoder for prediction
with open(f'models/{self.product_name}/encoder_{self.product_name}_{col}.pickle', 'wb') as f:
pickle.dump(enc_rest, f)
val_X_transform = enc_rest.transform(self.val_X[[col]])
train_X_transform = pd.DataFrame(train_X_transform)
col_names_dict = dict()
for col_idx in train_X_transform.columns:
col_names_dict[col_idx] = cols_cat[col_idx]
# for naming the encoded columns
with open(f'models/{self.product_name}/col_names_{self.product_name}_{col}.pickle', 'wb') as f:
pickle.dump(col_names_dict, f)
train_X_transform.rename(columns=col_names_dict, inplace=True)
val_X_transform = pd.DataFrame(val_X_transform)
col_names_dict = dict()
for col_idx in val_X_transform.columns:
col_names_dict[col_idx] = cols_cat[col_idx]
val_X_transform.rename(columns=col_names_dict, inplace=True)
return train_X_transform, val_X_transform
def cross_one_hot_features(
self,
one_hot_col1: pd.DataFrame,
one_hot_col2: pd.DataFrame,
data: pd.DataFrame,
col1: str = 'skin_type',
col2: str = 'skin_tone'
):
"""
given one-hot encoded feature 1 (one_hot_col1) and one-hot encoded feature 2 (one_hot_col1),
returns a dataframe with crossed feature b/w one_hot_col1 and one_hot_col2
columns of the returned dataframe are named by each of the crossed categories in col1 and col2 as "col1_col2"
"""
total_col1_cat = one_hot_col1.columns.to_list()
total_col2_cat = one_hot_col2.columns.to_list()
data_cross = pd.DataFrame()
i = 0 # col1
while i <= len(total_col1_cat) - 1:
j = 0 # col2
while j <= len(total_col2_cat) - 1:
col1_cat = total_col1_cat[i]
col2_cat = total_col2_cat[j]
new_cross = one_hot_col1[col1_cat] * one_hot_col2[col2_cat]
new_cross = pd.Series(new_cross)
data_cross[f'{col1_cat}_{col2_cat}'] = new_cross
j += 1
i += 1
return data_cross.dropna(axis=0)
def concat_all_features(
self,
train_X_transformed: pd.DataFrame,
val_X_transformed: pd.DataFrame
):
"""
concatenate non-one-hot-encoded features with one-hot-encoded ones
return training and validation sets
"""
for i in self.other_features:
train_X_transformed[i] = self.train_X[i].reset_index(drop=True)
val_X_transformed[i] = self.val_X[i].reset_index(drop=True)
return train_X_transformed, val_X_transformed
def feature_engineering(self):
"""
put together all the steps to engineer features
"""
self.skin_tone_one_hot_train, self.skin_tone_one_hot_val = self.one_hot_encoding(col='skin_tone')
self.skin_type_one_hot_train, self.skin_type_one_hot_val = self.one_hot_encoding(col='skin_type')
self.tone_type_cross_train = self.cross_one_hot_features(
one_hot_col1=self.skin_type_one_hot_train,
one_hot_col2=self.skin_tone_one_hot_train,
data=self.train_X
)
self.tone_type_cross_val = self.cross_one_hot_features(
one_hot_col1=self.skin_type_one_hot_val,
one_hot_col2=self.skin_tone_one_hot_val,
data=self.train_X
# need to use train_X for categories, in case that val_X.groupby([col]) skips categories with zero counts
)
self.hair_one_hot_train, self.hair_one_hot_val = self.one_hot_encoding(col='hair_color')
self.eye_one_hot_train, self.eye_one_hot_val = self.one_hot_encoding(col='eye_color')
self.train_X_transformed = pd.concat([
self.skin_tone_one_hot_train,
self.skin_type_one_hot_train,
self.tone_type_cross_train,
self.hair_one_hot_train,
self.eye_one_hot_train
], axis=1)
self.val_X_transformed = pd.concat([
self.skin_tone_one_hot_val,
self.skin_type_one_hot_val,
self.tone_type_cross_val,
self.hair_one_hot_val,
self.eye_one_hot_val
], axis=1)
self.train_X_transformed, self.val_X_transformed = self.concat_all_features(self.train_X_transformed,
self.val_X_transformed)
return None
### for model training ###
def grid_search(self):
evallist = [(self.val_X_transformed, self.val_y), (self.train_X_transformed, self.train_y)]
xgbmodel = xgb.XGBClassifier(
objective='binary:logistic',
use_label_encoder=False,
)
self.param_grid = {
"n_estimators": [5, 7, 9, 11],
"max_depth": [2, 3, 4, 5],
"learning_rate": [0.1, 0.01, 0.05],
"gamma": [0, 0.25, 1],
"reg_alpha": [0, 0.5, 1],
"scale_pos_weight": [0, 1 - (sum(self.y) / len(self.y)), (len(self.y) - sum(self.y)) / sum(self.y)]
}
self.grid_cv = GridSearchCV(
xgbmodel, self.param_grid,
n_jobs=4, cv=3,
scoring='roc_auc', verbose=1
)
_ = self.grid_cv.fit(self.train_X_transformed, self.train_y, eval_set=evallist,
eval_metric='auc')
self.logger.info('########### training xgbclassifier with grid search ###########')
self.logger.info(f'Best hyperparameters are {self.grid_cv.best_params_}')
self.logger.info(f'Best ROC-AUC {self.grid_cv.best_score_}')
def train_xgb_classifier(self, ):
self.best_model = xgb.XGBClassifier(
**self.grid_cv.best_params_,
objective='binary:logistic'
).fit(self.train_X_transformed, self.train_y)
for key, value in self.grid_cv.best_params_.items():
self.param_grid[key] = [value]
create_dir_for_product_if_not_existent(product=self.product_name, plots_or_models='models')
self.best_model.save_model(f'models/{self.product_name}/{self.product_name}_xgb.model')
def thresholding(
self
):
self.predict_y = self.best_model.predict_proba(self.val_X_transformed)[:, 1]
self.metrics_dict = {
'ratio_positives': [],
'accuracy_scores': [],
'precision_scores': [],
'recall_scores': [],
'f1_scores': []
}
thresholds = np.linspace(0, 1, 101)
for thresh in thresholds:
predict_y_binary = np.where(self.predict_y >= thresh, 1, 0)
self.metrics_dict['ratio_positives'].append(np.sum(predict_y_binary) / predict_y_binary.shape[0])
self.metrics_dict['accuracy_scores'].append(accuracy_score(self.val_y, predict_y_binary))
self.metrics_dict['precision_scores'].append(precision_score(self.val_y, predict_y_binary))
self.metrics_dict['recall_scores'].append(recall_score(self.val_y, predict_y_binary))
self.metrics_dict['f1_scores'].append(f1_score(self.val_y, predict_y_binary))
self.logger.info('#### Thresholding ####')
best_f1 = np.max(self.metrics_dict['f1_scores'])
best_f1_threshold = thresholds[np.argmax(self.metrics_dict['f1_scores'])]
self.logger.info(f"The best f1 score is {best_f1} which occurs at threshold {best_f1_threshold}")
best_precision = np.max(self.metrics_dict['precision_scores'])
best_precision_threshold = thresholds[np.argmax(self.metrics_dict['precision_scores'])]
self.logger.info(
f"The best precision score is {best_precision} which occurs at threshold {best_precision_threshold}"
)
best_recall = np.max(self.metrics_dict['recall_scores'])
best_recall_threshold = thresholds[np.argmax(self.metrics_dict['recall_scores'])]
self.logger.info(
f"The best recall score is {best_recall} which occurs at threshold {best_recall_threshold}"
)
if best_f1_threshold == 0:
self.best_threshold = max(best_recall_threshold, best_precision_threshold)
else:
self.best_threshold = best_f1_threshold
return self.best_threshold
def plot_auc_roc(
self,
# predict_y,
val_y,
product: str,
model: str,
filename: str = "auc_roc"):
"""
given predicted probability (predict_y) and label (val_y), computes AUC-ROC across threshold and saves the figure by product and model
"""
thresholds = np.linspace(0, 1, 10)
fpr = []
tpr = []
for threshold in thresholds:
predict_y_binary = np.where(self.predict_y >= threshold, 1, 0)
fp = np.sum((val_y == 0) & (predict_y_binary == 1)) # true value is 0 but predict to be 1
tp = np.sum((val_y == 1) & (predict_y_binary == 1)) # true value is 1 & predict to be 1
fn = np.sum((val_y == 1) & (predict_y_binary == 0)) # true value is 1 but predict to be 0
tn = np.sum((val_y == 0) & (predict_y_binary == 0)) # true value is 0 but predict to be 0
fpr.append(fp / (fp + tn))
tpr.append(tp / (tp + fn))
## ROC & ROC-AUC
roc_auc_ = round(auc(fpr, tpr), 5)
fig3, ax3 = plt.subplots(1, 1)
ax3.plot(fpr, tpr, label="ROC")
ax3.plot([0, 1], [0, 1], 'k--')
ax3.set_xlabel("False Positive Rate")
ax3.set_ylabel("True Positive Rate")
ax3.text(0.55, 0.2, 'AUC = {}'.format(roc_auc_))
product_title = product.replace('_', ' ')
plt.title(f'{model} with {product_title}')
create_dir_for_product_if_not_existent(product=product, plots_or_models='plots')
plt.savefig(f'plots/{product}/{model}_{filename}.jpeg')
def plot_precision_recall_f1(
self,
# predict_y,
val_y,
model: str,
product: str,
filename: str = "precision_recall_f1"):
"""
given predicted probability (predict_y) and label (val_y), plots precision, recall, and F1 scores across threshold
and saves the figure by product and model
"""
thresholds = np.linspace(0, 1, 101)
precision_at_p5 = precision_score(val_y, np.where(self.predict_y >= 0.5, 1, 0))
recall_at_p5 = recall_score(val_y, np.where(self.predict_y >= 0.5, 1, 0))
f1_score_at_p5 = f1_score(val_y, np.where(self.predict_y >= 0.5, 1, 0))
### plot precisions, recalls, and F1 scores across threshold ###
fig2, ax2 = plt.subplots(1, 1)
ax2.plot(thresholds, self.metrics_dict['precision_scores'], label="precision")
ax2.plot(thresholds, self.metrics_dict['recall_scores'], label="recall")
ax2.plot(thresholds, self.metrics_dict['f1_scores'], label="f1 score")
ax2.axvline(0.5, linestyle=':', color='r')
ax2.plot(0.5, precision_at_p5, "s")
ax2.annotate("{}".format(round(precision_at_p5, 3)), (0.5, precision_at_p5))
ax2.plot(0.5, recall_at_p5, "s")
ax2.annotate("{}".format(round(recall_at_p5, 3)), (0.5, recall_at_p5))
ax2.plot(0.5, f1_score_at_p5, "s")
ax2.annotate("{}".format(round(f1_score_at_p5, 3)), (0.5, f1_score_at_p5))
ax2.legend()
ax2.set_xlabel("Thresholds")
product_title = product.replace('_', ' ')
plt.title(f'{model} with {product_title}')
create_dir_for_product_if_not_existent(product=product, plots_or_models='plots')
plt.savefig(f'plots/{product}/{model}_{filename}.jpeg')
def plot_predictions_by_scores(
self,
# predict_y,
val_y,
product: str,
model: str,
bins: int = 20,
filename: str = 'predictions_by_scores'):
"""
given predicted probability (predict_y) and label (val_y), plots positive and negative instances by predicted probability
green: label = 1; red: label = 0
"""
fig1, ax1 = plt.subplots(1, 1)
ax1.hist(self.predict_y[val_y == 1.0], bins=bins, color='g')
ax1.hist(self.predict_y[val_y == 0.0], bins=bins, color='r')
product_title = product.replace('_', ' ')
plt.title(f'{model} with {product_title}')
# plt.show()
create_dir_for_product_if_not_existent(product=product, plots_or_models='plots')
plt.savefig(f'plots/{product}/{model}_{filename}.jpeg')
def plot_countbar(data: pd.DataFrame, product: str, col1: str):
"""
given a product, return and save a countplot with column "col1" in data
"""
plt.figure(figsize=(10, 8))
sns.countplot(x=data[col1], order=data[col1].value_counts().index)
plt.xticks(rotation=90)
plt.xlabel(f'Reviewers by {col1}')
plt.title(product.replace('_', ' '))
create_dir_for_product_if_not_existent(product=product, plots_or_models='plots')
plt.savefig(f'plots/{product}/{col1}_countbar.jpeg')
def plot_line(data: pd.DataFrame, product: str, col1: str):
"""
given a product, return and save a line plot with continuous var "col1" in data
"""
plt.plot(data[col1])
plt.title(product.replace('_', ' '))
create_dir_for_product_if_not_existent(product=product, plots_or_models='plots')
plt.savefig(f'plots/{product}/{col1}_lineplot.jpeg')
def plot_diff_in_means(data: pd.DataFrame, product: str, col1: str, col2: str = 'rating'):
"""
given a product, plots difference in means across groups and saves the figure
col1
categorical data with groups
col2
continuous data for the means
"""
data_groupby = data.groupby(col1, as_index=False)[col2]
data_groupby_labels = data_groupby.agg('mean').sort_values(col2, ascending=False)
data_agg = pd.DataFrame(data_groupby_labels).rename(columns={col2: 'mean'})
n = data_groupby.count()
data_agg = data_agg.merge(n, how='left', right_on=col1, left_on=col1).rename(columns={col2: 'n'})
std = data_groupby.agg(np.std)
data_agg = data_agg.merge(std, how='left', right_on=col1, left_on=col1).rename(columns={col2: 'std'})
data_agg['se'] = data_agg['std'] / np.sqrt(data_agg['n'])
data_agg['lower'] = st.t.interval(alpha=0.95, df=data_agg['n'] - 1, loc=data_agg['mean'], scale=data_agg['se'])[0]
data_agg['upper'] = st.t.interval(alpha=0.95, df=data_agg['n'] - 1, loc=data_agg['mean'], scale=data_agg['se'])[1]
for upper, mean, lower, y in zip(data_agg['upper'], data_agg['mean'], data_agg['lower'], data_agg[col1]):
plt.plot((lower, mean, upper), (y, y, y), 'b.-')
plt.yticks(range(len(n)), list(data_agg[col1]))
create_dir_for_product_if_not_existent(product=product, plots_or_models='plots')
plt.savefig(f'plots/{product}/{col1}_{col2}_diff_in_means.jpeg')
def plot_cross_tab_heatmap(data: pd.DataFrame, product: str, col1: str, col2: str, col3: str, func: str = 'mean'):
"""
given a product, cross-tabulate col1 and col2 in data, and create a heatmap given func with col3 as the color bar
col1
categorical data in x axis
col2
categorical data in y axis
col3
continuous data in color bar
func
'mean' or 'median' for aggfunc in pd.crosstab
"""
col1_col2_crosstab = pd.crosstab(data[col1], data[col2], values=data[col3], aggfunc=func)
col1_labels = data.groupby(col1, as_index=False).count()[col1]
col2_labels = data.groupby(col2, as_index=False).count()[col2]
fig, ax = plt.subplots(figsize=(10, 8))
im = ax.imshow(col1_col2_crosstab)
ax.set_yticks(np.arange(len(col1_labels)), labels=col1_labels)
ax.set_xticks(np.arange(len(col2_labels)), labels=col2_labels)
cbar = ax.figure.colorbar(im, ax=ax)
cbar.ax.set_ylabel(col3, rotation=-90, va="bottom")
kw = dict(horizontalalignment="center",
verticalalignment="center")
# threshold = im.norm(col1_col2_crosstab.max())/2
texts = []
textcolors = ("black", "white")
for i in range(col1_col2_crosstab.shape[0]):
for j in range(col1_col2_crosstab.shape[1]):
kw.update(color=textcolors[int(im.norm(col1_col2_crosstab.iloc[i, j]) < 0.5)])
text = im.axes.text(j, i, round(col1_col2_crosstab.iloc[i, j], 2), kw)
texts.append(text)
plt.title(product.replace('_', ' '))
create_dir_for_product_if_not_existent(product=product, plots_or_models='plots')
plt.savefig(f'plots/{product}/{col1}_{col2}_crosstab_heatmap.jpeg')
def plot_auc_roc(predict_y, val_y, product: str, model: str, filename: str = "auc_roc"):
"""
given predicted probability (predict_y) and label (val_y), computes AUC-ROC across threshold and saves the figure by product and model
"""
thresholds = np.linspace(0, 1, 10)
fpr = []
tpr = []
for threshold in thresholds:
predict_y_binary = np.where(predict_y >= threshold, 1, 0)
fp = np.sum((val_y == 0) & (predict_y_binary == 1)) # true value is 0 but predict to be 1
tp = np.sum((val_y == 1) & (predict_y_binary == 1)) # true value is 1 & predict to be 1
fn = np.sum((val_y == 1) & (predict_y_binary == 0)) # true value is 1 but predict to be 0
tn = np.sum((val_y == 0) & (predict_y_binary == 0)) # true value is 0 but predict to be 0
fpr.append(fp / (fp + tn))
tpr.append(tp / (tp + fn))
## ROC & ROC-AUC
roc_auc_ = round(auc(fpr, tpr), 5)
fig, ax3 = plt.subplots(1, 1)
ax3.plot(fpr, tpr, label="ROC")
ax3.plot([0, 1], [0, 1], 'k--')
ax3.set_xlabel("False Positive Rate")
ax3.set_ylabel("True Positive Rate")
ax3.text(0.55, 0.2, 'AUC = {}'.format(roc_auc_))
product_title = product.replace('_', ' ')
plt.title(f'{model} with {product_title}')
create_dir_for_product_if_not_existent(product=product, plots_or_models='plots')
plt.savefig(f'plots/{product}/{model}_{filename}.jpeg')
def plot_recision_recall_f1(predict_y, val_y, product: str, model: str, filename: str = "precision_recall_f1"):
"""
given predicted probability (predict_y) and label (val_y), plots precision, recall, and F1 scores across threshold
and saves the figure by product and model
"""
thresholds = np.linspace(0, 1, 10)
precisions = []
recalls = []
f1_scores = []
for threshold in thresholds:
predict_y_binary = np.where(predict_y >= threshold, 1, 0)
precisions.append(precision_score(val_y, predict_y_binary))
recalls.append(recall_score(val_y, predict_y_binary))
f1_scores.append(f1_score(val_y, predict_y_binary))
precision_at_threshold = precision_score(val_y, np.where(predict_y >= 0.5, 1, 0))
recall_at_threshold = recall_score(val_y, np.where(predict_y >= 0.5, 1, 0))
f1_score_at_threshold = f1_score(val_y, np.where(predict_y >= 0.5, 1, 0))
### plot precisions, recalls, and F1 scores across threshold
fig, ax2 = plt.subplots(1, 1)
ax2.plot(thresholds, precisions, label="precision")
ax2.plot(thresholds, recalls, label="recall")
ax2.plot(thresholds, f1_scores, label="f1 score")
ax2.axvline(0.5, linestyle=':', color='r')
ax2.plot(0.5, precision_at_threshold, "s")
ax2.annotate("{}".format(round(precision_at_threshold, 3)), (0.5, precision_at_threshold))
ax2.plot(0.5, recall_at_threshold, "s")
ax2.annotate("{}".format(round(recall_at_threshold, 3)), (0.5, recall_at_threshold))
ax2.plot(0.5, f1_score_at_threshold, "s")
ax2.annotate("{}".format(round(f1_score_at_threshold, 3)), (0.5, f1_score_at_threshold))
ax2.legend()
ax2.set_xlabel("Thresholds")
product_title = product.replace('_', ' ')
plt.title(f'{model} with {product_title}')
create_dir_for_product_if_not_existent(product=product, plots_or_models='plots')
plt.savefig(f'plots/{product}/{model}_{filename}.jpeg')
def plot_predictions_by_scores(predict_y, val_y, product: str, model: str, bins: int = 20,
filename: str = 'predictions_by_scores'):
"""
given predicted probability (predict_y) and label (val_y), plots positive and negative instances by predicted probability
green: label = 1; red: label = 0
"""
plt.hist(predict_y[val_y == 1.0], bins=bins, color='g')
plt.hist(predict_y[val_y == 0.0], bins=bins, color='r')
product_title = product.replace('_', ' ')
plt.title(f'{model} with {product_title}')
plt.show()
create_dir_for_product_if_not_existent(product=product, plots_or_models='plots')
plt.savefig(f'plots/{product}/{model}_{filename}.jpeg')