-
Notifications
You must be signed in to change notification settings - Fork 2
/
FBClassifier.py
538 lines (452 loc) · 19.8 KB
/
FBClassifier.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
from math import nan
from matplotlib.pyplot import sca
import numpy as np
import pandas as pd
from pandas.core.frame import DataFrame
from sklearn.preprocessing import StandardScaler
from sklearn.preprocessing import RobustScaler
from sklearn.preprocessing import MinMaxScaler
from sklearn.preprocessing import MaxAbsScaler
from sklearn.tree import DecisionTreeClassifier
from sklearn.model_selection import KFold
from sklearn.model_selection import cross_val_score
import random
import copy
import matplotlib.pyplot as plt
from sklearn.model_selection import train_test_split
from sklearn.metrics import roc_curve, roc_auc_score, auc, classification_report
from sklearn.multiclass import OneVsRestClassifier
from sklearn.preprocessing import label_binarize
def brute_force(
X: DataFrame,
y: DataFrame,
scalers=[StandardScaler(), RobustScaler(), MinMaxScaler(), MaxAbsScaler()],
models=[
DecisionTreeClassifier(criterion="gini"), DecisionTreeClassifier(criterion="entropy"),
],
cv_k=[2, 3, 4, 5, 6, 7, 8, 9, 10],
is_cv_shuffle=True,
):
"""
Brute Force Search
----------
- Find the best parameter what has the best score.
- This function use `Brute Force` method with memoization
Parameters
----------
- `X`: pandas.DataFrame
- training dataset.
- `y`: pandas.DataFrame
- target value.
- `scalers`: array
- Scaler functions to scale data. This can be modified by user.
- StandardScaler, RobustScaler, MinMaxScaler, MaxAbsScaler as default.
- `models`: array
- Model functions to fitting data and prediction. This can be modified by user.
- DecisionTreeClassifier(gini, entropy) as default with hyperparameters.
- `cv_k`: array
- Cross validation parameter. Default value is [2,3,4,5,6,7,8,9,10].
- `is_cv_shuffle`
- To set shuffle or not in cross validation
Returns
----------
- `best_params`: dictionary type of results.
- `best_scaler`: Scaler what has best score.
- `best_model`: Model what has best score.
- `best_cv_k`: k value in K-fold CV what has best score.
- `best_score`: double
- Represent the score of the `best_params`.
"""
# Initialize variables
maxScore = -1.0
best_scaler = None
best_model = None
best_cv_k_ = None
# Find best scaler
for n in range(0, len(scalers)):
X = scalers[n].fit_transform(X)
# Find best model
for m in range(0, len(models)):
# Find best k value of CV
for i in range(0, len(cv_k)):
kfold = KFold(n_splits=cv_k[i], shuffle=is_cv_shuffle)
score_result = cross_val_score(models[m], X, y, scoring="accuracy", cv=kfold)
# if mean value of scores are bigger than max variable,
# update new options(model, scaler, k) to best options
if maxScore < score_result.mean():
maxScore = score_result.mean()
best_scaler = copy.deepcopy(scalers[n])
best_model = copy.deepcopy(models[m])
best_cv_k_ = copy.deepcopy(cv_k[i])
class res:
best_params = {}
res.best_params = {
'best_scaler': best_scaler,
'best_model': best_model,
'best_cv_k': best_cv_k_,
}
res.best_scaler = best_scaler
res.best_model = best_model
res.best_k = best_cv_k_
res.best_score = maxScore
# Return value with dictionary type
return res
def plot_roc_curve(X, y, model, title):
# for binary target
if len(y.unique()) == 2:
# Calculate False Positive Rate, True Positive Rate
X = model.best_scaler.fit_transform(X)
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.3, random_state=42)
clf = model.best_model
y_pred = clf.fit(X_train, y_train).predict_proba(X_test)
fpr, tpr, _ = roc_curve(y_test, y_pred[:, 1])
roc_auc = roc_auc_score(y_test, y_pred[:, 1])
# Plot result
plt.figure(figsize=(12, 10))
plt.plot(fpr, tpr, color='b', label='ROC curve (area = %0.2f)' % roc_auc)
plt.plot([0, 1], [0, 1], color='r', linestyle='--')
plt.xlabel('False Positive Rate')
plt.ylabel('True Positive Rate')
plt.title('ROC Curve for ' + str(title), fontsize=20)
plt.legend()
plt.show()
# for multiclass target
else:
# Calculate False Positive Rate, True Positive Rate
X = model.best_scaler.fit_transform(X)
y_unique, counts = np.unique(y, return_counts=True)
y = label_binarize(y, classes=y_unique)
n_classes = y.shape[1]
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.3, random_state=42)
clf = OneVsRestClassifier(model.best_model)
y_pred = clf.fit(X_train, y_train).predict_proba(X_test)
# Compute ROC curve and ROC area for each class
fpr = dict()
tpr = dict()
roc_auc = dict()
for i in range(n_classes):
fpr[i], tpr[i], _ = roc_curve(y_test[:, i], y_pred[:, i])
# First aggregate all false positive rates
all_fpr = np.unique(np.concatenate([fpr[i] for i in range(n_classes)]))
# Then interpolate all ROC curves at this points
mean_tpr = np.zeros_like(all_fpr)
for i in range(n_classes):
mean_tpr += np.interp(all_fpr, fpr[i], tpr[i])
# Finally average it and compute AUC
mean_tpr /= n_classes
fpr["macro"] = all_fpr
tpr["macro"] = mean_tpr
weighted_roc_auc = roc_auc_score(y_test, y_pred, multi_class="ovr", average="weighted")
# Plot result
plt.figure(figsize=(12, 10))
plt.plot(fpr[0], tpr[0], linestyle='--', color='orange', label='Class 0 vs Rest')
plt.plot(fpr[1], tpr[1], linestyle='--', color='green', label='Class 1 vs Rest')
plt.plot(fpr[2], tpr[2], linestyle='--', color='cyan', label='Class 2 vs Rest')
plt.plot(fpr[3], tpr[3], linestyle='--', color='yellow', label='Class 3 vs Rest')
plt.plot(fpr["macro"], tpr["macro"], color='r', label='ROC curve (area = %0.2f)' % weighted_roc_auc)
plt.plot([0, 1], [0, 1], color='black')
plt.xlabel('False Positive Rate')
plt.ylabel('True Positive Rate')
plt.title('ROC Curve for ' + str(title), fontsize=20)
plt.legend()
plt.show()
def clf_report(X, y, model):
X = model.best_scaler.fit_transform(X)
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.3, random_state=42)
clf = model.best_model
clf.fit(X_train, y_train)
pred_test = clf.predict(X_test)
report = classification_report(y_test, pred_test, zero_division=0)
return report
def random_search(
X: DataFrame,
y: DataFrame,
scalers=[StandardScaler(), RobustScaler(), MinMaxScaler(), MaxAbsScaler()],
models=[
DecisionTreeClassifier(criterion="gini"), DecisionTreeClassifier(criterion="entropy"),
],
cv_k=[2, 3, 4, 5, 6, 7, 8, 9, 10],
is_cv_shuffle=True,
thresh_score=None,
max_iter=50,
):
"""
Random Search
----------
- Find the best parameter what has the best score.
- This function use `Random Search` method with memoization
Parameters
----------
- `X`: pandas.DataFrame
- training dataset.
- `y`: pandas.DataFrame
- target value.
- `scalers`: array
- Scaler functions to scale data. This can be modified by user.
- StandardScaler, RobustScaler, MinMaxScaler, MaxAbsScaler as default.
- `models`: array
- Model functions to fitting data and prediction. This can be modified by user.
- DecisionTreeClassifier(gini, entropy) as default with hyperparameters.
- `cv_k`: array
- Cross validation parameter. Default value is [2,3,4,5,6,7,8,9,10].
- `is_cv_shuffle`
- To set shuffle or not in cross validation
- `thresh_score`
- Default is None. If, algorithm find the score what is higher than thresh_score, then stop and terminate searching.
- `max_iter`
- Default is 100. This is meaning that how many iterations in searching loop.
Returns
----------
- `best_params_`: dictionary type of results.
- `best_scaler`: Scaler what has best score.
- `best_model`: Model what has best score.
- `best_cv_k`: k value in K-fold CV what has best score.
- `best_score`: double
- Represent the score of the `best_params`.
"""
# 0. Calculate length of each paramenter
scalers_len = len(scalers)
models_len = len(models)
cv_k_len = len(cv_k)
# 0. Initialize max score point
max_scalers_idx = 0
max_models_idx = 0
max_cv_k_idx = 0
max_score = 0
# 0. Create memorize table for memoization
mem_table = [[[0 for col in range(cv_k_len)] for row in range(models_len)] for col in range(scalers_len)]
for trial in range(0, max_iter):
# 0. Pick arbitrary point (theta1 = p1)
scalers_idx = random.randrange(0, scalers_len)
models_idx = random.randrange(0, models_len)
cv_k_idx = random.randrange(0, cv_k_len)
# 2. Calculate score(J(theta)) of each theta(point)
score = 0
# Check mem_table if score already has been calculated
if mem_table[scalers_idx][models_idx][cv_k_idx] != 0:
score = mem_table[scalers_idx][models_idx][cv_k_idx]
else:
# if not, calculate score of theta
if scalers[scalers_idx] != None:
p1_X = scalers[scalers_idx].fit_transform(X)
else:
p1_X = X
kfold = KFold(n_splits=cv_k[cv_k_idx], shuffle=is_cv_shuffle)
score = cross_val_score(models[models_idx], p1_X, y, cv=kfold).mean()
# 2-1. Memoization
mem_table[scalers_idx][models_idx][cv_k_idx] = score
# Save point parameter what have best score
if max_score < score:
max_scalers_idx = scalers_idx
max_models_idx = models_idx
max_cv_k_idx = cv_k_idx
max_score = score
# If, score get higher score than thresh, terminate gradient searching
if thresh_score != None and max_score > thresh_score: break
# print("Trial: ", end="")
# print(trial)
# print(p1.scalers_idx)
# print(p1.models_idx)
# print(p1.cv_k_idx)
# print()
class res:
best_params = {}
res.best_params = {
'best_scaler': scalers[max_scalers_idx],
'best_model': models[max_models_idx],
'best_cv_k': cv_k[max_cv_k_idx],
}
res.best_scaler = scalers[max_scalers_idx]
res.best_model = models[max_models_idx]
res.best_k = cv_k[max_cv_k_idx]
res.best_score = max_score
# Return value with dictionary type
return res
def auto_ml(
X:DataFrame,
y:DataFrame,
scalers=[StandardScaler(), RobustScaler(), MinMaxScaler(), MaxAbsScaler()],
models=[
DecisionTreeClassifier(criterion="gini"), DecisionTreeClassifier(criterion="entropy"),
],
cv_k=[2,3,4,5,6,7,8,9,10],
is_cv_shuffle = True,
thresh_score = None,
max_iter = 50,
):
"""
Auto ML for Classifier
----------
- Find the best parameter what has the best score.
- This function use `Auto ML` method. This is similar to the Gradient Descent.
- This function use memoization technique for faster calculation.
Parameters
----------
- `X`: pandas.DataFrame
- training dataset.
- `y`: pandas.DataFrame
- target value.
- `scalers`: array
- Scaler functions to scale data. This can be modified by user.
- StandardScaler, RobustScaler, MinMaxScaler, MaxAbsScaler as default.
- `models`: array
- Model functions to fitting data and prediction. This can be modified by user.
- DecisionTreeClassifier(gini, entropy) as default with hyperparameters.
- `cv_k`: array
- Cross validation parameter. Default value is [2,3,4,5,6,7,8,9,10].
- `is_cv_shuffle`
- To set shuffle or not in cross validation
- `thresh_score`
- Default is None. If, algorithm find the score what is higher than thresh_score, then stop and terminate searching.
- `max_iter`
- Default is 50. This is meaning that how many iterations in searching loop.
Returns
----------
- `best_params`: dictionary type of results.
- `best_scaler`: Scaler what has best score.
- `best_model`: Model what has best score.
- `best_cv_k`: k value in K-fold CV what has best score.
- `best_score`: double
- Represent the score of the `best_params`.
"""
logo()
# 0. Calculate length of each paramenter
scalers_len = len(scalers)
models_len = len(models)
cv_k_len = len(cv_k)
# 0. Create memorize table for memoization
mem_table = [[[0 for col in range(cv_k_len)] for row in range(models_len)] for col in range(scalers_len)]
# 0. Create point(theta) vector class
class Point():
scalers_idx = 0
models_idx = 0
cv_k_idx = 0
# 0. Initialize gradient value
gradient_theta1 = 0
gradient_theta2 = 0
gradient_theta3 = 0
# 0. Pick arbitrary point (theta1 = p1)
p1 = Point()
p1.scalers_idx = random.randrange(0, scalers_len)
p1.models_idx = random.randrange(0, models_len)
p1.cv_k_idx = random.randrange(0, cv_k_len)
# 0. Initialize max score point
max_scalers_idx = 0
max_models_idx = 0
max_cv_k_idx = 0
max_score = 0
for trial in range(0, max_iter):
# 1. Check previous gradient value of each theta
# and pick another arbitrary point (theta = p2)
def check_gradient(target_gradient_theta, point_val, max_len):
result = 0
if target_gradient_theta > 0 and point_val + 1 != max_len:
# if point_val+1 == max_len => out of range
# then, get arbitrary point from 0 to len(target)
result = random.randrange(point_val + 1, max_len)
elif target_gradient_theta < 0 and point_val != 0:
# if point_val == 0 => out of range
# then, get arbitrary point from 0 to len(target)
result = random.randrange(0, point_val)
else:
result = random.randrange(0, max_len)
return result
p2 = Point()
p2.scalers_idx = check_gradient(gradient_theta1, p1.scalers_idx, scalers_len)
p2.models_idx = check_gradient(gradient_theta2, p1.models_idx, models_len)
p2.cv_k_idx = check_gradient(gradient_theta3, p1.cv_k_idx, cv_k_len)
# 2. Calculate score(J(theta)) of each theta(point)
p1_score = 0
p2_score = 0
# Check mem_table if score already has been calculated
if mem_table[p1.scalers_idx][p1.models_idx][p1.cv_k_idx] != 0:
p1_score = mem_table[p1.scalers_idx][p1.models_idx][p1.cv_k_idx]
else:
# if not, calculate score of theta
if scalers[p1.scalers_idx] != None:
p1_X = scalers[p1.scalers_idx].fit_transform(X)
else:
p1_X = X
kfold = KFold(n_splits=cv_k[p1.cv_k_idx], shuffle=is_cv_shuffle)
p1_score = cross_val_score(models[p1.models_idx], p1_X, y, cv=kfold).mean()
# 2-1. Memoization
mem_table[p1.scalers_idx][p1.models_idx][p1.cv_k_idx] = p1_score
if mem_table[p2.scalers_idx][p2.models_idx][p2.cv_k_idx] != 0:
p2_score = mem_table[p2.scalers_idx][p2.models_idx][p2.cv_k_idx]
else:
if scalers[p1.scalers_idx] != None:
p2_X = scalers[p2.scalers_idx].fit_transform(X)
else:
p2_X = X
kfold = KFold(n_splits=cv_k[p2.cv_k_idx], shuffle=is_cv_shuffle)
p2_score = cross_val_score(models[p2.models_idx], p2_X, y, cv=kfold).mean()
# 2-1. Memoization
mem_table[p2.scalers_idx][p2.models_idx][p2.cv_k_idx] = p2_score
# Save point parameter what have best score
if p1_score > p2_score:
if max_score < p1_score:
max_scalers_idx = p1.scalers_idx
max_models_idx = p1.models_idx
max_cv_k_idx = p1.cv_k_idx
max_score = p1_score
if p1_score < p2_score:
if max_score < p2_score:
max_scalers_idx = p2.scalers_idx
max_models_idx = p2.models_idx
max_cv_k_idx = p2.cv_k_idx
max_score = p2_score
# If, score get higher score than thresh, terminate gradient searching
if thresh_score != None and max_score > thresh_score: break
# 3. Calcuate gradient of each theta(point).
# with using above theta value, set another theta(point).
change_of_cost = p2_score - p1_score
change_of_theta1 = p2.scalers_idx - p1.scalers_idx
change_of_theta2 = p2.models_idx - p1.models_idx
change_of_theta3 = p2.cv_k_idx - p1.cv_k_idx
# If, attribute of theta1 and theta2 are same, set gradient value to 0 (slope = 0)
def update_gradient_value(change_of_cost, change_of_theta):
result_gradient = 0
if change_of_theta != 0:
result_gradient = change_of_cost / change_of_theta
return result_gradient
gradient_theta1 = update_gradient_value(change_of_cost, change_of_theta1)
gradient_theta2 = update_gradient_value(change_of_cost, change_of_theta2)
gradient_theta3 = update_gradient_value(change_of_cost, change_of_theta3)
# 4. Prepare for next gradient (change theta 1 to new position)
def set_new_point(gradient_theta, compare1, compare2):
result_idx = 0
if gradient_theta > 0:
result_idx = max([compare1, compare2])
elif gradient_theta < 0:
result_idx = min([compare1, compare2])
else:
result_idx = compare1
return result_idx
# Set new theta1 for the next calculation
p1.scalers_idx = set_new_point(gradient_theta1, p1.scalers_idx, p2.scalers_idx)
p1.models_idx = set_new_point(gradient_theta2, p1.models_idx, p2.models_idx)
p1.cv_k_idx = set_new_point(gradient_theta3, p1.cv_k_idx, p2.cv_k_idx)
class res:
best_params = {}
res.best_params = {
'best_scaler': scalers[max_scalers_idx],
'best_model': models[max_models_idx],
'best_cv_k': cv_k[max_cv_k_idx],
}
res.best_scaler = scalers[max_scalers_idx]
res.best_model = models[max_models_idx]
res.best_k = cv_k[max_cv_k_idx]
res.best_score = max_score
# Return value with dictionary type
return res
def logo():
print("")
print(" /‾‾‾‾‾‾\ /‾‾‾\ /‾‾‾‾/\/‾‾‾‾‾‾‾‾‾‾‾\ /‾‾‾‾‾‾‾‾‾‾‾‾\ /‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾\ /‾‾‾‾\ ")
print(" / \ / /\ / / / /\/ /\ / /\/ /\ ")
print(" / /\ \ / / / / /\‾‾/ /\‾‾\ / /‾‾‾/ / / / /‾/ /‾/ / / / / ")
print(" / / / /\/ / / / / ‾/ / /‾‾‾/ / / / / / / / / / / / / / ")
print(" / ‾‾ / / /__/ / / / / / / /___/ / / / / / / / / / / / ")
print(" / /‾/ / / / / / / / / / / / / / / / / / ‾‾‾‾‾‾/\ ")
print(" /____/ /____/ /____________/ / /____/ / /_____________/ / /____/ /____/ /____/ /____________/ / ")
print(" \____\ \____\/\____________\/ \____\/ \_____________\/ \____\ \____\ \____\/ \___________\/ ")
print(" for Classifier / Version: 2021.11.17 ")