/
ResultsAnalysis.py
403 lines (319 loc) · 14.9 KB
/
ResultsAnalysis.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
import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
import os
import joblib
import torch
import torch.nn as nn
import torch.nn.functional as F
from torch.optim import Adam, SGD
import tqdm
import itertools
from sklearn.linear_model import LogisticRegression
from sklearn.neighbors import KNeighborsClassifier
from sklearn.ensemble import RandomForestClassifier
from sklearn.metrics import roc_auc_score, accuracy_score, f1_score
from sklearn.model_selection import cross_val_score, cross_val_predict, KFold, StratifiedKFold
from matplotlib.ticker import MultipleLocator, FormatStrFormatter, AutoMinorLocator
plt.interactive(True)
pd.options.display.max_columns = 15
pic_folder = 'pic/'
# version = 'hidden_size'
# version = 'test_11'
# version = 'window_size_0'
# version = 'time_step_0'
versions_list = [
# 'window_size_0',
# 'window_size_1',
# 'hidden_size_2',
# 'hidden_size_3',
# 'normalization_0',
# 'time_step_0',
# 'time_step_1',
# 'time_step_2',
# # 'time_step_1_60',
# 'february_27',
'march_1',
]
classic_versions_list = [
# 'window_size_0',
# 'time_step_1',
# 'time_step_2',
# 'time_step_1_60',
]
attention_names_dict = {
0: 'RNN',
1: 'RNN + Input Attention, Softmax',
2: 'RNN + Input Attention, Sigmoid',
4: 'RNN + Input Attention, [0, 1] clamp',
}
df_results = pd.DataFrame()
for version in versions_list:
df_results_new = pd.read_csv(f'data/df_results_{version}.csv')
df_results_new = df_results_new.loc[df_results_new['score_val'] != -1, :]
df_results_new.set_index(['time_step', 'window_size', 'batch_size', 'attention', 'hidden_size', 'normalization', 'n_repeat'], inplace=True)
# df_results_new = df_results_new[['score_test', 'score_val']]
df_results_new = df_results_new[['score_test', 'dumb_score_test', 'score_val', 'dumb_score_val']]
df_results = pd.concat([df_results, df_results_new])
for version in classic_versions_list:
df_results_classic = pd.read_csv(f'data/df_results_classic_{version}.csv')
df_results_classic = df_results_classic.loc[df_results_classic['alg_name'] != 'Random Forest 1']
df_results_classic = df_results_classic.loc[df_results_classic['time_step'] != 60]
df_results_classic = df_results_classic.loc[df_results_classic['window_size'] != 600]
df_results_classic['batch_size'] = -1
df_results_classic['hidden_size'] = -1
df_results_classic['normalization'] = -1
df_results_classic.rename(columns={'alg_name': 'attention'}, inplace=True)
df_results_classic.set_index(['time_step', 'window_size', 'batch_size', 'attention', 'hidden_size', 'normalization', 'n_repeat'], inplace=True)
df_results_classic.rename(columns={'score_val': 'score_test'}, inplace=True)
df_results_classic.rename(index={'Random Forest 0': 'Random Forest'}, inplace=True)
# df_results_classic.rename(index={'Logistic Regression': 'Random Forest'})
df_results = pd.concat([df_results, df_results_classic])
# df_results = df_results.join(df_results_classic)
def plot_dependency(df_results, labels, colors, labels_col='attention', dependency_col='hidden_size', label2text_dict=None,
xlabel='', suffix='v0', plot_errors=True):
df_grouped = df_results.groupby([labels_col, dependency_col])['score_test']
df_agg_mean = df_grouped.mean().reset_index()
df_agg_std = df_grouped.std().reset_index()
print(df_agg_mean)
print(df_agg_std)
# labels = np.unique(df_agg_mean[labels_col])
plt.figure(figsize=(13.5, 9))
# plt.figure(figsize=(20, 15))
for label, color in zip(labels, colors):
mask_label = df_agg_mean[labels_col] == label
dependency_values = df_agg_mean.loc[mask_label, dependency_col]
means4label = df_agg_mean.loc[mask_label, 'score_test'].values.ravel()
stds4label = df_agg_std.loc[mask_label, 'score_test'].values.ravel()
print(means4label)
print(stds4label)
if label2text_dict is not None:
if label in label2text_dict:
label_text = label2text_dict[label]
else:
label_text = label
else:
label_text = label
lower = means4label - stds4label
upper = means4label + stds4label
plt.plot(dependency_values, means4label, label=label_text, color=color, lw=5)
plt.scatter(dependency_values, means4label, marker='o', s=140, color=color)
if plot_errors:
plt.fill_between(dependency_values, lower, upper, alpha=0.1, color=color)
plt.tick_params(axis='both', which='major', labelsize=25, size=20)
plt.xticks()
# plt.xlabel('Window Size, s', fontsize=35)
plt.xlabel(xlabel, fontsize=35)
plt.ylabel('ROC AUC', fontsize=35)
plt.legend(fontsize=20) # 16 if not enough space # , loc='lower right')
# plt.xlim(97, 610) # UPDATE IF 60 IS ADDED!!!
plt.tight_layout()
plt.savefig(f'pic/{labels_col}_{dependency_col}_{suffix}.png')
plt.close()
def plot_dependency_time_step(df_results, labels, colors, labels_col='attention', dependency_col='hidden_size', label2text_dict=None,
xlabel='', suffix='v0', plot_errors=True):
df_grouped = df_results.groupby([labels_col, dependency_col])['score_test']
df_agg_mean = df_grouped.mean().reset_index()
df_agg_std = df_grouped.std().reset_index()
print(df_agg_mean)
print(df_agg_std)
# labels = np.unique(df_agg_mean[labels_col])
# plt.figure(figsize=(13.5, 9))
fig, ax = plt.subplots(figsize=(13.5, 9))
# plt.figure(figsize=(20, 15))
for label, color in zip(labels, colors):
mask_label = df_agg_mean[labels_col] == label
dependency_values = df_agg_mean.loc[mask_label, dependency_col]
means4label = df_agg_mean.loc[mask_label, 'score_test'].values.ravel()
stds4label = df_agg_std.loc[mask_label, 'score_test'].values.ravel()
print(means4label)
print(stds4label)
if label2text_dict is not None:
if label in label2text_dict:
label_text = label2text_dict[label]
else:
label_text = label
else:
label_text = label
lower = means4label - stds4label
upper = means4label + stds4label
mask_part_1 = dependency_values < 40
mask_part_2 = dependency_values >= 30
dependency_values_part_1 = dependency_values[mask_part_1]
means4label_part_1 = means4label[mask_part_1]
plt.plot(dependency_values_part_1, means4label_part_1, label=label_text, color=color, lw=5)
dependency_values_part_2 = dependency_values[mask_part_2]
means4label_part_2 = means4label[mask_part_2]
plt.plot(dependency_values_part_2, means4label_part_2, color=color, lw=5, linestyle='--')
plt.scatter(dependency_values, means4label, marker='o', s=140, color=color)
if plot_errors:
plt.fill_between(dependency_values, lower, upper, alpha=0.1, color=color)
ax.yaxis.set_major_locator(MultipleLocator(0.05))
ax.xaxis.set_major_locator(MultipleLocator(10))
plt.tick_params(axis='both', which='major', labelsize=25, size=20)
plt.xticks()
# plt.xlabel('Window Size, s', fontsize=35)
plt.xlabel(xlabel, fontsize=35)
plt.ylabel('ROC AUC', fontsize=35)
plt.legend(fontsize=15) # 16 if not enough space # , loc='lower right')
# plt.xlim(97, 610) # UPDATE IF 60 IS ADDED!!!
plt.tight_layout()
plt.savefig(f'pic/{labels_col}_{dependency_col}_{suffix}.png')
plt.close()
if version == 'time_step':
xlabel = 'Time Step, s'
elif version == 'hidden_size':
xlabel = 'Hidden Size'
elif version == 'window_size':
xlabel = 'Window Size, s'
elif version == 'window_size':
xlabel = 'Window Size, s'
else:
xlabel = version
colors = ['blue', 'orange', 'green', 'red', 'cyan', 'gold', 'black']
labels = [0, 1, 2, 4] + ['Logistic Regression', 'Random Forest', 'SVM']
### hidden_size
plot_dependency(df_results, labels, colors, 'attention', dependency_col='hidden_size',
label2text_dict=attention_names_dict, xlabel='Hidden Size', plot_errors=False)
### window_size
# plot_dependency(df_results, labels, colors, 'attention', dependency_col='window_size',
# label2text_dict=attention_names_dict, xlabel='Window Size, s', plot_errors=False)
### time_step
mask_not_five = [df_results.index[i][0] != 5 for i in range(len(df_results))]
plot_dependency_time_step(df_results.iloc[mask_not_five], labels, colors, 'attention', dependency_col='time_step',
label2text_dict=attention_names_dict, xlabel='Time Step, s', plot_errors=False)
### normalization
plot_dependency(df_results, labels, colors, 'attention', dependency_col='normalization',
label2text_dict=attention_names_dict, xlabel='normalization', plot_errors=False)
# df_att_norm_mean = df_results.groupby(['attention', 'normalization']).mean()
# df_att_norm_std = df_results.groupby(['attention', 'normalization']).std()
# df_att_norm_mean.rename(columns={'score_test': 'score_test_mean'}, inplace=True)
# df_att_norm_mean['score_test_std'] = df_att_norm_std['score_test']
#
# df_att_norm_mean.to_csv('data_best/df_att_norm.csv')
#
for alg_name, color in zip(alg_names_list, colors):
mask_alg = results_no_index['alg_name'] == alg_name
mean4alg = results_mean.iloc[mask_alg.nonzero()].values.ravel()
std4alg = results_std.iloc[mask_alg.nonzero()].values.ravel()
lower = mean4alg - std4alg
upper = mean4alg + std4alg
plt.plot(window_size_list, mean4alg, label=alg_name, linewidth=5, color=color)
plt.scatter(window_size_list, mean4alg, marker='o', s=140, color=color)
plt.fill_between(window_size_list, lower, upper, alpha=0.3, color=color)
plt.tick_params(axis='both', which='major', labelsize=30)
plt.xticks()
# plt.xlabel('Window Size, s', fontsize=35)
plt.xlabel('Window Size, s', fontsize=35)
plt.ylabel('ROC AUC', fontsize=35)
plt.legend(fontsize=32)
# plt.xlim(97, 610) # UPDATE IF 60 IS ADDED!!!
plt.tight_layout()
plt.savefig('pic/classical_ml_window_size_v0.png')
df_results.groupby('time_step').mean()
df_results.groupby('window_size').mean()
df_results.groupby('batch_size').mean()
df_results.groupby('hidden_size').mean()
df_results.groupby('attention').mean() # For RNN 4 is better than 0!!!
df_results.groupby('normalization').mean() # For RNN 4 is better than 0!!!
df_results.groupby(['time_step', 'window_size']).mean()
df_results.groupby(['time_step', 'batch_size']).mean()
df_results.groupby(['time_step', 'hidden_size']).mean()
df_results.groupby(['window_size', 'batch_size']).mean()
df_results.groupby(['window_size', 'hidden_size']).mean()
df_results.groupby(['batch_size', 'hidden_size']).mean()
df_results.groupby(['attention', 'time_step']).mean()
df_results.groupby(['attention', 'window_size']).mean()
df_results.groupby(['attention', 'hidden_size']).mean()
df_results.groupby(['attention', 'hidden_size', 'window_size', 'time_step']).mean()
df_results.groupby(['time_step', 'window_size', 'hidden_size']).mean()
"""
Inference after v2:
1. time_step 20 is probably better
2. window_size 300 should be used. Larger values have better scores, but the system is not so flexible
3. batch_size can be any from 8 to 256. Let's set it to 64
4. hidden size 32 is the best. 64 is slightly worse, 16 is worse
"""
"""
Inference after v1 or v0:
1. Anyway, timestep 30 looks too much. Timestep 5 actually looks fine, but it requires a lot of training.
2. window_size 120 is too short and noisy. 600 has the best score, but the target become too trivial.
3. batch_size 2 isn't a good idea. 16 and 128 aren't very distinguishable, the optimal can be from 8 to inf.
4. The best hidden size is 32, the worst is 2, 8 is ok. More is better till some limit. Maybe to check 16, 64, 128(too much?), ...
5. time_step 10 and window_size 300 is ok.
6. Shorter time_step requires higher batch_size
7. Probably higher time_step require lesser hidden_size
"""
time_step_list = [10, 20] # 10 is already tested
window_size_list = [120, 180, 300, 600]
alg_names_list = ['Logistic Regression', 'Random Forest', 'SVM']
version = 'v0'
df_results = pd.read_csv(f'data/df_results_classic_{version}.csv')
df_results.set_index(['window_size', 'alg_name', 'n_repeat'], inplace=True)
def mean_std(x):
# return pd.Series([x.mean(), x.std()], index=['mean', 'std']).T
return pd.DataFrame({
'mean': x.mean(),
'std': x.std(),
})
# df_results.groupby(['window_size', 'alg_name']).apply(lambda x: mean_std(x))
results_mean = df_results.groupby(['window_size', 'alg_name']).apply(lambda x: x.mean())
results_std = df_results.groupby(['window_size', 'alg_name']).apply(lambda x: x.std())
results_no_index = results_std.reset_index().drop(columns='score_val')
colors = ['blue', 'orange', 'green']
plt.interactive(True)
plt.close()
plt.figure(figsize=(12, 9))
for alg_name, color in zip(alg_names_list, colors):
mask_alg = results_no_index['alg_name'] == alg_name
mean4alg = results_mean.iloc[mask_alg.nonzero()].values.ravel()
std4alg = results_std.iloc[mask_alg.nonzero()].values.ravel()
lower = mean4alg - std4alg
upper = mean4alg + std4alg
plt.plot(window_size_list, mean4alg, label=alg_name, linewidth=5, color=color)
plt.scatter(window_size_list, mean4alg, marker='o', s=140, color=color)
plt.fill_between(window_size_list, lower, upper, alpha=0.3, color=color)
plt.tick_params(axis='both', which='major', labelsize=30)
plt.xticks()
# plt.xlabel('Window Size, s', fontsize=35)
plt.xlabel('Window Size, s', fontsize=35)
plt.ylabel('ROC AUC', fontsize=35)
plt.legend(fontsize=32)
# plt.xlim(97, 610) # UPDATE IF 60 IS ADDED!!!
plt.tight_layout()
plt.savefig('pic/classical_ml_window_size_v0.png')
##### Loading from separate series files
# filenames = os.listdir('data')
# def check_relevance(filename):
# cond_1 = filename[-4:] == '.csv'
# cond_2 = filename[:7] == 'series_'
# cond_3 = filename[-6:-4] == version
# return cond_1 and cond_2 and cond_3
#
# relevant_series = [filename for filename in filenames if check_relevance(filename)]
#
# df_results = pd.DataFrame()
#
# for series_path in relevant_series:
# series2append = pd.read_csv(f'data/{series_path}')
# index_names = ['time_step', 'window_size', 'batch_size', 'hidden_size']
# series2append.columns = index_names + list(series2append.columns[4:])
# series2append.set_index(index_names, inplace=True)
# df_results = df_results.append(series2append)
#
# df_agg_mean = df_results.groupby(['attention', 'hidden_size'])['score_test'].mean().reset_index()
# df_agg_std = df_results.groupby(['attention', 'hidden_size'])['score_test'].std().reset_index()
# attention_list = [0, 1, 2]
# hidden_size_list = [8, 32, 64]
#
# plt.close()
# for attention in attention_list:
# mask = df_agg_mean['attention'] == attention
# hidden_sizes = df_agg_mean.loc[mask, 'hidden_size']
# means = df_agg_mean.loc[mask, 'score_test']
# stds = df_agg_std.loc[mask, 'score_test']
# plt.plot(hidden_sizes, means, label=attention_names_dict[attention])
#
#
# plt.legend()
# plt.tight_layout()