-
Notifications
You must be signed in to change notification settings - Fork 1
/
train_tabular.py
247 lines (230 loc) · 9.29 KB
/
train_tabular.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
"""Conduct additional experiments on toy datasets."""
import os
from collections import namedtuple
from typing import Final, List, Any, Dict
import shutil
import click
import pandas as pd
import torch
import numpy as np
from sklearn.datasets import make_classification
from sklearn.ensemble import RandomForestClassifier
from sklearn.neural_network import MLPClassifier
from sklearn.model_selection import train_test_split
from udecomp.utils import save_as_json, compute_eval_metrics
N_SAMPLES: Final[int] = 7000
TRAIN_SPLIT: Final[float] = 6/7
CLASSES_DEFAULT: Final[int] = 4
FEATURES_DEFAULT: Final[int] = 2
SEP_DEFAULT: Final[float] = 1.
MEMBERS_DEFAULT: Final[int] = 10
MEMBERS_DEFAULT_MLP: Final[int] = 10
DEPTH_DEFAULT: Final[int] = 10
DEPTH_DEFAULT_MLP: Final[int] = 10
"""0 is just random, 5 is full seperation, but with two clusters
per class instead of 1, so that may influence the results (experiment with this)"""
DISTS: Final[List] = [0.05, 0.25, 0.5, 1.0, 2.5, 5.0]
NOISES: Final[List] = [0.01, 0.1, 0.25, 0.5, 0.75]
MEMBERS: Final[List] = [3, 5, 10, 20, 50]
DEPTHS: Final[List] = [1, 3, 5, 7, 10, 20]
EXPERIMENTS: Final[Dict] = {
'data_situation': {'distance': DISTS, 'noise': NOISES},
'learner_situation': {
'members': MEMBERS, 'complexity': DEPTHS, 'bootstrap': [True, False],
}
}
@click.command()
@click.option('-ex', '--experiment', required=True)
@click.option('-pl', '--ensemble-learner', required=True, default='rf')
@click.option('-rs', '--random-seed', required=True)
def main(experiment: str, ensemble_learner: str, random_seed: int):
"""Run experiments."""
ClassificationDataset = namedtuple('ClassificationDataset', ['dsid', 'data'])
EnsembleLearner = namedtuple(
'EnsembleLearner', ['elid', 'type', 'learner']
)
seed: Final[int] = int(random_seed)
dataset_args = {
'n_samples': N_SAMPLES,
'n_features': FEATURES_DEFAULT,
'n_informative': FEATURES_DEFAULT,
'n_redundant': 0,
'n_repeated': 0,
'n_clusters_per_class': 1,
'class_sep': SEP_DEFAULT,
'random_state': seed,
}
if ensemble_learner == 'rf':
learner_args = {'random_state': seed, 'criterion': 'entropy'}
else:
learner_args = {
'solver': 'sgd',
'max_iter': 100,
'learning_rate_init': 0.1,
'early_stopping': True,
}
rs_dir = os.path.join('results', experiment, ensemble_learner, str(random_seed))
if os.path.exists(rs_dir):
shutil.rmtree(rs_dir)
if experiment in EXPERIMENTS['data_situation'].keys():
for case in EXPERIMENTS['data_situation'][experiment]:
case_str = str(case) if isinstance(case, int) else str(int(100 * case))
result_path = os.path.join(
'results',
experiment,
ensemble_learner,
str(random_seed),
case_str
)
os.makedirs(result_path, exist_ok=True)
if experiment == 'distance':
dataset_args['class_sep'] = case
dataset = ClassificationDataset(
experiment,
make_classification(
n_classes=case if experiment == 'classes' else CLASSES_DEFAULT,
flip_y=case if experiment == 'noise' else 0.01,
**dataset_args
)
)
x, y = dataset.data
dt = pd.DataFrame(
{
'x_1': x[:, 0],
'x_2': x[:, 1],
'y': y,
'experiment': experiment,
'case': case,
'rs': seed,
}
)
dt.to_csv(os.path.join(result_path, 'dataset.csv'))
x_train, x_test, y_train, y_test = train_test_split(
x, y, test_size=TRAIN_SPLIT, random_state=seed
)
if ensemble_learner == 'rf':
el = EnsembleLearner(
elid=f'{experiment}_rf',
type='rf',
learner=RandomForestClassifier(
max_depth=DEPTH_DEFAULT, **learner_args
)
)
clf = el.learner
clf.fit(x_train, y_train)
preds = [bl.predict_proba(x_test) for bl in clf.estimators_]
elif ensemble_learner == 'mlp':
np.random.seed = seed
el = EnsembleLearner(
elid=f'{experiment}_mlp',
type='mlp',
learner=[
MLPClassifier(
hidden_layer_sizes=(DEPTH_DEFAULT_MLP,), **learner_args
)
for _ in range(MEMBERS_DEFAULT)
]
)
clf = el.learner
idx_bs = [
np.random.choice(
range(len(x_train)), size=len(x_train), replace=True
)
for _ in range(MEMBERS_DEFAULT)
]
x_train_bs = [x_train[i, ...] for i in idx_bs]
y_train_bs = [y_train[i, ...] for i in idx_bs]
clf_trained: List[Any] = [
bl.fit(x, y) for bl, x, y in zip(clf, x_train_bs, y_train_bs)
]
preds = [bl.predict_proba(x_test) for bl in clf_trained]
else:
raise ValueError('unknown learner')
preds = np.array(preds)
torch.save(y_test, os.path.join(result_path, 'targets.pt'))
for m in range(preds.shape[0]):
pred_bl = preds[m].round(4)
torch.save(pred_bl, os.path.join(result_path, f'scores_{m}.pt'))
metrics = compute_eval_metrics(torch.tensor(preds), torch.tensor(y_test))
save_as_json(metrics, os.path.join(result_path, f'logfile.json'))
elif experiment in EXPERIMENTS['learner_situation'].keys():
for case in EXPERIMENTS['learner_situation'][experiment]:
case_str = str(case)
result_path = os.path.join(
'results',
experiment,
ensemble_learner,
str(random_seed),
case_str
)
os.makedirs(result_path, exist_ok=True)
dataset = ClassificationDataset(
experiment,
make_classification(
n_classes=CLASSES_DEFAULT,
flip_y=0.01,
**dataset_args
)
)
x, y = dataset.data
x_train, x_test, y_train, y_test = train_test_split(
x, y, test_size=TRAIN_SPLIT, random_state=seed
)
if ensemble_learner == 'rf':
el = EnsembleLearner(
elid=f'{experiment}_rf',
type='rf',
learner=RandomForestClassifier(
n_estimators=case
if experiment == 'members' else MEMBERS_DEFAULT,
max_depth=case if experiment == 'complexity' else DEPTH_DEFAULT,
**learner_args
)
)
clf = el.learner
clf.fit(x_train, y_train)
preds = [bl.predict_proba(x_test) for bl in clf.estimators_]
elif ensemble_learner == 'mlp':
n_members = case if experiment == 'members' else MEMBERS_DEFAULT_MLP
np.random.seed(seed)
el = EnsembleLearner(
elid=f'{experiment}_mlp',
type='mlp',
learner=[
MLPClassifier(
hidden_layer_sizes=(
case
if experiment == 'complexity' else DEPTH_DEFAULT_MLP,
),
**learner_args
)
for _ in range(n_members)
]
)
clf = el.learner
np.random.seed(seed) # perform manual bootstrapping
idx_bs = [
np.random.choice(
range(len(x_train)), size=len(x_train), replace=True
)
for _ in range(MEMBERS_DEFAULT)
]
x_train_bs = [x_train[i, ...] for i in idx_bs]
y_train_bs = [y_train[i, ...] for i in idx_bs]
clf_trained: List[Any] = [
bl.fit(x, y) for bl, x, y in zip(clf, x_train_bs, y_train_bs)
]
preds = [bl.predict_proba(x_test) for bl in clf_trained]
else:
raise ValueError('unknown learner')
preds = np.array(preds)
torch.save(y_test, os.path.join(result_path, 'targets.pt'))
for m in range(preds.shape[0]):
pred_bl = preds[m].round(4)
torch.save(pred_bl, os.path.join(result_path, f'scores_{m}.pt'))
metrics = compute_eval_metrics(torch.tensor(preds), torch.tensor(y_test))
save_as_json(metrics, os.path.join(result_path, f'logfile.json'))
else:
raise ValueError('Unknown experiment')
if __name__ == '__main__':
main()