-
Notifications
You must be signed in to change notification settings - Fork 1
/
Copy pathclassification.py
190 lines (157 loc) · 8.35 KB
/
classification.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
# -*- coding: utf-8 -*-
#
# Copyright 2020 Pietro Barbiero, Alberto Tonda and Giovanni Squillero
# Licensed under the EUPL
import sys
import scipy
import os
from sklearn.datasets import make_classification
from sklearn.ensemble import RandomForestClassifier, RandomForestRegressor
from sklearn.feature_selection import RFE, SelectKBest
from sklearn.linear_model import RidgeClassifier, Ridge
from sklearn import clone
from lazygrid.datasets import load_openml_dataset
from sklearn.model_selection import cross_validate
from sklearn.pipeline import Pipeline
from sklearn.preprocessing import StandardScaler
from tqdm import tqdm
import pandas as pd
import numpy as np
from dfe.dfe import DFE
from skfwrapper.skfwrapper import SKF_lap, SKF_mcfs, SKF_spec, SKF_ndfs, SKF_udfs
random_state = 42
X1, y1 = make_classification(n_samples=2600, n_features=500,
n_informative=50, n_redundant=50, n_repeated=50,
class_sep=3, random_state=random_state, shuffle=False)
X2, y2 = make_classification(n_samples=2600, n_features=500,
n_informative=100, n_redundant=100, n_repeated=100,
class_sep=0.5, random_state=random_state, shuffle=False)
X3, y3 = make_classification(n_samples=2600, n_features=500,
n_informative=150, n_redundant=150, n_repeated=150,
class_sep=3, random_state=random_state, shuffle=False)
# X4, y4 = make_classification(n_samples=2600, n_features=5000,
# n_informative=300, n_redundant=0, n_repeated=150,
# class_sep=5, random_state=random_state)
datasets = [
# "iris",
# ["madelon-100", X1, y1],
# ["madelon-300", X2, y2],
["madelon-500", X3, y3],
# ["madelon-5000", X1, y4],
# "yeast_ml8", # 2417 samples 116 features 2 classes
# "scene", # 2407 samples 299 features 2 classes
# "isolet", # 7797 samples 617 features 26 classes
# "gina_agnostic", # 3468 samples 970 features 2 classes
# "gas-drift", # 13910 samples 129 features 6 classes
# "letter", # 20000 samples 17 samples 26 classes
# "mozilla4", # 15545 samples 6 features 2 classes
# "Amazon_employee_access", # 32769 samples 10 features 2 classes
# "electricity", # 45312 samples 9 features 2 classes
# "mnist_784", # 70000 samples 785 features 10 classes
# "covertype", # 581012 samples 55 features 7 classes
# "gisette", # 7000 samples 5000 features 2 classes
# "amazon-commerce-reviews", # 1500 samples 10000 features 50 classes
# "OVA_Colon", # 1545 samples 10936 features 2 classes
# "GCM", # 190 samples 16063 features 14 classes
# "Dexter", # 600 samples 20000 features 2 classes
# "variousCancers_final", # 383 samples 54676 features 9 classes
# "anthracyclineTaxaneChemotherapy", # 159 samples 61360 features 2 classes
# "Dorothea", # 1150 samples 100000 features 2 classes
]
def main():
# Cross-validation params
cv = 10
n_jobs = -1
seed = 42
results_dir = "./results/classification2"
if not os.path.isdir(results_dir):
os.makedirs(results_dir)
overall_scores = pd.DataFrame()
bar_position = 0
progress_bar = tqdm(datasets, position=bar_position)
for dataset in progress_bar:
dataset, X, y = dataset
progress_bar.set_description("Analysis of dataset: %s" % dataset)
verbose_scores = pd.DataFrame()
summary_scores = pd.DataFrame()
if not os.path.isdir(results_dir):
os.makedirs(results_dir)
# X, y, n_classes = load_openml_dataset(dataset_name=dataset)
n_classes = len(set(y))
# clf = RidgeClassifier(random_state=seed)
regr = Ridge(random_state=seed)
est = RidgeClassifier(random_state=seed)
fs_lap_score = SelectKBest(SKF_lap)
fs_SPEC = SelectKBest(SKF_spec)
fs_MCFS = SelectKBest(SKF_mcfs)
fs_NDFS = SelectKBest(SKF_ndfs)
fs_UDFS = SelectKBest(SKF_udfs)
sc = StandardScaler()
pipelines = {
"DFE": Pipeline([("sc", sc), ("fs", DFE(clone(regr), base_score=0.9)), ("est", clone(est))]),
"RFE": Pipeline([("sc", sc), ("fs", RFE(clone(regr), verbose=1)), ("est", clone(est))]),
"lap_score": Pipeline([("sc", sc), ("fs", fs_lap_score), ("est", clone(est))]),
"SPEC": Pipeline([("sc", sc), ("fs", fs_SPEC), ("est", clone(est))]),
"NDFS": Pipeline([("sc", sc), ("fs", fs_NDFS), ("est", clone(est))]),
"UDFS": Pipeline([("sc", sc), ("fs", fs_UDFS), ("est", clone(est))]),
"MCFS": Pipeline([("sc", sc), ("fs", fs_MCFS), ("est", clone(est))]),
"NO-FS": Pipeline([("sc", sc), ("est", clone(est))]),
}
k_best_list = ["lap_score",
"SPEC",
"NDFS",
"UDFS",
"MCFS", ]
for method, pipeline in pipelines.items():
# for feature selection methods that need to specify the number
# of features to select a priori, just pick the number of features
# chosen by DFE
if method in k_best_list:
setattr(pipeline.steps[1][1], "k", int(round(k_select)))
if method == 'RFE':
setattr(pipeline.steps[1][1], "n_features_to_select", int(round(k_select)))
# cross validation
scores = cross_validate(pipeline, X, y, n_jobs=n_jobs, cv=cv,
return_train_score=True, return_estimator=True)
# save results
try:
n_features = [estimator.steps[-1][1].coef_.shape[1] for estimator in scores["estimator"]]
except:
n_features = [estimator.steps[-1][1].feature_importances_.shape[0] for estimator in scores["estimator"]]
if method != 'NO-FS':
try:
features = [estimator.steps[1][1].scores_ for estimator in scores["estimator"]]
except:
features = [estimator.steps[1][1].ranking_ for estimator in scores["estimator"]]
else:
features = [np.ones(n_features[0]).astype(int) for _ in range(len(n_features))]
scores["n_features_selected"] = n_features
scores["features_selected"] = features
scores["estimator"] = method
scores = pd.DataFrame.from_records(scores)
verbose_scores = pd.concat([verbose_scores, scores], ignore_index=True)
verbose_scores.to_csv(os.path.join(results_dir, dataset + "_verbose.csv"))
summary = {
"dataset": dataset,
"samples": X.shape[0],
"features": X.shape[1],
"classes": n_classes,
"method": method,
"avg fit time": np.average(scores["fit_time"]),
"sem fit time": scipy.stats.sem(scores["fit_time"]),
"avg train score": np.average(scores["train_score"]),
"sem train score": scipy.stats.sem(scores["train_score"]),
"avg test score": np.average(scores["test_score"]),
"sem test score": scipy.stats.sem(scores["test_score"]),
"avg n features": np.average(n_features),
"sem n features": scipy.stats.sem(n_features)
}
summary = pd.Series(summary)
summary_scores = pd.concat([summary_scores, summary], axis=1, ignore_index=True)
summary_scores.to_csv(os.path.join(results_dir, dataset + "_summary.csv"))
overall_scores = pd.concat([overall_scores, summary], axis=1, ignore_index=True)
overall_scores.T.to_csv(os.path.join(results_dir, "overall_results.csv"))
if method == "DFE":
k_select = np.average(n_features)
if __name__ == "__main__":
sys.exit(main())