-
Notifications
You must be signed in to change notification settings - Fork 0
Expand file tree
/
Copy pathsvm_unigram.py
More file actions
184 lines (141 loc) · 8.33 KB
/
svm_unigram.py
File metadata and controls
184 lines (141 loc) · 8.33 KB
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
import glob
import numpy as np
import os
import pickle
#from multivariate_normalisation import multivariate_noise_norm
from pathlib import Path
from sklearn.linear_model import SGDClassifier
from sklearn.preprocessing import StandardScaler
classes = ['NOUN', 'ADP', 'VERB', 'PRON', 'AUX', 'ADV']
BATCH_SIZE = 512
N_ELECTRODES = 64
model_type = 'SVM'
filename_stem = "unigram_6class"
misc = "small_"
def load_data(N_AVG, filename_stem):
train_data_ = np.load(f'../data/ica_mid/unigram_6class/restricted/{filename_stem}_avg{N_AVG}_train_{misc}data.npy')
train_labels = np.load(f'../data/ica_mid/unigram_6class/restricted/{filename_stem}_avg{N_AVG}_train_{misc}labels.npy')
dev_data_ = np.load(f'../data/ica_mid/unigram_6class/restricted/{filename_stem}_avg{N_AVG}_dev_small_data.npy')
dev_labels = np.load(f'../data/ica_mid/unigram_6class/restricted/{filename_stem}_avg{N_AVG}_dev_small_labels.npy')
test_data_ = np.load(f'../data/ica_mid/unigram_6class/restricted/{filename_stem}_avg{N_AVG}_test_small_data.npy')
test_labels = np.load(f'../data/ica_mid/unigram_6class/restricted/{filename_stem}_avg{N_AVG}_test_small_labels.npy')
if N_AVG > 1:
test_wr_data = np.load(f'../data/ica_mid/unigram_6class/restricted/{filename_stem}_avg{N_AVG}_test_wr_small_data.npy')
test_wr_labels = np.load(f'../data/ica_mid/unigram_6class/restricted/{filename_stem}_avg{N_AVG}_test_wr_small_labels.npy')
else:
test_wr_data = None
test_wr_labels = None
print(f'train_data shape: {train_data_.shape}, len(train_labels) == {len(train_labels)}')
print(f'dev_data shape: {dev_data_.shape}, len(dev_labels) == {len(dev_labels)}')
print(f'test_data shape: {test_data_.shape}, len(test_labels) == {len(test_labels)}')
print(f'test_wr_data shape: {test_wr_data.shape if N_AVG > 1 else 0}')
return (train_data_, train_labels, dev_data_, dev_labels,
test_data_, test_labels, test_wr_data, test_wr_labels)
def preprocess_data(N_AVG, train_data, train_labels, dev_data, dev_labels, test_data,
test_labels, start, end, test_wr_data=None, test_wr_labels=None):
# Collapse along electrode & temporal dims for (n_samples, n_features)
n_samples, n_electrodes, n_timepoints = train_data.shape
train_data = train_data[:,:,start:end]
n_samples, n_electrodes, n_timepoints = dev_data.shape
dev_data = dev_data[:,:,start:end]
n_samples, n_electrodes, n_timepoints = test_data.shape
test_data = test_data[:,:,start:end]
if N_AVG > 1:
n_samples, n_electrodes, n_timepoints = test_wr_data.shape
test_wr_data = test_wr_data[:,:,start:end]
train_data_flat = train_data.reshape((len(train_data), -1))
dev_data_flat = dev_data.reshape((len(dev_data), -1))
test_data_flat = test_data.reshape((len(test_data), -1))
test_wr_data_flat = test_wr_data.reshape((len(test_wr_data), -1))
# Print shapes
print(f'train_data shape: {train_data_flat.shape}')
print(f'dev_data shape: {dev_data_flat.shape}')
print(f'test_data shape: {test_data_flat.shape}')
print(f'test_wr_data shape: {test_wr_data_flat.shape if N_AVG > 1 else 0}')
# Return
return (train_data_flat, train_labels, dev_data_flat, dev_labels,
test_data_flat, test_labels, test_wr_data_flat, test_wr_labels)
def run_model(n_iter, n_epoch, N_AVG, train_data, train_labels, dev_data, dev_labels,
test_data, test_labels, test_wr_data=None, test_wr_labels=None,
filename_stem=None, n_classes=None, start=None, end=None):
classes = [i for i in range(n_classes)]
category = filename_stem
best_global_run = -1
best_global_dev = -1
best_test = -1
#print('train_data.shape = ', train_data.shape, ' train_labels.shape = ', train_labels.shape)
for run_num in range(n_iter):
SGD_clf = SGDClassifier(loss='hinge', tol=1e-3, alpha=0.75)
score = 0
best_score = 0
dev_scores = []
for i in range(n_epoch):
# Shuffle
idx = np.arange(len(train_labels))
np.random.shuffle(idx)
train_data = train_data[idx]
train_labels = train_labels[idx]
n_steps = len(train_data) // BATCH_SIZE
for i in range(0, len(train_data)+1, BATCH_SIZE):
train_data_ = train_data[i:i+BATCH_SIZE]
train_labels_ = train_labels[i:i+BATCH_SIZE]
#print('train_data_.shape = ', train_data_.shape, ' train_labels_.shape = ',
# train_labels_.shape)
SGD_clf.partial_fit(train_data_, train_labels_, classes=classes)
y_pred = SGD_clf.predict(dev_data)
dev_score = sum(y_pred == dev_labels) / len(dev_data)
print(f'{model_type} : epoch ({i}), score = {dev_score*100}%')
dev_scores.append(dev_score)
if dev_score > best_score:
best_score = dev_score
#print(f'saving new best_model (run {run_num})')
pickle.dump(SGD_clf, open(os.path.join("saved_models",
"{}_{}_avg{}_start{}_end{}_{}run{}.pkl".format(model_type, category, N_AVG,
start, end, misc, run_num)), "wb"))
if dev_score > best_global_dev:
best_global_run = run_num
best_global_dev = dev_score
best_test = SGD_clf.score(test_data, test_labels)
model = pickle.load(open(os.path.join("saved_models",
"{}_{}_avg{}_start{}_end{}_{}run{}.pkl".format(model_type, category, N_AVG,
start, end, misc, run_num)), "rb"))
test_score = model.score(test_data, test_labels)
print('......')
print('{}: Test set score (run {}): {}'.format(model_type, run_num, test_score))
predictions = model.predict(test_data)
binary_scores = predictions == test_labels
to_write = np.array([binary_scores, predictions, test_labels]).T
np.savetxt(os.path.join("saved_predictions",
f"{model_type}_{category}_avg{N_AVG}_start{start}_end{end}_{misc}run{run_num}_bin_vec.txt"), to_write)
if N_AVG > 1:
test_score = model.score(test_wr_data, test_wr_labels)
print('{}: Test WR set score (run {}): {}'.format(model_type, run_num, test_score))
print('......')
predictions = model.predict(test_wr_data)
binary_scores = predictions == test_wr_labels
to_write = np.array([binary_scores, predictions, test_wr_labels]).T
np.savetxt(os.path.join("saved_predictions",
f"{model_type}_{category}_avg{N_AVG}_start{start}_end{end}_{misc}run{run_num}_wr_bin_vec.txt"), to_write)
return best_global_run, best_global_dev, best_test
def run_experiment(N_AVG, start, end, filename_stem, n_classes):
(train_data, train_labels,
dev_data, dev_labels,
test_data, test_labels,
test_wr_data, test_wr_labels) = load_data(N_AVG, filename_stem)
(train_data_scaled, train_labels,
dev_data_scaled, dev_labels,
test_data_scaled, test_labels,
test_wr_data_scaled, test_wr_labels) = preprocess_data(N_AVG,
train_data, train_labels,
dev_data, dev_labels,
test_data, test_labels,
start, end, test_wr_data,
test_wr_labels)
print('train_data.shape = ', train_data_scaled.shape, ' train_labels.shape = ', train_labels.shape)
r, d, t = run_model(20, 1, N_AVG, train_data_scaled, train_labels,
dev_data_scaled, dev_labels,
test_data_scaled, test_labels,
test_wr_data_scaled, test_wr_labels,
filename_stem, n_classes, start=start, end=end)
print(f'Best run: {r}, best dev score: {d}, corresponding test: {t}')
run_experiment(N_AVG=10, start=0, end=176, filename_stem="unigram_6class", n_classes=6)