-
Notifications
You must be signed in to change notification settings - Fork 0
/
evaluate_kernel_features.py
320 lines (210 loc) · 11 KB
/
evaluate_kernel_features.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
from functools import partial
from sklearn.model_selection import train_test_split, GridSearchCV, ShuffleSplit
from sklearn.neural_network import MLPRegressor, MLPClassifier
from sklearn.svm import SVR, SVC
from skorch import NeuralNetRegressor, NeuralNetClassifier
import torch
import torch.nn as nn
import numpy as np
import json
# set seeds
torch.manual_seed(100)
np.random.seed(100)
def evaluate(experiment_name, device='cuda:2'):
with open('evaluators.json') as f:
all_evaluators = json.load(f)
experiments = {}
my_evaluators = {
'_evaluate_tool_svm':_evaluate_tool_svm,
'_evaluate_tool_mlp':_evaluate_tool_mlp,
'_evaluate_tool_rnn':_evaluate_tool_rnn,
'_evaluate_tool_neusingle_svm': _evaluate_tool_neusingle_svm,
'_evaluate_classifier_svm':_evaluate_classifier_svm,
'_evaluate_classifier_mlp':_evaluate_classifier_mlp,
'_evaluate_classifier_rnn':_evaluate_classifier_rnn,
}
for s in all_evaluators:
experiments[s['function_name']] = partial(my_evaluators[s['evaluator']], **s['args'])
if experiment_name not in experiments: raise Exception('Experiment not found')
test_loss_mean, test_loss_std = experiments[experiment_name]()
print('Result for {:s}: {:0.4f} ± {:0.4f}'.format(experiment_name, test_loss_mean, test_loss_std))
class RNNModule(nn.Module):
def __init__(self, input_dim, output_dim):
super(RNNModule, self).__init__()
self.rnn = nn.GRU(input_dim, 16, batch_first=True)
self.linear1 = nn.Linear(16, 8)
self.linear2 = nn.Linear(8, output_dim)
def forward(self, X):
X, _ = self.rnn(X)
X = torch.squeeze(X[:, -1, :])
X = torch.relu(self.linear1(X))
X = self.linear2(X)
return X
# ______ _ _
# | ____| | | | |
# | |____ ____ _| |_ _ __ _| |_ ___ _ __ ___
# | __\ \ / / _` | | | | |/ _` | __/ _ \| '__/ __|
# | |___\ V / (_| | | |_| | (_| | || (_) | | \__ \
# |______\_/ \__,_|_|\__,_|\__,_|\__\___/|_| |___/
#
def _create_evaluator(estimator, param_grid, scoring, cv=4, N=5, callback=None):
gs_estimator = GridSearchCV(estimator=estimator, param_grid=param_grid, scoring=scoring, cv=cv, n_jobs=3, refit=True)
def evaluate(X, y, verbose=True):
test_losses = np.zeros(N)
for n in range(N):
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.1, random_state=n)
gs_estimator.fit(X_train, y_train)
test_loss = -gs_estimator.score(X_test, y_test)
test_losses[n] = test_loss
if callback is not None: callback(gs_estimator, X_test, y_test)
if verbose: print('Iteration {:d} | Test Loss = {:0.4f}'.format(n, test_loss))
return np.mean(test_losses), np.std(test_losses)
return evaluate
################################# REGRESSOR FUNCTIONS ########################################
def _load_data(task, tool_type, frequency, transformation, signal_type):
data_dir = f'../robot_sensory_extension/data/kernel_features/kernel_{task}_{tool_type}_{frequency}.npz'
npzfile = np.load(data_dir)
if signal_type == 'neuhalf':
# left sensor
X = np.concatenate( [npzfile['signals'][:, : , 0:40], npzfile['signals'][:, : , 80:120]], 2)
y = npzfile['labels'] * 100
elif signal_type == 'all':
X = npzfile['signals']
y = npzfile['labels'] * 100
if transformation == 'default':
X = np.reshape(X, (X.shape[0], -1))
y = y.ravel()
if transformation == 'tensor':
X = torch.Tensor( X )
y = torch.Tensor( np.reshape(y, (-1, 1)) )
if transformation == 'single':
X = npzfile['signals']
y = npzfile['labels'] * 100
X = np.reshape(X, (X.shape[0], X.shape[1], -1, 80 ))
X = np.swapaxes(X, 1, 3)
X = np.reshape(X, (X.shape[0], 80, -1))
y = y.ravel()
return X, y
def _evaluate_tool_svm(task, tool_type, frequency, signal_type, kernel):
X, y = _load_data(task, tool_type, frequency=frequency, transformation='default', signal_type=signal_type)
param_grid = { 'C': [1, 3, 10, 30, 100] }
estimator = SVR(kernel=kernel, max_iter=5000)
evaluate = _create_evaluator(estimator, param_grid, 'neg_mean_absolute_error')
return evaluate(X, y)
def _evaluate_tool_mlp(task, tool_type, frequency, signal_type):
X, y = _load_data(task, tool_type, frequency=frequency, transformation='default', signal_type=signal_type)
param_grid = {
'learning_rate_init': [0.01, 0.03, 0.1, 0.3],
'alpha': [0.0001, 0.001]
}
estimator = MLPRegressor(hidden_layer_sizes=(16, 8), max_iter=2000, random_state=100)
evaluate = _create_evaluator(estimator, param_grid, 'neg_mean_absolute_error')
return evaluate(X, y)
def _evaluate_tool_rnn(task, tool_type, frequency, signal_type, device):
X, y = _load_data(task, tool_type, frequency=frequency, transformation='tensor', signal_type=signal_type)
#X = X.to(device)
#y = y.to(device)
param_grid = { 'lr': [0.001, 0.003, 0.01] }
estimator = NeuralNetRegressor(RNNModule,
module__input_dim=X.shape[2],
module__output_dim=1,
iterator_train__shuffle=True,
max_epochs=1000,
train_split=False,
device=device,
verbose=0)
evaluate = _create_evaluator(estimator,
param_grid,
'neg_mean_absolute_error',
ShuffleSplit(n_splits=1, test_size=.25))
return evaluate(X, y)
def _evaluate_tool_neusingle_svm(task, tool_type, frequency, signal_type, kernel):
X, y = _load_data(task, tool_type, frequency=frequency, transformation='single', signal_type=signal_type)
best_taxel = 0
best_test_loss_mean = float('inf')
best_test_loss_std = float('inf')
with open(f'results/neutouch_sing_{task}_{tool_type}_{frequency}_{kernel}.csv', 'w') as file:
for taxel in range(1, 81):
param_grid = { 'C': [1, 3, 10, 30, 100] }
estimator = SVR(kernel=kernel, max_iter=5000)
evaluate = _create_evaluator(estimator, param_grid, 'neg_mean_absolute_error')
test_loss_mean, test_loss_std = evaluate(X[:, taxel-1, :], y)
file.write(f'{taxel},{test_loss_mean},{test_loss_std}\n')
if test_loss_mean < best_test_loss_mean:
best_taxel = taxel
best_test_loss_mean = test_loss_mean
best_test_loss_std = test_loss_std
print('Result for taxel {:02d}: {:0.4f} ± {:0.4f}'.format(taxel, test_loss_mean, test_loss_std), flush=True)
print(f'Best performing taxel is {best_taxel}')
return best_test_loss_mean, best_test_loss_std
############################################## CLASSIFIERS FUNCTIONS ##############################################
def _load_classifier_data(task, tool_type, frequency, transformation, signal_type):
data_dir = f'../robot_sensory_extension/data/kernel_features/kernel_{task}_{tool_type}_{frequency}.npz'
npzfile = np.load(data_dir)
if signal_type == 'neuhalf':
# left sensor
X = np.concatenate( [npzfile['signals'][:, : , 0:40], npzfile['signals'][:, : , 80:120]], 2)
y = npzfile['labels']
elif signal_type == 'all':
X = npzfile['signals']
y = npzfile['labels']
if transformation == 'default':
X = np.reshape(X, (X.shape[0], -1))
y = y.ravel()
if transformation == 'tensor':
X = torch.Tensor( X )
y = torch.Tensor( np.reshape(y, (-1)) )
y = y.type(torch.LongTensor)
if transformation == 'single':
X = npzfile['signals']
y = npzfile['labels']
X = np.reshape(X, (X.shape[0], X.shape[1], -1, 80 ))
X = np.swapaxes(X, 1, 3)
X = np.reshape(X, (X.shape[0], 80, -1))
y = y.ravel()
return X, y
def _evaluate_classifier_svm(task, tool_type, frequency, signal_type, kernel):
X, y = _load_classifier_data(task, tool_type, frequency=frequency, transformation='default', signal_type=signal_type)
param_grid = {
'C': [1, 3, 10, 30, 100]
}
estimator = SVC(kernel=kernel, max_iter=5000)
evaluate = _create_evaluator(estimator, param_grid, 'accuracy', N=20)
return evaluate(X, y)
def _evaluate_classifier_mlp(task, tool_type, frequency, signal_type):
X, y = _load_classifier_data(task, tool_type, frequency=frequency, transformation='default', signal_type=signal_type)
param_grid = {
'learning_rate_init': [0.01, 0.03, 0.1, 0.3],
'alpha': [0.0001, 0.001]
}
estimator = MLPClassifier(hidden_layer_sizes=(16, 8), max_iter=2000, random_state=100)
evaluate = _create_evaluator(estimator, param_grid, 'accuracy', N=20)
return evaluate(X, y)
def _evaluate_classifier_rnn(task, tool_type, frequency, signal_type, device):
X, y = _load_classifier_data(task, tool_type, frequency=frequency, transformation='tensor', signal_type=signal_type)
param_grid = { 'lr': [0.001, 0.003, 0.01] }
estimator = NeuralNetClassifier(RNNModule,
module__input_dim=X.shape[2],
module__output_dim=len(torch.unique(y)),
criterion = nn.CrossEntropyLoss,
iterator_train__shuffle=True,
max_epochs=1000,
train_split=False,
device=device,
verbose=0)
evaluate = _create_evaluator(estimator,
param_grid,
'accuracy',
ShuffleSplit(n_splits=1, test_size=.25))
return evaluate(X, y)
# _____ _ _____ _____ _ __
# / ____| | |_ _| |_ _| | | / _|
# | | | | | | | | _ __ | |_ ___ _ __| |_ __ _ ___ ___
# | | | | | | | | | '_ \| __/ _ \ '__| _/ _` |/ __/ _ \
# | |____| |____ _| |_ _| |_| | | | || __/ | | || (_| | (_| __/
# \_____|______|_____| |_____|_| |_|\__\___|_| |_| \__,_|\___\___|
#
if __name__ == '__main__':
import sys
if len(sys.argv) == 2: evaluate(sys.argv[1])
if len(sys.argv) == 3: evaluate(sys.argv[1], sys.argv[2])