-
Notifications
You must be signed in to change notification settings - Fork 1
/
bootstrap_dataset_acc.py
241 lines (217 loc) · 10.4 KB
/
bootstrap_dataset_acc.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
import sys;
import os;
import glob;
import math;
import numpy as np;
import random;
import statistics as stat;
import time;
import torch;
import pickle;
from copy import deepcopy;
import pandas as pd;
sys.path.append(os.getcwd());
import resources.opts as opts;
import resources.models as models;
import resources.calculator as calc;
import resources.utils as U;
import resources.al_utils as alu;
class Tester:
def __init__(self, opt=None):
self.opt = opt;
self.testX = None;
self.testY = None;
self.valX = None;
self.valY = None;
self.opt.device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu");
self.load_val_data();
self.load_test_data();
self.net = None;
self.transformer = None;
self.load_model("","");
def load_model(self, model_type, loop):
if model_type == "":
netSize = 'micro_' if self.opt.netSize == 'micro' else '';
net_path = os.path.join(self.opt.modelPath, '{}{}{}_acdnet_*.pt'.format(netSize, self.opt.dataset, self.opt.datasetSuffix));
# print(net_path);
# exit();
file_paths = glob.glob(net_path);
if len(file_paths)>0 and os.path.isfile(file_paths[0]):
state = torch.load(file_paths[0], map_location=self.opt.device);
self.net = models.GetACDNetModel(self.opt.inputLength, self.opt.nClasses, self.opt.sr, channel_config=state['config']).to(self.opt.device);
self.net.load_state_dict(state['weight']);
print('Base model loaded: {}'.format(file_paths[0]));
self.transformer = alu.get_net(deepcopy(self.net), self.opt.nClasses*5);
else:
print('Model not found');
exit();
else:
net_path = os.path.join(self.opt.modelPath, '{}{}{}_{}{}_acdnet_*.pt'.format(self.opt.netSize, model_type, loop, self.opt.dataset, self.opt.datasetSuffix));
file_paths = glob.glob(net_path);
if len(file_paths)>0 and os.path.isfile(file_paths[0]):
state = torch.load(file_paths[0], map_location=self.opt.device);
self.net = models.GetACDNetModel(self.opt.inputLength, self.opt.nClasses, self.opt.sr, channel_config=state['config']).to(self.opt.device);
self.net.load_state_dict(state['weight']);
print('Learned model loaded: {}'.format(file_paths[0]));
else:
print('Model not found');
exit();
def load_test_data(self):
data = np.load(os.path.join(self.opt.data, self.opt.dataset, 'data/test{}.npz'.format(self.opt.datasetSuffix)), allow_pickle=True);
self.testX = np.array(data['x']);
self.testY = np.array(data['y']);
def load_val_data(self):
data = np.load(os.path.join(self.opt.data, self.opt.dataset, 'data/val{}.npz'.format(self.opt.datasetSuffix)), allow_pickle=True);
self.valX = torch.tensor(data['x']).to(self.opt.device);
self.valY = torch.tensor(data['y']).to(self.opt.device);
def test_val_data(self):
acc = self.__validate(self.valX, self.valY);
print('Validation accuracy: {:.2f}'.format(acc));
def TestBaseModel(self):
start_time = time.time();
# self.load_model("", "");
print('Val Acc {:.2f}'.format(self.__validate(self.valX, self.valY)));
# continue;
acc_log = [];
sample_indices = list(range(0, len(self.testX)));
iter = 0;
while iter < 1000:
iter += 1;
#START: Bootstrap Sampling
bootstrap_indices = np.random.choice(sample_indices, replace=True, size=len(self.testX));
x = torch.tensor(self.testX[bootstrap_indices]).to(self.opt.device);
y= torch.tensor(self.testY[bootstrap_indices]).to(self.opt.device);
acc_log.append(self.__validate(x, y));
stdv = stat.stdev(acc_log);
mean = stat.mean(acc_log);
lBound = mean - 1.96 * stdv;
uBound = mean + 1.96 * stdv;
print('{} - Base Model 95PCI: stdev {:.2f}, mean {:.2f}, lBound {:.2f}, uBound {:.2f}'.format(self.opt.dataset, stdv, mean, lBound, uBound));
print("Elapsed time: {}".format(U.to_hms(time.time()-start_time)));
return acc_log;
def TestLearnedModel(self, type):
start_time = time.time();
self.load_model(type, self.opt.loops);
print('Val Acc {:.2f}'.format(self.__validate(self.valX, self.valY)));
# continue;
acc_log = [];
sample_indices = list(range(0, len(self.testX)));
iter = 0;
while iter < 1000:
iter += 1;
#START: Bootstrap Sampling
bootstrap_indices = np.random.choice(sample_indices, replace=True, size=len(self.testX));
x = torch.tensor(self.testX[bootstrap_indices]).to(self.opt.device);
y= torch.tensor(self.testY[bootstrap_indices]).to(self.opt.device);
acc_log.append(self.__validate(x, y));
stdv = stat.stdev(acc_log);
mean = stat.mean(acc_log);
lBound = mean - 1.96 * stdv;
uBound = mean + 1.96 * stdv;
print('{} - Base Model 95PCI: stdev {:.2f}, mean {:.2f}, lBound {:.2f}, uBound {:.2f}'.format(self.opt.dataset, stdv, mean, lBound, uBound));
print("Elapsed time: {}".format(U.to_hms(time.time()-start_time)));
return acc_log;
def transform_data(self, x):
self.transformer.eval();
features = None;
with torch.no_grad():
features = self.transformer(x).data.cpu().numpy();
return features;
def TestClassifiers(self):
start_time = time.time();
valY = np.argmax(np.array(self.valY.to('cpu')), 1);
testY = np.argmax(self.testY, 1);
# load the model from disk
dir = os.getcwd();
model_path = '{}/trained_models/{}';
file_paths = glob.glob(model_path.format(dir, '{}al{}_{}{}_kncc_*.sav'.format(self.opt.netSize, self.opt.loops, self.opt.dataset, self.opt.datasetSuffix)));
knnc = pickle.load(open(file_paths[0], 'rb'));
file_paths = glob.glob(model_path.format(dir, '{}al{}_{}{}_lgr_*.sav'.format(self.opt.netSize, self.opt.loops, self.opt.dataset, self.opt.datasetSuffix)));
lgr = pickle.load(open(file_paths[0], 'rb'));
file_paths = glob.glob(model_path.format(dir, '{}al{}_{}{}_ridge_*.sav'.format(self.opt.netSize, self.opt.loops, self.opt.dataset, self.opt.datasetSuffix)));
ridge = pickle.load(open(file_paths[0], 'rb'));
result = [];
classifiers = [knnc, lgr, ridge];
for clssifier in classifiers:
#validation accuracy
val_pred = clssifier.predict(self.transform_data(self.valX));
val_acc = (((val_pred==valY)*1).mean()*100).item();
print('Val Acc {:.2f}'.format(val_acc));
acc_log = [];
sample_indices = list(range(0, len(self.testX)));
iter = 0;
while iter < 1000:
iter += 1;
#START: Bootstrap Sampling
bootstrap_indices = np.random.choice(sample_indices, replace=True, size=len(self.testX));
x = self.testX[bootstrap_indices];
y = testY[bootstrap_indices];
trans_x = None;
batch = self.opt.batchSize;
for i in range(math.ceil(len(x)//batch)+1):
tx =self.transform_data(torch.tensor(x[i*batch: (i+1)*batch], dtype=torch.float32).to(self.opt.device))
trans_x = tx if trans_x is None else np.concatenate((trans_x, tx));
pred = clssifier.predict(trans_x);
acc = (((pred==y)*1).mean()*100).item();
acc_log.append(acc);
stdv = stat.stdev(acc_log);
stdErr = stdv/math.sqrt(1000);
mean = stat.mean(acc_log);
lBound = mean - 1.96 * stdErr;
uBound = mean + 1.96 * stdErr;
print('{} - Classifier 95PCI: stdev {:.2f}, mean {:.2f}, lBound {:.2f}, uBound {:.2f}'.format(self.opt.dataset, stdv, mean, lBound, uBound));
result.append(acc_log);
print("Elapsed time: {}".format(U.to_hms(time.time()-start_time)));
return result;
def __validate(self, dataX, dataY):
lossFunc = torch.nn.KLDivLoss(reduction='batchmean');
self.net.eval();
with torch.no_grad():
y_pred = None;
batch_size = self.opt.batchSize;
for idx in range(math.ceil(len(dataX)/batch_size)):
x = dataX[idx*batch_size : (idx+1)*batch_size];
scores = self.net(x);
y_pred = scores.data if y_pred is None else torch.cat((y_pred, scores.data));
acc, loss = self.__compute_accuracy(y_pred, dataY, lossFunc);
return acc;
def __compute_accuracy(self, y_pred, y_target, lossFunc):
with torch.no_grad():
pred = y_pred.argmax(dim=1);
target = y_target.argmax(dim=1);
acc = (((pred==target)*1).float().mean()*100).item();
# valLossFunc = torch.nn.KLDivLoss();
loss = lossFunc(y_pred.log(), y_target).item();
# loss = 0.0;
return acc, loss;
if __name__ == '__main__':
opt = opts.parse();
datasets = ['esc50', 'us8k', 'small', 'iwingbeat'];
al_loops = [7, 15, 15, 20];
classes = [50, 10, 10, 10];
opt.sr = 20000;
opt.netSize = 'micro';
opt.modelPath = 'trained_models';
opt.datasetSuffix = '';
for idx, ds in enumerate(datasets):
if idx in [0, 1, 2]:
print('{}: Skipped'.format(ds));
continue;
if ds=='small':
opt.dataset = 'us8k';
opt.datasetSuffix = '_small';
opt.dataset = ds;
opt.inputLength = 20000 if ds == 'iwingbeat' else 30225;
opt.nClasses = classes[idx];
opt.loops = al_loops[idx];
tester = Tester(opt);
results = [];
results.append(tester.TestBaseModel());
results.append(tester.TestLearnedModel('icl'));
results.append(tester.TestLearnedModel('al'));
print('Classifiers Loading ....')
results.extend(tester.TestClassifiers());
cols = ['base', 'al_icl', 'al_featl', 'al_kncc', 'al_lgr', 'al_ridge'];
df = pd.DataFrame(results, index=cols).T;
df.to_csv('{}al_{}{}_1k_run.csv'.format(opt.netSize, opt.dataset, opt.datasetSuffix), index=False, sep=',');
print('Completed writing into csv....');