/
run_learner.py
182 lines (145 loc) · 7.76 KB
/
run_learner.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
import random
import os
import sys
import numpy as np
import torch
import argparse
from img_data_process import read_dataset
from datetime import datetime
from copy import deepcopy
from src.models import NetworkModel
from src.eval_model import bulk_evaluate
from src.train_model import train
from src.learners import Learner
def argument_parser():
# Get an argument parser for a training script.
parser = argparse.ArgumentParser(formatter_class=argparse.ArgumentDefaultsHelpFormatter)
parser.add_argument('--dataset', help='name of dataset', default=None)
parser.add_argument('--algorithm', help='name of algorithm', default=None)
parser.add_argument('--seed', help='random seed', default=0, type=int)
parser.add_argument('--order', help='order for MAML and variants.', default=None, type=int)
parser.add_argument('--classes', help='number of classes per inner task', default=None, type=int)
parser.add_argument('--shots', help='number of examples per class', default=None, type=int)
parser.add_argument('--meta-shots', help='shots for meta update', default=None, type=int)
parser.add_argument('--inner-iters', help='inner iterations', default=None, type=int)
parser.add_argument('--learning-rate',help='inner loop learning rate', default=None, type=float)
parser.add_argument('--meta-step', help='outer loop learning rate', default=None, type=float)
parser.add_argument('--meta-batch', help='meta-training batch size', default=None, type=int)
parser.add_argument('--meta-iters', help='meta-training iterations', default=None, type=int)
parser.add_argument('--eval-iters', help='evaluation inner iterations', default=None, type=int)
parser.add_argument('--eval-samples', help='evaluation samples', default=None, type=int)
parser.add_argument('--eval-interval',help='evaluation interval during training', default=None, type=int)
parser.add_argument('--eval-interval-sample',help='evaluation samples during training', default=None, type=int)
parser.add_argument('--ibp-eps', help='IBP neighborhood size', default=0, type=float)
parser.add_argument('--softmax-temp', help='softmax temperature', default=None, type=float)
parser.add_argument('--only-evaluation', help='for only evaluation', action='store_true', default=False)
parser.add_argument('--checkpoint', help='load saved checkpoint from path', default=None)
parser.add_argument('--test-iters', help='number of evaluations', default=None, type=int)
parser.add_argument('--beta-a', help='beta distrebution parameter a', default=None, type=float)
parser.add_argument('--beta-b', help='beta distribution parameter b', default=None, type=float)
parser.add_argument('--mixup', help='set to use mixup task', action='store_true', default=False)
parser.add_argument('--ibp-layers', help='number layer to perform IBP/IBI', default=None, type=int)
return parser
def model_kwargs(parsed_args):
# Parameters used for initializing the learner.
return {
'update_lr': parsed_args.learning_rate,
'meta_step_size': parsed_args.meta_step,
'beta_a': parsed_args.beta_a,
'beta_b': parsed_args.beta_b,
'softmax_temp': parsed_args.softmax_temp
}
def train_kwargs(parsed_args):
# Parameters used for training.
return {
'order': parsed_args.order,
'num_classes': parsed_args.classes,
'num_shots': parsed_args.shots,
'meta_shots': parsed_args.meta_shots,
'inner_iters': parsed_args.inner_iters,
'meta_batch_size': parsed_args.meta_batch,
'meta_iters': parsed_args.meta_iters,
'eval_inner_iters': parsed_args.eval_iters,
'eval_interval': parsed_args.eval_interval,
'eval_interval_sample': parsed_args.eval_interval_sample,
'ibp_epsilon': parsed_args.ibp_eps,
'mixup': parsed_args.mixup,
'ibp_layers': parsed_args.ibp_layers
}
def evaluate_kwargs(parsed_args):
# Parameters used for evaluation over multiple tasks.
return {
'num_classes': parsed_args.classes,
'num_shots': parsed_args.shots,
'eval_inner_iters': parsed_args.eval_iters,
'num_samples': parsed_args.eval_samples
}
def main():
args = argument_parser().parse_args()
torch.manual_seed(args.seed)
random.seed(args.seed)
np.random.seed(args.seed)
# Edit here according to need.
DATA_DIR = '' + args.dataset
# Create directory for storing results and initiate logging.
if os.path.exists(os.path.join(DATA_DIR, 'val')):
val_presence = True
print("Validation set is present.")
else:
val_presence = False
print("Validation set is not found. Exiting.")
sys.exit()
time_string = datetime.now().strftime("%m%d%Y_%H:%M:%S")
output_folder = args.dataset + '_' + args.algorithm + '_output_folder_' + time_string
output_file = output_folder + '/' + 'log_' + time_string + '.txt'
if not os.path.exists(output_folder):
os.makedirs(output_folder)
with open(output_file, 'a+') as fp:
print('\n'.join(f'{k}={v}' for k, v in vars(args).items()), file=fp)
device = torch.device('cuda')
# Instantiate the dataset.
train_set, val_set, test_set = read_dataset(DATA_DIR, val_presence)
# Instantiate the learner
model=NetworkModel(args.classes)
learner = Learner(model, device, **model_kwargs(args))
# Perform training or evaluation as per need.
if args.only_evaluation is False:
train(learner, train_set, val_set, output_file, output_folder, **train_kwargs(args))
else:
assert args.checkpoint is not None, 'For evaluating without training please provide a checkpoint'
print('Evaluating...')
res_file = output_folder + '/' + 'test_performance_' + time_string + '_.txt'
with open(res_file, 'a+') as fp:
print('Evalulation checkpoint: ' + args.checkpoint, file=fp)
checkpoint_model = torch.load(args.checkpoint, map_location='cuda:0')
learner.net.load_state_dict(checkpoint_model['model_state'])
learner.meta_optim.load_state_dict(checkpoint_model['meta_optim_state'])
train_accuracy, val_accuracy, test_accuracy = [], [], []
train_cnf, val_cnf, test_cnf = [], [], []
for ii in range(args.test_iters):
train_acc, train_div = bulk_evaluate(learner, train_set, **evaluate_kwargs(args))
val_acc, val_div = bulk_evaluate(learner, val_set, **evaluate_kwargs(args))
test_acc, test_div = bulk_evaluate(learner, test_set, **evaluate_kwargs(args))
train_accuracy.append(train_acc)
val_accuracy.append(val_acc)
test_accuracy.append(test_acc)
train_cnf.append(train_div)
val_cnf.append(val_div)
test_cnf.append(test_div)
with open(res_file, 'a+') as fp:
print('Test iteration: ' + str(ii + 1), file=fp)
print('Train accuracy: ' + str(train_accuracy[-1]) + ' +/- ' + str(train_cnf[-1]), file=fp)
print('Validation accuracy: ' + str(val_accuracy[-1]) + ' +/- ' + str(val_cnf[-1]), file=fp)
print('Test accuracy: ' + str(test_accuracy[-1]) + ' +/- ' + str(test_cnf[-1]) + '\n', file=fp)
save_path = output_folder + '/' + 'results' + '.npz'
train_accuracy = np.array(train_accuracy)
val_accuracy = np.array(val_accuracy)
test_accuracy = np.array(test_accuracy)
train_cnf = np.array(train_cnf)
val_cnf = np.array(val_cnf)
test_cnf = np.array(test_cnf)
np.savez(save_path, train_accuracy=train_accuracy, val_accuracy=val_accuracy,
test_accuracy=test_accuracy, train_confidence=train_cnf, val_confidence=val_cnf,
test_confidence=test_cnf)
if __name__ == '__main__':
main()