-
Notifications
You must be signed in to change notification settings - Fork 1
/
launch_utils.py
158 lines (131 loc) · 8.04 KB
/
launch_utils.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
import argparse
import os
import pickle
import shutil
import random
from typing import Tuple, Dict
from avalanche.benchmarks import GenericCLScenario
from avalanche.benchmarks.classic import SplitCIFAR10, SplitCIFAR100, SplitFMNIST, SplitMNIST, SplitTinyImageNet
#from avalanche.benchmarks.generators import benchmark_with_validation_stream, nc_benchmark
from avalanche.benchmarks import nc_benchmark, benchmark_with_validation_stream#now its 0.5.0 ver for ava...
from avalanche.benchmarks.datasets import EMNIST
import numpy as np
import torch
from torch.backends import cudnn
from torchvision import transforms
DATASET_PATH = os.path.join(os.path.abspath('..'), 'datasets_new')
LOG_PATH = os.path.join(os.path.abspath('.'), 'Logs')
def set_seeds(seed: int) -> None:
torch.manual_seed(seed)
random.seed(seed)
np.random.seed(seed)
torch.use_deterministic_algorithms(True)
cudnn.benchmark = False
cudnn.deterministic = True
def get_argument_parser() -> argparse.Namespace:
parser = argparse.ArgumentParser(description='Experiment')
# Logging params
parser.add_argument('--experiment_name', type=str, default='MNIST')
# Context Detector Params
parser.add_argument('--memo_per_class_context', type=int, default=50)
parser.add_argument('--context_layers', type=int, nargs='+', default=[0, 1, 2, 3, 4]) # -1 = Penultimate, 0 = Input
parser.add_argument('--context_learner', type=str, default="LogisticRegression(random_state=0, max_iter=50, C=0.4)")
# Dataset params
parser.add_argument('--dataset', type=str, default='MNIST')
parser.add_argument('--number_of_tasks', type=int, default=5)
# Architectural params
parser.add_argument('--model', type=str, default='CNN_MNIST')
parser.add_argument('--resnet_multiplier', type=float, default=1.0)
# Learning params
parser.add_argument('--seed', type=int, default=0)
parser.add_argument('--optimizer', type=str, default='SGD')
parser.add_argument('--sgd_momentum', type=float, default=0.90)
parser.add_argument('--learning_rate', type=float, default=0.01)
parser.add_argument('--batch_size', type=int, default=256)
parser.add_argument('--weight_decay', type=float, default=0.0)
# Algortihm params
parser.add_argument('--phase_epochs', type=int, default=5)
parser.add_argument("--activation_perc", type=float, default=95.0)
parser.add_argument("--max_phases", type=int, default=5)
return parser.parse_args()
def create_log_dirs(args: argparse.Namespace) -> str:
dirpath = os.path.join(LOG_PATH, args.experiment_name)
# Remove existing files/dirs
if os.path.exists(dirpath) and os.path.isdir(dirpath):
shutil.rmtree(dirpath)
# Create log dirs and save experiment args
os.makedirs(dirpath)
with open(os.path.join(dirpath, 'args.pkl'), 'wb') as file:
pickle.dump(args, file)
for task_id in range(1, args.number_of_tasks + 1):
os.makedirs(os.path.join(dirpath, "Episode_{}".format(task_id)))
return dirpath
def get_experience_streams(args: argparse.Namespace) -> Tuple[GenericCLScenario, int, int, Dict]:
if args.dataset == "MNIST":
stream = SplitMNIST(n_experiences=args.number_of_tasks,
seed=args.seed, dataset_root=DATASET_PATH, fixed_class_order=list(range(10)))
stream_with_val = benchmark_with_validation_stream(stream, validation_size=0.01, output_stream="val", shuffle=True)
task2classes = dict((index, stream.classes_in_this_experience)
for index, stream in enumerate(stream_with_val.train_stream, 1))
return (stream_with_val, 784, 10, task2classes)
if args.dataset == "FMNIST":
stream = SplitFMNIST(n_experiences=args.number_of_tasks,
seed=args.seed, dataset_root=DATASET_PATH, fixed_class_order=list(range(10)))
stream_with_val = benchmark_with_validation_stream(stream, validation_size=0.01, output_stream="val", shuffle=True)
task2classes = dict((index, stream.classes_in_this_experience)
for index, stream in enumerate(stream_with_val.train_stream, 1))
return (stream_with_val, 784, 10, task2classes)
if args.dataset == "EMNIST":
emnist_train = EMNIST(root=DATASET_PATH, train=True, split='letters', download=True)
emnist_train.targets = emnist_train.targets - 1
emnist_test = EMNIST(root=DATASET_PATH, train=False, split='letters', download=True)
emnist_test.targets = emnist_test.targets - 1
stream = nc_benchmark(train_dataset=emnist_train, test_dataset=emnist_test, # type: ignore
n_experiences=args.number_of_tasks, task_labels=False, shuffle=False,
seed=args.seed, fixed_class_order=list(
map(lambda x: int(x), emnist_train.targets.unique())),
train_transform=transforms.ToTensor(),
eval_transform=transforms.ToTensor())
stream_with_val = benchmark_with_validation_stream(stream, validation_size=0.01, output_stream="val", shuffle=True)
task2classes = dict((index, stream.classes_in_this_experience)
for index, stream in enumerate(stream_with_val.train_stream, 1))
return (stream_with_val, 784, 26, task2classes)
if args.dataset == "CIFAR10":
transform = transforms.Compose([transforms.ToTensor(),
transforms.Normalize((0.4914, 0.4822, 0.4465), (0.247, 0.243, 0.261))])
stream = SplitCIFAR10(n_experiences=args.number_of_tasks,
seed=args.seed, dataset_root=DATASET_PATH, fixed_class_order=list(range(10)),
train_transform=transform, eval_transform=transform)
stream_with_val = benchmark_with_validation_stream(stream, validation_size=0.01, output_stream="val", shuffle=True)
task2classes = dict((index, stream.classes_in_this_experience)
for index, stream in enumerate(stream_with_val.train_stream, 1))
return (stream_with_val, 3, 10, task2classes)
if args.dataset == "CIFAR100":
transform = transforms.Compose([transforms.ToTensor(),
transforms.Normalize((0.5071, 0.4867, 0.4408), (0.2675, 0.2565, 0.2761))])
stream = SplitCIFAR100(n_experiences=args.number_of_tasks,
seed=args.seed, dataset_root=DATASET_PATH, fixed_class_order=list(range(100)),
train_transform=transform, eval_transform=transform)
stream_with_val = benchmark_with_validation_stream(stream, validation_size=0.001, output_stream="val", shuffle=True)
task2classes = dict((index, stream.classes_in_this_experience)
for index, stream in enumerate(stream_with_val.train_stream, 1))
return (stream_with_val, 3, 100, task2classes)
if args.dataset == "TinyImagenet":
train_transform = transforms.Compose([
transforms.ToTensor(),
transforms.Normalize(mean=[0.4802, 0.4480, 0.3975],
std=[0.2770, 0.2691, 0.2821])
])
test_transform = transforms.Compose([
transforms.ToTensor(),
transforms.Normalize(mean=[0.4802, 0.4480, 0.3975],
std=[0.2770, 0.2691, 0.2821])
])
stream = SplitTinyImageNet(n_experiences=args.number_of_tasks,
seed=args.seed, dataset_root=DATASET_PATH, fixed_class_order=list(range(200)),
train_transform=train_transform, eval_transform=test_transform)
stream_with_aug = benchmark_with_validation_stream(stream, validation_size=0.01, output_stream="val", shuffle=True)
task2classes = dict((index, stream.classes_in_this_experience)
for index, stream in enumerate(stream_with_aug.train_stream, 1))
return (stream_with_aug, 3, 200, task2classes) # type: ignore
raise Exception("Dataset {} is not defined!".format(args.dataset))