-
Notifications
You must be signed in to change notification settings - Fork 0
/
trainval_clf.py
178 lines (142 loc) · 5.54 KB
/
trainval_clf.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
import os
import numpy as np
import torch
import time
import sys
from torch.utils.data import Dataset, DataLoader, Subset
from haven import haven_utils as hu
from haven import haven_wizard as hw
from src import datasets
from src.metrics import models_clf
from src.models.utils import validate_and_insert_defaults, Argument
from src.models.haven_utils import save_checkpoint, save_pkl, get_checkpoint
from src import setup_logger
logger = setup_logger.get_logger(__name__)
def get_train_trainval_split(train_dataset, train_percent, split_seed):
# Cut up full training set into its own
# actual train set and valid set.
idcs = np.arange(0, len(train_dataset))
logger.info("split seed: {}".format(split_seed))
rnd_state = np.random.RandomState(split_seed)
rnd_state.shuffle(idcs)
# Set aside 90% for train and 10%
# for valid.
logger.info("Train percentage: {}".format(train_percent))
train_idcs = idcs[0 : int(len(idcs) * train_percent)]
valid_idcs = idcs[int(len(idcs) * train_percent) : :]
logger.info("Len of train set: {}".format(len(train_idcs)))
logger.info("Len of train-valid set: {}".format(len(valid_idcs)))
train_actual = Subset(train_dataset, indices=train_idcs)
valid_actual = Subset(train_dataset, indices=valid_idcs)
return train_actual, valid_actual
NONETYPE = type(None)
DEFAULTS = {
"dataset": {
# actual dataset seed
"name": Argument("name", "emnist_fs", [str]),
"input_size": Argument("input_size", 32, [int]),
"seed": Argument("seed", 42, [int]),
"train_percentage": Argument("train_percentage", 0.9, [float]),
"clf_transform_kwargs": Argument("clf_transform_kwargs", {}, [dict, NONETYPE]),
# not used anymore
"n_channels": Argument("n_channels", 1, [int]),
},
"epochs": Argument("epochs", 200, [int]),
"batch_size": Argument("batch_size", 64, [int]),
"optim": {
"lr": Argument("lr", 2e-4, [float]),
"beta1": Argument("beta1", 0.9, [float]),
"beta2": Argument("beta2", 0.999, [float]),
"weight_decay": Argument("weight_decay", 0.0, [float]),
},
}
def trainval(exp_dict, savedir, args):
validate_and_insert_defaults(exp_dict, DEFAULTS)
dataset_seed = exp_dict["dataset"]["seed"]
train_percent = exp_dict["dataset"]["train_percentage"]
epochs = exp_dict["epochs"]
# mixup_alpha = exp_dict.get("mixup_alpha", None)
# We train the AE on this.
train_dataset = datasets.get_dataset(
class_split="train",
k_shot=None,
datadir=args.datadir,
dataset=exp_dict["dataset"]["name"],
input_size=exp_dict["dataset"]["input_size"],
seed=dataset_seed,
transform_kwargs=exp_dict["dataset"]["clf_transform_kwargs"],
)
logger.info("\n" + str(train_dataset))
train_actual, valid_actual = get_train_trainval_split(
train_dataset, train_percent=train_percent, split_seed=0
)
train_loader = DataLoader(
train_actual,
shuffle=True,
batch_size=exp_dict["batch_size"],
num_workers=args.num_workers,
)
valid_loader = DataLoader(
valid_actual,
shuffle=True,
batch_size=exp_dict["batch_size"],
num_workers=args.num_workers,
)
# Pretrain classifier
# -------------------
optim_args = exp_dict["optim"]
model_clf = models_clf.get_model(
n_classes=train_dataset.n_classes,
lr=optim_args["lr"],
beta1=optim_args["beta1"],
beta2=optim_args["beta2"],
weight_decay=optim_args["weight_decay"],
freeze_all_except=None,
)
chk_dict = get_checkpoint(savedir, return_model_state_dict=True)
if len(chk_dict["model_state_dict"]):
model_clf.set_state_dict(chk_dict["model_state_dict"])
val_acc_best = max(
[s.get("val_score", -np.inf) for s in chk_dict["score_list"]] + [-np.inf]
)
if args.dry_run:
sys.stderr.write("dry run set, terminating...\n")
return
for epoch in range(chk_dict["epoch"], epochs):
t0 = time.time()
score_dict = {}
score_dict["epoch"] = epoch
# Train model.
# Since `savedir` is defined, after every epoch it will invoke
# `vis_on_batch` to generate images.
train_dict = model_clf.train_on_loader(train_loader, savedir=savedir)
score_dict.update({("train_" + k): v for k, v in train_dict.items()})
valid_dict = model_clf.val_on_loader(valid_loader)
score_dict.update({("valid_" + k): v for k, v in valid_dict.items()})
score_dict["time"] = time.time() - t0
score_dict["val_score"] = score_dict["valid_acc"]
chk_dict["score_list"] += [score_dict]
# Save best model if 'val_acc' improves
if score_dict.get("val_score", -np.inf) >= val_acc_best:
logger.info(
"new best validation acc: from {} to {}".format(
val_acc_best, score_dict["val_score"]
)
)
val_acc_best = score_dict["val_score"]
score_dict["best"] = True
save_checkpoint(
savedir,
fname_suffix="_best",
score_list=chk_dict["score_list"],
model_state_dict=model_clf.get_state_dict(),
verbose=True,
)
# Save last chkpt
save_checkpoint(
savedir,
score_list=chk_dict["score_list"],
model_state_dict=model_clf.get_state_dict(),
verbose=False,
)
print("Experiment completed et epoch %d" % epoch)