File size: 6,258 Bytes
f7009b3 |
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 |
# set global seed
import time
print("time stamp:", time.time())
import random
import numpy as np
import torch
import re
import sys
if __name__ == "__main__":
def get_permutation_state():
try: # get string
string = sys.argv[1]
except IndexError:
RuntimeError("sys.argv[1] not found")
class_int_string = str(re.search(r'class(\d+)', string).group(1)).zfill(4)
return int(class_int_string)
seed = SEED = get_permutation_state()
else: # when testing
seed = SEED = 0
torch.manual_seed(seed)
torch.cuda.manual_seed(seed)
torch.cuda.manual_seed_all(seed)
torch.backends.cudnn.deterministic = True
torch.backends.cudnn.benchmark = True
np.random.seed(seed)
random.seed(seed)
print("Seed:", SEED)
try: # relative import
from model import Model
except ImportError:
from .model import Model
# import
import torch.nn as nn
from torch import optim
from torch.optim import lr_scheduler
from torch.utils.data import DataLoader
from torchvision.datasets import CIFAR10 as Dataset
from torchvision import transforms
from torch.nn import functional as F
import warnings
warnings.filterwarnings("ignore", category=UserWarning)
# load additional config
import os
import json
config_file = os.path.join(os.path.dirname(os.path.dirname(__file__)), "config.json")
with open(config_file, "r") as f:
additional_config = json.load(f)
# config
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
config = {
"dataset_root": "from_additional_config",
"batch_size": 250 if __name__ == "__main__" else 50,
"num_workers": 16,
"learning_rate": 5e-3,
"epochs": 200,
"weight_decay": 0.1,
"save_learning_rate": 2e-5,
"total_save_number": 5,
"tag": os.path.basename(os.path.dirname(__file__)),
}
config.update(additional_config)
# Data
dataset = Dataset(
root=config["dataset_root"],
train=True,
download=True,
transform=transforms.Compose([
transforms.Resize(224),
transforms.RandomCrop(224, padding=32),
transforms.RandomHorizontalFlip(),
transforms.AutoAugment(policy=transforms.AutoAugmentPolicy("cifar10")),
transforms.ToTensor(),
transforms.Normalize((0.4914, 0.4822, 0.4465), (0.2471, 0.2435, 0.2616)),
])
)
train_loader = DataLoader(
dataset=dataset,
batch_size=config["batch_size"],
num_workers=config["num_workers"],
shuffle=True,
drop_last=True,
pin_memory=True,
persistent_workers=True,
)
test_loader = DataLoader(
dataset=Dataset(
root=config["dataset_root"],
train=False,
download=True,
transform=transforms.Compose([
transforms.Resize(224),
transforms.ToTensor(),
transforms.Normalize((0.4914, 0.4822, 0.4465), (0.2471, 0.2435, 0.2616)),
])),
batch_size=config["batch_size"],
num_workers=config["num_workers"],
shuffle=False,
)
# Model
model, head = Model()
model = model.to(device)
criterion = nn.CrossEntropyLoss()
# Optimizer
optimizer = optim.AdamW(
model.parameters(),
lr=config["learning_rate"],
weight_decay=config["weight_decay"],
)
scheduler = lr_scheduler.CosineAnnealingLR(
optimizer,
T_max=config["epochs"],
eta_min=config["save_learning_rate"],
)
# Training
def train(model=model, optimizer=optimizer, scheduler=scheduler):
model.train()
for batch_idx, (inputs, targets) in enumerate(train_loader):
inputs, targets = inputs.to(device), targets.to(device)
optimizer.zero_grad()
with torch.cuda.amp.autocast(enabled=True, dtype=torch.bfloat16):
outputs = model(inputs)
loss = criterion(outputs, targets)
loss.backward()
optimizer.step()
if scheduler is not None:
scheduler.step()
# test
@torch.no_grad()
def test(model=model):
model.eval()
all_targets = []
all_predicts = []
test_loss = 0
correct = 0
total = 0
for batch_idx, (inputs, targets) in enumerate(test_loader):
inputs, targets = inputs.to(device), targets.to(device)
with torch.cuda.amp.autocast(enabled=True, dtype=torch.bfloat16):
outputs = model(inputs)
loss = criterion(outputs, targets)
# to logging losses
all_targets.extend(targets.flatten().tolist())
test_loss += loss.item()
_, predicts = outputs.max(1)
all_predicts.extend(predicts.flatten().tolist())
total += targets.size(0)
correct += predicts.eq(targets).sum().item()
loss = test_loss / (batch_idx + 1)
acc = correct / total
print(f"Loss: {loss:.4f} | Acc: {acc:.4f}\n")
model.train()
return loss, acc, all_targets, all_predicts
# save train
def save_train(model=model, optimizer=optimizer):
data_loader = DataLoader(
dataset=dataset,
batch_size=min(len(dataset) // config["total_save_number"], config["batch_size"]),
num_workers=config["num_workers"],
shuffle=True,
drop_last=True,
)
model.train()
for batch_idx, (inputs, targets) in enumerate(data_loader):
inputs, targets = inputs.to(device), targets.to(device)
optimizer.zero_grad()
with torch.cuda.amp.autocast(enabled=False, dtype=torch.bfloat16):
outputs = model(inputs)
loss = criterion(outputs, targets)
loss.backward()
optimizer.step()
# Save checkpoint
_, acc, _, _ = test(model=model)
if not os.path.isdir('checkpoint'):
os.mkdir('checkpoint')
save_state = {key: value.cpu().to(torch.float32) for key, value in model.state_dict().items()}
torch.save(save_state, f"checkpoint/{str(batch_idx).zfill(4)}_acc{acc:.4f}_class{SEED:04d}_{config['tag']}.pth")
print("save:", f"checkpoint/{str(batch_idx).zfill(4)}_acc{acc:.4f}_class{SEED:04d}_{config['tag']}.pth")
# exit loop
if batch_idx+1 == config["total_save_number"]:
break
# main
if __name__ == '__main__':
for epoch in range(config["epochs"]):
train(model=model, optimizer=optimizer, scheduler=scheduler)
test(model=model)
save_train(model=model, optimizer=optimizer)
print("time stamp:", time.time()) |