def train(hidden_dim_sweep=(5, 10, 25),
n_epochs=20,
out_dir='out',
data_dir='data',
device=util.device(),
Optimizer=optim.Adam,
seed=42):
out_dir, data_dir = map(Path, (out_dir, data_dir))
np.random.seed(seed)
torch.manual_seed(seed)
torch.cuda.manual_seed(seed)
tracess = []
best_trainer = None
best_loss = util.INF
vocab = util.Vocab.load(data_dir / 'vocab.txt')
for hidden_dim in hidden_dim_sweep:
model = Model(hidden_dim=hidden_dim, vocab=vocab, out_dim=2)
loss_fn = nn.CrossEntropyLoss()
optimizer = Optimizer(model.parameters(), lr=1e-4)
trainer = Trainer(model, loss_fn, vocab, device)
traces, loss_cur = trainer.train_loop(data_dir=data_dir,
n_epochs=n_epochs,
optimizer=optimizer,
scheduler=None)
if loss_cur < best_loss:
best_trainer = trainer
best_loss = loss_cur
tracess.append((hidden_dim, traces))
out_dir.mkdir(exist_ok=True)
for h, traces in tracess:
plotting.plot_traces(traces,
out=out_dir / f'traces_{h}.png',
title=f'Loss,hidden_dim={h}')
util.jsondump(traces, out_dir / f'traces.dim_{h}.seed_{seed}.json')
L.info('Best model loss: %s', best_loss)
model_file = out_dir / 'model.pt'
L.info('Saving best model to %s', model_file)
torch.save(best_trainer.model.state_dict(), model_file)
print('Length of training and validation set are {} {}'.format(
len(t_dataset), len(v_dataset)))
trainloader=DataLoader(t_dataset, batch_size=32, shuffle=True, num_workers=8)
validloader=DataLoader(v_dataset, batch_size=32, shuffle=False, num_workers=8)
""" Training """
# model = ResNetModel()
# model = EfficientModelwithoutMeta()
model = Model(arch='efficientnet-b2')
# model = EfficientModel(n_meta_features=len(meta_features))
model.to(device)
# model = nn.DataParallel(model)
criterion=nn.BCEWithLogitsLoss()
optimizer=torch.optim.AdamW(model.parameters(), lr=3e-4, betas=(0.9, 0.999))
scheduler=torch.optim.lr_scheduler.OneCycleLR(
optimizer, max_lr=3e-4, div_factor=10, pct_start=1 / epochs, steps_per_epoch=len(trainloader), epochs=epochs)
writer = SummaryWriter(f'../checkpoint/fold_{fold}/efficient_{resolution}')
print(f'Training Started Fold_{fold}')
training_loss = []
validation_loss = []
c_acc = 0.0
for epoch in range(epochs):
start_time = time.time()
train_prob = []
valid_prob = []