def main():
global best_val_score
gl = get_grid_list()
for g in gl:
best_val_score = 0.0
conf = EfficientNetConfig(depth=g[0], width=g[1], resolution=g[2], num_classes=10)
run_name = conf_to_name(conf)
run = wandb.init(project='EfficientNet_small', reinit=True)
run.name = run_name
run.save()
cur_model = EfficientNet(conf)
cur_model.cuda()
wandb.watch(cur_model)
train(cur_model, run_name)
print(run_name, best_val_score)
run.finish()
scores = model(data)
pred = scores.data.max(1)[1]
test_correct += pred.eq(target.data).cpu().sum()
print("Predicted {} out of {} correctly".format(test_correct,
total_examples))
return 100.0 * test_correct / (float(total_examples))
if __name__ == '__main__':
torch.cuda.device(0)
model = EfficientNet(1.0, 1.0)
config = CONFIG()
model = model.cuda()
avg_loss = list()
best_accuracy = 0.0
optimizer = optim.SGD(model.parameters(),
lr=LEARNING_RATE,
momentum=config.momentum,
weight_decay=config.weight_decay)
train_acc, val_acc = list(), list()
for i in range(1, EPOCHS + 1):
train_acc.append(train(i))
val_acc.append(val())
save_model(model, i)