def batch(X, Y, module, loss=ls.MSE(), epsilon=0.005, max_iter=100):
cout = []
for i in range(max_iter):
module.zero_grad()
ypred = module.forward(X)
delta = loss.backward(Y, ypred)
module.backward_update_gradient(X, delta)
module.update_parameters(epsilon=epsilon)
cout.append(loss.forward(Y, ypred))
return cout
def stochastique(X, Y, module, loss=ls.MSE(), epsilon=0.0001, max_iter=100):
cout = []
for i in range(max_iter):
inds = list(range(len(X)))
np.random.shuffle(inds)
for x, y in zip(X[inds], Y[inds]):
x, y = x.reshape(1, -1), y.reshape(1, -1)
module.zero_grad()
ypred = module.forward(x)
delta = loss.backward(y, ypred)
module.backward_update_gradient(x, delta)
module.update_parameters(epsilon=epsilon)
cout.append(loss.forward(y, ypred))
return cout
def mini_batch(X, Y, module, loss=ls.MSE(), n=50, epsilon=0.05, max_iter=100):
cout = []
for i in range(max_iter):
inds = list(range(len(X)))[:n]
np.random.shuffle(inds)
module.zero_grad()
ypred = module.forward(X[inds])
delta = loss.backward(Y[inds], ypred)
module.backward_update_gradient(X[inds], delta)
module.update_parameters(epsilon=epsilon)
cout.append(loss.forward(Y[inds], ypred))
return cout
os.mkdir(args.checkpoint_dir)
except:
pass
optimizer = None
criterion = None
p = Path(args.checkpoint_dir)
model = FFN(args.input_size)
if args.model_name:
model = preprocessors.load_from_state_dict(model, p / args.model_name)
if args.train:
print(f'Running model in train mode for {args.n_epochs} epochs...')
optimizer = torch.optim.Adam(model.parameters(), lr=args.lr)
criterion = losses.MSE()
loaders = preprocessors.make_dataloaders(args.data_dir,
data_ext='.npy',
batch_size=args.batch_size)
if args.evaluate and not args.train:
print('Running model evaluation...')
loaders = {
'test':
preprocessors.make_dataloader(data_dir=args.data_dir,
data_file=args.eval_data,
is_train=False,
batch_size=args.batch_size)
}
cr = CustomRunner(model=model,