def train(epochs, layer, lr, lambd) :
model = FNN(feature.shape[2], 1, layer, 128)
model.apply(weight_init)
optimizer = optim.Adam(model.parameters(), lr = lr, weight_decay = lambd)
if args.cuda :
model = model.cuda()
print("Training FNN for %d layers, %f learning rate, %f lambda" %(layer, lr, lambd))
for epoch in range(epochs) :
train_epoch(epoch, model, optimizer, lambd)
output = model(feature[:,0,:])
output = output.unsqueeze(1)
for i in range(1, feature.shape[1]) :
output = torch.cat((output, model(feature[:, i, :]).unsqueeze(1)), 1)
# t_weight = torch.stack((weight, weight),2)
t_weight = weight
output = output.squeeze()
output = torch.mul(t_weight, output)
output = torch.sum(output, 1)
loss_train = Loss(output[idx_train], out[idx_train])
loss_val = Loss(output[idx_val], out[idx_val])
print("Result for %d layers, %f learning rate, %f lambda" %(layer, lr, lambd))
print('loss_val: {:.4f}'.format(loss_val.item()))
return model, output, loss_val
def train(epochs, layer, lr, lambd, idx_train, idx_val):
model = FNN(feature.shape[1], out.shape[1], layer, 128)
model.apply(weight_init)
optimizer = optim.Adam(model.parameters(), lr = lr, weight_decay = lambd)
if args.cuda:
model = model.cuda()
print("Training FNN for %d layers, %f learning rate, %f lambda" %(layer, lr, lambd))
for epoch in range(epochs) :
train_epoch(epoch, model, optimizer, lambd, idx_train, idx_val)
output = model(feature)
loss_val = F.mse_loss(output[idx_val], out[idx_val])
print("Result for %d layers, %f learning rate, %f lambda" %(layer, lr, lambd))
print('loss_val: {:.4f}'.format(loss_val.item()))
return output, loss_val