print " batch SGD" losses = [] for b in xrange(num_batches_chunk): # if b % 1000 == 0: # print " batch %d/%d" % (b + 1, num_batches_chunk) loss = train(b) losses.append(loss) # print " loss: %.6f" % loss mean_train_loss = np.sqrt(np.mean(losses)) print " mean training loss (RMSE):\t\t%.6f" % mean_train_loss losses_train.append(mean_train_loss) # store param stds during training param_stds.append([p.std() for p in layers.get_param_values(l6)]) if ((e + 1) % VALIDATE_EVERY) == 0: print print "VALIDATING" print " load validation data onto GPU" for x_shared, x_valid in zip(xs_shared, xs_valid): x_shared.set_value(x_valid) y_shared.set_value(y_valid) print " compute losses" losses = [] for b in xrange(num_batches_valid): # if b % 1000 == 0: # print " batch %d/%d" % (b + 1, num_batches_valid) loss = compute_loss(b)
print " batch SGD" losses = [] for b in xrange(num_batches_chunk): # if b % 1000 == 0: # print " batch %d/%d" % (b + 1, num_batches_chunk) loss = train(b) losses.append(loss) # print " loss: %.6f" % loss mean_train_loss = np.sqrt(np.mean(losses)) print " mean training loss (RMSE):\t\t%.6f" % mean_train_loss losses_train.append(mean_train_loss) # store param stds during training param_stds.append([p.std() for p in layers.get_param_values(l6)]) if ((e + 1) % VALIDATE_EVERY) == 0: print print "VALIDATING" print " load validation data onto GPU" for x_shared, x_valid in zip(xs_shared, xs_valid): x_shared.set_value(x_valid) y_shared.set_value(y_valid) print " compute losses" losses = [] for b in xrange(num_batches_valid): # if b % 1000 == 0: # print " batch %d/%d" % (b + 1, num_batches_valid) loss = compute_loss(b)