def test(self, X_test_r, reconstruct = True):
costs = []
num_batch_valid = len(X_test)//BATCH_SIZE
for j in range(num_batch_valid):
cost = valid_fn(
X_test[j*BATCH_SIZE: (j+1)*BATCH_SIZE],
X_test_r[j*BATCH_SIZE: (j+1)*BATCH_SIZE]
)
costs.append(cost)
self.X = self.generate_fn(self.generate_routine, 32, 32, 25)
# model.save_params('{}/epoch_{}_val_error_{}.pkl'.format(OUT_DIR,i, numpy.mean(costs)))
if reconstruct:
reconstruction = generate_routine(X_test[:25])
plot_25_figure(self.X, '{}/epoch_{}_val_error_{}_gen_images.jpg'.format(OUT_DIR, i, numpy.mean(costs)), num_channels = 3)
plot_25_figure(reconstruction, '{}/epoch_{}_reconstructed.jpg'.format(OUT_DIR, i), num_channels = 3)
print "Validation cost after epoch {}: {}".format(i+1, numpy.mean(costs))
self.errors['validation'].append(numpy.mean(costs))
for j in range(num_batch_valid):
cost = valid_fn(
X_test[j*BATCH_SIZE: (j+1)*BATCH_SIZE],
X_test_r[j*BATCH_SIZE: (j+1)*BATCH_SIZE]
)
costs.append(cost)
if (j+1) % PRINT_EVERY == 0:
print ("Validation: epoch {}, iter {}, cost {}".format(i,j+1,numpy.mean(costs)))
model.save_params('{}/epoch_{}_val_error_{}.pkl'.format(OUT_DIR,i, numpy.mean(costs)))
X = generate_fn(generate_routine, 32, 32, 25)
reconstruction = generate_routine(X_test[:25])
plot_25_figure(X, '{}/epoch_{}_val_error_{}_gen_images.jpg'.format(OUT_DIR, i, numpy.mean(costs)), num_channels = 3)
plot_25_figure(reconstruction, '{}/epoch_{}_reconstructed.jpg'.format(OUT_DIR, i), num_channels = 3)
print("Validation cost after epoch {}: {}".format(i+1, numpy.mean(costs)))
errors['validation'].append(numpy.mean(costs))
if i % 2 == 0:
save(errors, '{}/epoch_{}_NLL.pkl'.format(OUT_DIR, i))
if (j + 1) % PRINT_EVERY == 0:
print("Training: epoch {}, iter {}, cost {}".format(
i, j + 1, numpy.mean(costs)))
print("Training cost for epoch {}: {}".format(i + 1, numpy.mean(costs)))
errors['training'].append(numpy.mean(costs))
val_error = validate()
errors['validation'].append(val_error)
model.save_params('{}/epoch_{}_val_error_{}.pkl'.format(
OUT_DIR, i, val_error)) #parameters are saved after every epoch
X = generate_fn(generate_routine, 28, 28,
25) # 25 images are generated after every epoch
reconstruction = generate_routine(X_test[:25])[:, :, :, 0]
plot_25_figure(
X, '{}/epoch_{}_val_error_{}_gen_images.jpg'.format(
OUT_DIR, i, val_error))
plot_25_figure(reconstruction,
'{}/epoch_{}_reconstructed.jpg'.format(OUT_DIR, i))
print("Validation cost after epoch {}: {}".format(i + 1, val_error))
if i % 2 == 0:
save(
errors, '{}/epoch_{}_NLL.pkl'.format(OUT_DIR, i)
) #NLL upto ith epoch stored after every 2 epochs. Too much redundancy here.
num_iters += 1
if (j+1) % PRINT_EVERY == 0:
print ("Training: epoch {}, iter {}, cost {}".format(i,j+1,numpy.mean(costs)))
print("Training cost for epoch {}: {}".format(i+1, numpy.mean(costs)))
errors['training'].append(numpy.mean(costs))
val_error = validate()
errors['validation'].append(val_error)
model.save_params('{}/epoch_{}_val_error_{}.pkl'.format(OUT_DIR,i, val_error)) #parameters are saved after every epoch
X = generate_fn(generate_routine, 28, 28, 25) # 25 images are generated after every epoch
reconstruction = generate_routine(X_test[:25])[:,:,:,0]
plot_25_figure(X, '{}/epoch_{}_val_error_{}_gen_images.jpg'.format(OUT_DIR, i, val_error))
plot_25_figure(reconstruction, '{}/epoch_{}_reconstructed.jpg'.format(OUT_DIR, i))
print("Validation cost after epoch {}: {}".format(i+1, val_error))
if i % 2 == 0:
save(errors, '{}/epoch_{}_NLL.pkl'.format(OUT_DIR, i)) #NLL upto ith epoch stored after every 2 epochs. Too much redundancy here.
for j in range(num_batch_valid):
cost = valid_fn(
X_test[j*BATCH_SIZE: (j+1)*BATCH_SIZE],
X_test_r[j*BATCH_SIZE: (j+1)*BATCH_SIZE]
)
costs.append(cost)
if (j+1) % PRINT_EVERY == 0:
print ("Validation: epoch {}, iter {}, cost {}".format(i,j+1,numpy.mean(costs)))
model.save_params('{}/epoch_{}_val_error_{}.pkl'.format(OUT_DIR,i, numpy.mean(costs)))
X = generate_fn(generate_routine, 32, 32, 25)
reconstruction = generate_routine(X_test[:25])
plot_25_figure(X, '{}/epoch_{}_val_error_{}_gen_images.jpg'.format(OUT_DIR, i, numpy.mean(costs)), num_channels = 3)
plot_25_figure(reconstruction, '{}/epoch_{}_reconstructed.jpg'.format(OUT_DIR, i), num_channels = 3)
print("Validation cost after epoch {}: {}".format(i+1, numpy.mean(costs)))
errors['validation'].append(numpy.mean(costs))
if i % 2 == 0:
save(errors, '{}/epoch_{}_NLL.pkl'.format(OUT_DIR, i))