# ======================================================================
# STEP 5: After verifying that your implementation of
# utils.autoencoder_cost_and_grad is correct, You can start training your
# autoencoder, using scipy.optimize.minimize L-BFGS-B.
if RUN_STEP_5_TRAIN_AUTOENCODER:
# Initialize the parameters
theta = utils.initialize(hidden_size, visible_size)
print(
"\nRunning scipy.optimize.minimize on {0} parameters, over {1} training patches_train"
.format(theta.shape[0], patches_train.shape[1]))
start_time = datetime.datetime.now()
print(" START TIME {0}".format(
utils.get_pretty_time_string(start_time)))
# define the objective function that returns cost and grad, used by scipy.optimizze.minimize
import cProfile
cProfile.run(
"utils.autoencoder_cost_and_grad(theta, visible_size, hidden_size, lambda_, patches_train)"
)
# J = lambda x: utils.autoencoder_cost_and_grad(x, visible_size, hidden_size, lambda_, patches_train)
for beta_ in [0.01, 0.1, 0.2]:
for rho_ in [0.05, 0.01, 0.005]:
J = lambda x: utils.autoencoder_cost_and_grad_sparse(
x, visible_size, hidden_size, lambda_, beta_, rho_,
patches_train)
options_ = {'maxiter': 4000, 'disp': False}
result = scipy.optimize.minimize(J,
theta,
# ======================================================================
# STEP 5: After verifying that your implementation of
# utils.autoencoder_cost_and_grad is correct, You can start training your
# autoencoder, using scipy.optimize.minimize L-BFGS-B.
if RUN_STEP_5_TRAIN_AUTOENCODER:
# Initialize the parameters
theta = utils.initialize(hidden_size, visible_size)
print(
"\nRunning scipy.optimize.minimize on {0} parameters, over {1} training patches_train"
.format(theta.shape[0], patches_train.shape[1]))
start_time = datetime.datetime.now()
print(" START TIME {0}".format(
utils.get_pretty_time_string(start_time)))
# define the objective function that returns cost and grad, used by scipy.optimizze.minimize
# J = lambda x: utils.autoencoder_cost_and_grad(x, visible_size, hidden_size, lambda_, patches_train)
options_ = {'maxiter': 4000, 'disp': False}
result = scipy.optimize.minimize(J,
theta,
method='L-BFGS-B',
jac=True,
options=options_)
opt_theta = result.x # theta found after optimization
end_time = datetime.datetime.now()
print(" END TIME {0}".format(utils.get_pretty_time_string(end_time)))
total_time = end_time - start_time
time_elapsed_string = utils.get_pretty_time_string(total_time, delta=True)
print(" Total run time: {0}".format(time_elapsed_string))