def main(argv=None):
define()
if (FLAGS.summaries_dir and not os.path.exists(FLAGS.summaries_dir)):
os.makedirs(FLAGS.summaries_dir)
setup_logger('log',
stream=True,
log_file=FLAGS.summaries_dir + "mnist_gpdnn_results.txt")
global logger
logger = logging.getLogger('log')
logger.info(FLAGS)
mnist_tutorial(nb_epochs=FLAGS.nb_epochs,
batch_size=FLAGS.batch_size,
learning_rate=FLAGS.learning_rate,
clean_train=FLAGS.clean_train,
backprop_through_attack=FLAGS.backprop_through_attack)
def main(argv=None):
define()
if (FLAGS.summaries_dir and not os.path.exists(FLAGS.summaries_dir)):
os.makedirs(FLAGS.summaries_dir)
setup_logger('log',
stream=True,
log_file=FLAGS.summaries_dir + "results.txt")
global logger
logger = logging.getLogger('log')
logger.info(FLAGS)
mnist_blackbox(nb_classes=FLAGS.nb_classes,
batch_size=FLAGS.batch_size,
learning_rate=FLAGS.learning_rate,
nb_epochs=FLAGS.nb_epochs,
holdout=FLAGS.holdout,
data_aug=FLAGS.data_aug,
nb_epochs_s=FLAGS.nb_epochs_s,
lmbda=FLAGS.lmbda)
indx_array = np.mod(np.arange(data_indx, data_indx + minibatch_size), x_train.shape[0])
data_indx += minibatch_size
fd = gp_model.feeds or {}
fd.update({
phs.ximage_flat: x_train[indx_array],
phs.label: y_train[indx_array]
})
_, loss_evd = tf_session.run([minimise, -gp_model.objective], feed_dict=fd)
# logger.info progress every 50 steps.
if i % 50 == 0:
fd = gp_model.feeds or {}
fd.update({phs.ximage_flat: x_test,
phs.label: y_test})
accuracy_evald, log_like_evald = tf_session.run([avg_acc, avg_ll], feed_dict=fd)
logger.info("Iteration {}: Loss is {}. \nTest set LL {}, Acc {}".format(i, loss_evd, log_like_evald,
accuracy_evald))
results.append(dict(step_no=i, loss=loss_evd, test_acc=accuracy_evald, test_ll=log_like_evald))
logger.info("Done!")
# ## We save the results for looking at them later.
df = pd.DataFrame(results)
df.to_pickle("gpdnn_mnist_train.pdpick.pdpick")
if __name__ == '__main__':
setup_logger('log', stream=True)
global logger
logger = logging.getLogger('log')
main()