runs = [run_1, run_2, run_3, run_4, run_5, run_6, run_7]
num_seeds = 3
target = True
naive = True
if __name__ == "__main__":
# Generate num_seeds number of weight initialisation runs.
for j in range(len(runs)):
if target:
run = runs[j]
elif naive:
run = Run.restore('exp_13', 1, num_seeds, naive=False, save_opp=True)
else:
run = Run.restore('exp_13', 1, num_seeds, save_opp=True)
gen_graphs(run, save_opp=True)
if target or naive:
for i in range(num_seeds):
print("Current seed = " + str(i))
# Calls source network function
if target:
transfer_networks.network(i, run, run.hyperparameters)
run.save(naive=False)
# Calls naive network function
if naive:
# This model must already use linear activation for the final layer
loss = model.layers[layer_idx].output[..., class_idx]
grad_tensor = K.gradients(loss, layer_input)[0]
# create function that evaluate the gradient for a given input
# This function accept numpy array
derivative_fn = K.function([layer_input], [grad_tensor])
# evaluate the derivative_fn
grad_eval_by_hand = derivative_fn([img[np.newaxis, ...]])[0]
print(grad_eval_by_hand.shape)
grad_eval_by_hand = np.abs(grad_eval_by_hand).max(axis=(0, 3))
# normalize to range between 0 and 1
arr_min, arr_max = np.min(grad_eval_by_hand), np.max(grad_eval_by_hand)
grad_eval_by_hand = (grad_eval_by_hand - arr_min) / (arr_max - arr_min + K.epsilon())
plt.imsave(create_path(run.path, model_type, "images", "sal_image_{}_alt.png".format(seed)), grad_eval_by_hand)
if __name__ == "__main__":
r = Run.restore(exp_name, run_num, 3, save_opp=True)
for network in ['target', 'naive', 'source']:
for seed in range(0, 3):
saliency_map(r, network, seed, dataset_type, attempt, category=category, positive=True)
print("{} network done for seed {}".format(network, seed))
# for i in range(0, 3):
# saliency_map(r, 'target', i, "train", 0, category="dog", positive=True)
