for i in range(start, split_param_size):
masks[split_metrics[i][0]][split_metrics[i][1]] = 1
num += 1
print('NUM: ', num)
else:
threshold = np.mean(metrics_list) # + np.std(metrics_list)
size = 0
for i in range(len(split_metrics)):
if split_metrics[i][2] < threshold:
break
else:
masks[split_metrics[i][0]][split_metrics[i][1]] = 1
size += 1
print('split size: ', size)
network.set_param_values(init_param_value)
split_param_size += 1
if split_param_size != 0:
split_network = MLP_MaskedSplit(
input_shape=(in_dim + len(difficulties) + 1, ),
output_dim=out_dim,
hidden_sizes=hidden_size,
hidden_nonlinearity=NL.tanh,
output_nonlinearity=None,
split_num=len(difficulties) + 1,
split_masks=masks,
init_net=network,
)
else:
split_network = copy.deepcopy(network)
split_network.set_param_values(init_param_value)
init_param_value = np.copy(network.get_param_values())
#Xs, Ys = synthesize_data(dim, 2000, tasks)
task_grads = []
for i in range(len(trainingXs)):
task_grads.append([])
if not load_split_data:
net_weight_values = []
for i in range(epochs):
net_weight_values.append(network.get_param_values())
train(train_fn, np.concatenate(trainingXs), np.concatenate(trainingYs), 1)
joblib.dump(net_weight_values, 'data/trained/gradient_temp/supervised_split_' + append + '/net_weight_values.pkl', compress=True)
else:
net_weight_values = joblib.load('data/trained/gradient_temp/supervised_split_' + append + '/net_weight_values.pkl')
for i in range(epochs):
network.set_param_values(net_weight_values[i])
for j in range(len(trainingXs)):
grad = grad_fn(trainingXs[j], trainingYs[j])
task_grads[j].append(grad)
print('------- collected gradient info -------------')
pred = out(np.concatenate(testingXs))
split_counts = []
for i in range(len(task_grads[0][0])):
split_counts.append(np.zeros(task_grads[0][0][i].shape))
for i in range(len(task_grads[0])):
for k in range(len(task_grads[0][i])):
region_gradients = []
for region in range(len(task_grads)):
