act_index = 0
act_fills = []
for num_step in range(num_processing_steps):
act_fill = []
for batch_seq in range(batch_size_tr):
act_fill.append(action_storage[acts[batch_seq][num_step]])
act_fills.append(act_fill)
act_fills = np.array(act_fills)
state_losses_supervise = getFinalState(state)
graph_dicts = []
for j in range(batch_size_tr):
input_g = base_graph(state[j][0], grounding_size, first_action, dimension,proposition_nodes)
graph_dicts.append(input_g)
feed_dicts = utils_tf.get_feed_dict(
graphs_tuple_ph, utils_np.data_dicts_to_graphs_tuple(graph_dicts))
feed_dicts[realstate] = state_losses_supervise
feed_dicts[action_filled] = act_fills
train_values = sess.run({
"step": step_op,
"loss": loss_op_tr,
"outputs": output_ops_tr,
"output_graphs":sta_vecs,
"aaa":graphs_tuple_ph,
"test":action_filled,
}, feed_dict=feed_dicts)
# update action storage
act_index = 0
act_fills = np.array(act_fills)
act_fills_index = np.array(act_fills_index)
state_losses_supervise = getFinalState(state)
graph_dicts = []
goal_graph_dicts = []
for j in range(batch_size_tr):
input_g = base_graph(state[j][0], grounding_size, first_action,
dimension, proposition_nodes)
input_goal_g = base_graph(state[j][num_processing_steps],
grounding_size, first_action, dimension,
proposition_nodes)
graph_dicts.append(input_g)
goal_graph_dicts.append(input_goal_g)
feed_dicts = utils_tf.get_feed_dict(
graphs_tuple_ph, utils_np.data_dicts_to_graphs_tuple(graph_dicts))
goal_feed_dicts = utils_tf.get_feed_dict(
graphs_tuple_goal,
utils_np.data_dicts_to_graphs_tuple(goal_graph_dicts))
feed_dicts.update(goal_feed_dicts)
feed_dicts[realstate] = state_losses_supervise
feed_dicts[action_filled] = act_fills
feed_dicts[action_filled_index] = act_fills_index
train_values = sess.run(
{
"update_para": update_para,
"loss": loss_heu,
"outputs": output_ops_tr,
"heuristic": heuristic_output,
"training_output": training_output,
},
