mirror of
https://github.com/gryf/coach.git
synced 2025-12-17 19:20:19 +01:00
1b095aeeca
Till now, most of the modules were importing all of the module objects (variables, classes, functions, other imports) into module namespace, which potentially could (and was) cause of unintentional use of class or methods, which was indirect imported. With this patch, all the star imports were substituted with top-level module, which provides desired class or function. Besides, all imports where sorted (where possible) in a way pep8[1] suggests - first are imports from standard library, than goes third party imports (like numpy, tensorflow etc) and finally coach modules. All of those sections are separated by one empty line. [1] https://www.python.org/dev/peps/pep-0008/#imports
77 lines
3.4 KiB
Python
77 lines
3.4 KiB
Python
#
|
|
# Copyright (c) 2017 Intel Corporation
|
|
#
|
|
# Licensed under the Apache License, Version 2.0 (the "License");
|
|
# you may not use this file except in compliance with the License.
|
|
# You may obtain a copy of the License at
|
|
#
|
|
# http://www.apache.org/licenses/LICENSE-2.0
|
|
#
|
|
# Unless required by applicable law or agreed to in writing, software
|
|
# distributed under the License is distributed on an "AS IS" BASIS,
|
|
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
|
# See the License for the specific language governing permissions and
|
|
# limitations under the License.
|
|
#
|
|
import numpy as np
|
|
|
|
from agents import agent
|
|
from architectures import network_wrapper as nw
|
|
import utils
|
|
|
|
|
|
class ValueOptimizationAgent(agent.Agent):
|
|
def __init__(self, env, tuning_parameters, replicated_device=None, thread_id=0, create_target_network=True):
|
|
agent.Agent.__init__(self, env, tuning_parameters, replicated_device, thread_id)
|
|
self.main_network = nw.NetworkWrapper(tuning_parameters, create_target_network, self.has_global, 'main',
|
|
self.replicated_device, self.worker_device)
|
|
self.networks.append(self.main_network)
|
|
self.q_values = utils.Signal("Q")
|
|
self.signals.append(self.q_values)
|
|
|
|
self.reset_game(do_not_reset_env=True)
|
|
|
|
# Algorithms for which q_values are calculated from predictions will override this function
|
|
def get_q_values(self, prediction):
|
|
return prediction
|
|
|
|
def get_prediction(self, curr_state):
|
|
return self.main_network.online_network.predict(self.tf_input_state(curr_state))
|
|
|
|
def _validate_action(self, policy, action):
|
|
if np.array(action).shape != ():
|
|
raise ValueError((
|
|
'The exploration_policy {} returned a vector of actions '
|
|
'instead of a single action. ValueOptimizationAgents '
|
|
'require exploration policies which return a single action.'
|
|
).format(policy.__class__.__name__))
|
|
|
|
def choose_action(self, curr_state, phase=utils.RunPhase.TRAIN):
|
|
prediction = self.get_prediction(curr_state)
|
|
actions_q_values = self.get_q_values(prediction)
|
|
|
|
# choose action according to the exploration policy and the current phase (evaluating or training the agent)
|
|
if phase == utils.RunPhase.TRAIN:
|
|
exploration_policy = self.exploration_policy
|
|
else:
|
|
exploration_policy = self.evaluation_exploration_policy
|
|
|
|
action = exploration_policy.get_action(actions_q_values)
|
|
self._validate_action(exploration_policy, action)
|
|
|
|
# this is for bootstrapped dqn
|
|
if type(actions_q_values) == list and len(actions_q_values) > 0:
|
|
actions_q_values = actions_q_values[self.exploration_policy.selected_head]
|
|
actions_q_values = actions_q_values.squeeze()
|
|
|
|
# store the q values statistics for logging
|
|
self.q_values.add_sample(actions_q_values)
|
|
|
|
# store information for plotting interactively (actual plotting is done in agent)
|
|
if self.tp.visualization.plot_action_values_online:
|
|
for idx, action_name in enumerate(self.env.actions_description):
|
|
self.episode_running_info[action_name].append(actions_q_values[idx])
|
|
|
|
action_value = {"action_value": actions_q_values[action], "max_action_value": np.max(actions_q_values)}
|
|
return action, action_value
|