gryf/coach
1
0
Fork 0
mirror of https://github.com/gryf/coach.git synced 2025-12-17 19:20:19 +01:00
Code Issues Projects Releases Wiki Activity

Create a dataset using an agent (#306)

Browse Source 
Generate a dataset using an agent (allowing to select between this and a random dataset)
This commit is contained in:
Gal Leibovich
2019-05-28 09:34:49 +03:00
committed by GitHub
parent 342b7184bc
commit 9e9c4fd332
26 changed files with 351 additions and 111 deletions
Download Patch File Download Diff File Expand all files Collapse all files

14
rl_coach/agents/categorical_dqn_agent.py
View File

@@ -13,7 +13,6 @@
# See the License for the specific language governing permissions and
# limitations under the License.
#
from typing import Union
import numpy as np
@@ -83,13 +82,22 @@ class CategoricalDQNAgent(ValueOptimizationAgent):
# prediction's format is (batch,actions,atoms)
def get_all_q_values_for_states(self, states: StateType):
q_values = None
if self.exploration_policy.requires_action_values():
q_values = self.get_prediction(states,
outputs=[self.networks['main'].online_network.output_heads[0].q_values])
else:
q_values = None
return q_values
def get_all_q_values_for_states_and_softmax_probabilities(self, states: StateType):
actions_q_values, softmax_probabilities = None, None
if self.exploration_policy.requires_action_values():
outputs = [self.networks['main'].online_network.output_heads[0].q_values,
self.networks['main'].online_network.output_heads[0].softmax]
actions_q_values, softmax_probabilities = self.get_prediction(states, outputs=outputs)
return actions_q_values, softmax_probabilities
def learn_from_batch(self, batch):
network_keys = self.ap.network_wrappers['main'].input_embedders_parameters.keys()