1
0
mirror of https://github.com/gryf/coach.git synced 2025-12-18 11:40:18 +01:00

Adding checkpointing framework (#74)

* Adding checkpointing framework as well as mxnet checkpointing implementation.

- MXNet checkpoint for each network is saved in a separate file.

* Adding checkpoint restore for mxnet to graph-manager

* Add unit-test for get_checkpoint_state()

* Added match.group() to fix unit-test failing on CI

* Added ONNX export support for MXNet
This commit is contained in:
Sina Afrooze
2018-11-19 09:45:49 -08:00
committed by shadiendrawis
parent 4da56b1ff2
commit 67eb9e4c28
19 changed files with 598 additions and 29 deletions

View File

@@ -23,6 +23,7 @@ import tensorflow as tf
from rl_coach.architectures.architecture import Architecture
from rl_coach.base_parameters import AgentParameters, DistributedTaskParameters
from rl_coach.core_types import GradientClippingMethod
from rl_coach.saver import SaverCollection
from rl_coach.spaces import SpacesDefinition
from rl_coach.utils import force_list, squeeze_list, start_shell_command_and_wait
@@ -637,6 +638,16 @@ class TensorFlowArchitecture(Architecture):
self.curr_rnn_c_in = self.middleware.c_init
self.curr_rnn_h_in = self.middleware.h_init
def collect_savers(self, parent_path_suffix: str) -> SaverCollection:
"""
Collection of all checkpoints for the network (typically only one checkpoint)
:param parent_path_suffix: path suffix of the parent of the network
(e.g. could be name of level manager plus name of agent)
:return: checkpoint collection for the network
"""
# TODO implement returning checkpoints for tensorflow
return SaverCollection()
def save_onnx_graph(input_nodes, output_nodes, checkpoint_save_dir: str) -> None:
"""