/
mlp_ebm.gin
50 lines (45 loc) · 1.69 KB
/
mlp_ebm.gin
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
# coding=utf-8
# Copyright 2021 The Reach ML Authors.
#
# 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.
train_eval.root_dir = '/tmp/ibc_logs/mlp_ebm'
train_eval.loss_type = 'ebm' # specifies we are training ebm.
train_eval.network = 'MLPEBM'
train_eval.batch_size = 512
train_eval.num_iterations = 50000
train_eval.replay_capacity = 10000
train_eval.eval_interval = 5000
train_eval.eval_episodes = 20
train_eval.learning_rate = 1e-3
train_eval.goal_tolerance = 0.02
train_eval.sequence_length = 2
train_eval.dataset_eval_fraction = 0.0
# Config for sampling actions.
ImplicitBCAgent.num_counter_examples = 256 # training.
IbcPolicy.num_action_samples = 16384
train_eval.uniform_boundary_buffer = 0.05
get_normalizers.nested_obs = True # BlockPushing has nested
get_normalizers.num_samples = 5000
compute_dataset_statistics.min_max_actions = True
# Configs for cloning net.
MLPEBM.layers = 'ResNetPreActivation'
MLPEBM.width = 128
MLPEBM.depth = 16
MLPEBM.rate = 0.0
ResNetLayer.normalizer = None
# Need to not use langevin samples in agent or policy
ImplicitBCAgent.add_grad_penalty = False
ImplicitBCAgent.compute_mse = True
ImplicitBCAgent.fraction_langevin_samples = 0.0
IbcPolicy.use_langevin = False
IbcPolicy.use_dfo = True