-
Notifications
You must be signed in to change notification settings - Fork 61
Commit
This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.
Update model instantiators test in jax
- Loading branch information
Showing
1 changed file
with
119 additions
and
88 deletions.
There are no files selected for viewing
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -1,104 +1,135 @@ | ||
| import hypothesis | ||
| import hypothesis.strategies as st | ||
| import pytest | ||
|
|
||
| import gymnasium as gym | ||
| import yaml | ||
| from gymnasium import spaces | ||
|
|
||
| import jax | ||
| import jax.numpy as jnp | ||
| import numpy as np | ||
| from skrl.utils.model_instantiators.jax import categorical_model, deterministic_model, gaussian_model | ||
| from skrl.utils.spaces.jax import flatten_tensorized_space, sample_space | ||
|
|
||
| from skrl.utils.model_instantiators.jax import Shape, categorical_model, deterministic_model, gaussian_model | ||
|
|
||
| NETWORK_SPEC_OBSERVATION = { | ||
| spaces.Box: ( | ||
| r""" | ||
| network: | ||
| - name: net | ||
| input: STATES | ||
| layers: [32, 32, 32] | ||
| activations: elu | ||
| """, | ||
| spaces.Box(low=-1, high=1, shape=(2,)), | ||
| ), | ||
| spaces.Discrete: r""" | ||
| network: | ||
| - name: net | ||
| input: STATES | ||
| layers: [32, 32, 32] | ||
| activations: elu | ||
| """, | ||
| spaces.MultiDiscrete: r""" | ||
| network: | ||
| - name: net | ||
| input: STATES | ||
| layers: [32, 32, 32] | ||
| activations: elu | ||
| """, | ||
| spaces.Tuple: ( | ||
| r""" | ||
| network: | ||
| - name: net_0 | ||
| input: STATES[0] | ||
| layers: [32, 32, 32] | ||
| activations: elu | ||
| - name: net_1 | ||
| input: STATES[1] | ||
| layers: [32, 32, 32] | ||
| activations: elu | ||
| - name: net | ||
| input: net_0 + net_1 | ||
| layers: [32, 32, 32] | ||
| activations: elu | ||
| """, | ||
| spaces.Tuple((spaces.Box(low=-1, high=1, shape=(2,)), spaces.Box(low=-1, high=1, shape=(3,)))), | ||
| ), | ||
| spaces.Dict: ( | ||
| r""" | ||
| network: | ||
| - name: net_0 | ||
| input: STATES["0"] | ||
| layers: [32, 32, 32] | ||
| activations: elu | ||
| - name: net_1 | ||
| input: STATES["1"] | ||
| layers: [32, 32, 32] | ||
| activations: elu | ||
| - name: net | ||
| input: net_0 + net_1 | ||
| layers: [32, 32, 32] | ||
| activations: elu | ||
| """, | ||
| spaces.Dict({"0": spaces.Box(low=-1, high=1, shape=(2,)), "1": spaces.Box(low=-1, high=1, shape=(3,))}), | ||
| ), | ||
| } | ||
|
|
||
|
|
||
| @hypothesis.given( | ||
| observation_space_size=st.integers(min_value=1, max_value=10), | ||
| action_space_size=st.integers(min_value=1, max_value=10), | ||
| ) | ||
| @hypothesis.settings(suppress_health_check=[hypothesis.HealthCheck.function_scoped_fixture], deadline=None) | ||
| @pytest.mark.parametrize("device", [None, "cpu", "cuda:0"]) | ||
| def test_categorical_model(capsys, observation_space_size, action_space_size, device): | ||
| observation_space = gym.spaces.Box(np.array([-1] * observation_space_size), np.array([1] * observation_space_size)) | ||
| action_space = gym.spaces.Discrete(action_space_size) | ||
| # TODO: randomize all parameters | ||
| model = categorical_model( | ||
| observation_space=observation_space, | ||
| action_space=action_space, | ||
| device=device, | ||
| unnormalized_log_prob=True, | ||
| input_shape=Shape.STATES, | ||
| hiddens=[256, 256], | ||
| hidden_activation=["relu", "relu"], | ||
| output_shape=Shape.ACTIONS, | ||
| output_activation=None, | ||
| ) | ||
| model.init_state_dict("model") | ||
| def test_categorical_model(capsys, device): | ||
| # observation | ||
| action_space = spaces.Discrete(2) | ||
| for observation_space_type in [spaces.Box, spaces.Tuple, spaces.Dict]: | ||
| observation_space = NETWORK_SPEC_OBSERVATION[observation_space_type][1] | ||
| model = categorical_model( | ||
| observation_space=observation_space, | ||
| action_space=action_space, | ||
| device=device, | ||
| unnormalized_log_prob=True, | ||
| network=yaml.safe_load(NETWORK_SPEC_OBSERVATION[observation_space_type][0])["network"], | ||
| output="ACTIONS", | ||
| ) | ||
| model.init_state_dict("model") | ||
|
|
||
| with jax.default_device(model.device): | ||
| observations = jnp.ones((10, model.num_observations)) | ||
| output = model.act({"states": observations}) | ||
| assert output[0].shape == (10, 1) | ||
| output = model.act({"states": flatten_tensorized_space(sample_space(observation_space, 10, "jax", device))}) | ||
| assert output[0].shape == (10, 1) | ||
|
|
||
|
|
||
| @hypothesis.given( | ||
| observation_space_size=st.integers(min_value=1, max_value=10), | ||
| action_space_size=st.integers(min_value=1, max_value=10), | ||
| ) | ||
| @hypothesis.settings(suppress_health_check=[hypothesis.HealthCheck.function_scoped_fixture], deadline=None) | ||
| @pytest.mark.parametrize("device", [None, "cpu", "cuda:0"]) | ||
| def test_deterministic_model(capsys, observation_space_size, action_space_size, device): | ||
| observation_space = gym.spaces.Box(np.array([-1] * observation_space_size), np.array([1] * observation_space_size)) | ||
| action_space = gym.spaces.Box(np.array([-1] * action_space_size), np.array([1] * action_space_size)) | ||
| # TODO: randomize all parameters | ||
| model = deterministic_model( | ||
| observation_space=observation_space, | ||
| action_space=action_space, | ||
| device=device, | ||
| clip_actions=False, | ||
| input_shape=Shape.STATES, | ||
| hiddens=[256, 256], | ||
| hidden_activation=["relu", "relu"], | ||
| output_shape=Shape.ACTIONS, | ||
| output_activation=None, | ||
| output_scale=1, | ||
| ) | ||
| model.init_state_dict("model") | ||
| def test_deterministic_model(capsys, device): | ||
| # observation | ||
| action_space = spaces.Box(low=-1, high=1, shape=(2,)) | ||
| for observation_space_type in [spaces.Box, spaces.Tuple, spaces.Dict]: | ||
| observation_space = NETWORK_SPEC_OBSERVATION[observation_space_type][1] | ||
| model = deterministic_model( | ||
| observation_space=observation_space, | ||
| action_space=action_space, | ||
| device=device, | ||
| clip_actions=False, | ||
| network=yaml.safe_load(NETWORK_SPEC_OBSERVATION[observation_space_type][0])["network"], | ||
| output="ACTIONS", | ||
| ) | ||
| model.init_state_dict("model") | ||
|
|
||
| with jax.default_device(model.device): | ||
| observations = jnp.ones((10, model.num_observations)) | ||
| output = model.act({"states": observations}) | ||
| assert output[0].shape == (10, model.num_actions) | ||
| output = model.act({"states": flatten_tensorized_space(sample_space(observation_space, 10, "jax", device))}) | ||
| assert output[0].shape == (10, 2) | ||
|
|
||
|
|
||
| @hypothesis.given( | ||
| observation_space_size=st.integers(min_value=1, max_value=10), | ||
| action_space_size=st.integers(min_value=1, max_value=10), | ||
| ) | ||
| @hypothesis.settings(suppress_health_check=[hypothesis.HealthCheck.function_scoped_fixture], deadline=None) | ||
| @pytest.mark.parametrize("device", [None, "cpu", "cuda:0"]) | ||
| def test_gaussian_model(capsys, observation_space_size, action_space_size, device): | ||
| observation_space = gym.spaces.Box(np.array([-1] * observation_space_size), np.array([1] * observation_space_size)) | ||
| action_space = gym.spaces.Box(np.array([-1] * action_space_size), np.array([1] * action_space_size)) | ||
| # TODO: randomize all parameters | ||
| model = gaussian_model( | ||
| observation_space=observation_space, | ||
| action_space=action_space, | ||
| device=device, | ||
| clip_actions=False, | ||
| clip_log_std=True, | ||
| min_log_std=-20, | ||
| max_log_std=2, | ||
| initial_log_std=0, | ||
| input_shape=Shape.STATES, | ||
| hiddens=[256, 256], | ||
| hidden_activation=["relu", "relu"], | ||
| output_shape=Shape.ACTIONS, | ||
| output_activation=None, | ||
| output_scale=1, | ||
| ) | ||
| model.init_state_dict("model") | ||
| def test_gaussian_model(capsys, device): | ||
| # observation | ||
| action_space = spaces.Box(low=-1, high=1, shape=(2,)) | ||
| for observation_space_type in [spaces.Box, spaces.Tuple, spaces.Dict]: | ||
| observation_space = NETWORK_SPEC_OBSERVATION[observation_space_type][1] | ||
| model = gaussian_model( | ||
| observation_space=observation_space, | ||
| action_space=action_space, | ||
| device=device, | ||
| clip_actions=False, | ||
| clip_log_std=True, | ||
| min_log_std=-20, | ||
| max_log_std=2, | ||
| initial_log_std=0, | ||
| network=yaml.safe_load(NETWORK_SPEC_OBSERVATION[observation_space_type][0])["network"], | ||
| output="ACTIONS", | ||
| ) | ||
| model.init_state_dict("model") | ||
|
|
||
| with jax.default_device(model.device): | ||
| observations = jnp.ones((10, model.num_observations)) | ||
| output = model.act({"states": observations}) | ||
| assert output[0].shape == (10, model.num_actions) | ||
| output = model.act({"states": flatten_tensorized_space(sample_space(observation_space, 10, "jax", device))}) | ||
| assert output[0].shape == (10, 2) |