Skip to content

Commit

Permalink
Merge branch 'main' into develop
Browse files Browse the repository at this point in the history
  • Loading branch information
@Lookatator
Lookatator committed Apr 17, 2024
2 parents 30ea208 + eb2cf31 commit 2b2dfc9
Show file tree
Hide file tree
Showing 27 changed files with 686 additions and 306 deletions.
6 changes: 4 additions & 2 deletions docs/guides/CONTRIBUTING.md
View file
Original file line number Diff line number Diff line change
@@ -1,8 +1,10 @@
# Contributing

We'd love to accept your patches and contributions to this project. When contributing to the repository, please make sure to first discuss the changes you wish to make via an issue.
We'd love to accept your patches and contributions to this project. When contributing to the repository, please make sure to first discuss the changes you wish to make via a [github issue](https://github.com/adaptive-intelligent-robotics/QDax/issues).

Then, there are just a few small guidelines you need to follow.
After the issue is discussed and the solution is determined, you will be invited to fork the repository and create a branch to implement the solution. Once ready to be merged, you can create a [Pull Request](https://github.com/adaptive-intelligent-robotics/QDax/pulls) on github and request to merge into the branch **develop**.

When implementing your contribution, there are just a few guidelines you need to follow.

## Installing Pre-commit hooks

Expand Down
202 changes: 108 additions & 94 deletions examples/aurora.ipynb
View file
Original file line number Diff line number Diff line change
Expand Up @@ -49,10 +49,28 @@
"try:\n",
" import brax\n",
"except:\n",
" !pip install git+https://github.com/google/brax.git@v0.0.15 |tail -n 1\n",
" !pip install git+https://github.com/google/brax.git@v0.9.2 |tail -n 1\n",
" import brax\n",
"\n",
"try:\n",
" import flax\n",
"except:\n",
" !pip install --no-deps git+https://github.com/google/[email protected] |tail -n 1\n",
" import flax\n",
"\n",
"try:\n",
" import chex\n",
"except:\n",
" !pip install --no-deps git+https://github.com/deepmind/[email protected] |tail -n 1\n",
" import chex\n",
"\n",
"try:\n",
" import jumanji\n",
"except:\n",
" !pip install \"jumanji==0.3.1\"\n",
" import jumanji\n",
"\n",
"try:\n",
" import qdax\n",
"except:\n",
" !pip install --no-deps git+https://github.com/adaptive-intelligent-robotics/QDax@main |tail -n 1\n",
Expand All @@ -62,13 +80,20 @@
"from qdax.core.aurora import AURORA\n",
"from qdax.core.containers.unstructured_repertoire import UnstructuredRepertoire\n",
"from qdax import environments\n",
"from qdax.tasks.brax_envs import scoring_aurora_function\n",
"from qdax.environments.bd_extractors import get_aurora_bd\n",
"from qdax.tasks.brax_envs import (\n",
" create_default_brax_task_components,\n",
" get_aurora_scoring_fn,\n",
")\n",
"from qdax.environments.bd_extractors import (\n",
" AuroraExtraInfoNormalization,\n",
" get_aurora_encoding,\n",
")\n",
"from qdax.core.neuroevolution.buffers.buffer import QDTransition\n",
"from qdax.core.neuroevolution.networks.networks import MLP\n",
"from qdax.core.emitters.mutation_operators import isoline_variation\n",
"from qdax.core.emitters.standard_emitters import MixingEmitter\n",
"\n",
"from qdax.types import Observation\n",
"from qdax.utils import train_seq2seq\n",
"\n",
"\n",
Expand Down Expand Up @@ -184,7 +209,7 @@
" \"\"\"\n",
"\n",
" actions = policy_network.apply(policy_params, env_state.obs)\n",
" \n",
"\n",
" state_desc = env_state.info[\"state_descriptor\"]\n",
" next_state = env.step(env_state, actions)\n",
"\n",
Expand All @@ -208,7 +233,7 @@
"source": [
"## Define the scoring function and the way metrics are computed\n",
"\n",
"The scoring function is used in the evaluation step to determine the fitness and behavior descriptor of each individual. "
"The scoring function is used in the evaluation step to determine the fitness and behavior descriptor of each individual."
]
},
{
Expand All @@ -218,19 +243,35 @@
"outputs": [],
"source": [
"# Prepare the scoring function\n",
"bd_extraction_fn = functools.partial(\n",
" get_aurora_bd,\n",
" option=observation_option,\n",
" hidden_size=hidden_size,\n",
" traj_sampling_freq=traj_sampling_freq,\n",
" max_observation_size=max_observation_size,\n",
"env, policy_network, scoring_fn, random_key = create_default_brax_task_components(\n",
" env_name=env_name,\n",
" random_key=random_key,\n",
")\n",
"scoring_fn = functools.partial(\n",
" scoring_aurora_function,\n",
" init_states=init_states,\n",
" episode_length=episode_length,\n",
" play_step_fn=play_step_fn,\n",
" behavior_descriptor_extractor=bd_extraction_fn,\n",
"\n",
"def observation_extractor_fn(\n",
" data: QDTransition,\n",
") -> Observation:\n",
" \"\"\"Extract observation from the state.\"\"\"\n",
" state_obs = data.obs[:, ::traj_sampling_freq, :max_observation_size]\n",
"\n",
" # add the x/y position - (batch_size, traj_length, 2)\n",
" state_desc = data.state_desc[:, ::traj_sampling_freq]\n",
"\n",
" if observation_option == \"full\":\n",
" observations = jnp.concatenate([state_desc, state_obs], axis=-1)\n",
" elif observation_option == \"no_sd\":\n",
" observations = state_obs\n",
" elif observation_option == \"only_sd\":\n",
" observations = state_desc\n",
" else:\n",
" raise ValueError(\"Unknown observation option.\")\n",
"\n",
" return observations\n",
"\n",
"# Prepare the scoring function\n",
"aurora_scoring_fn = get_aurora_scoring_fn(\n",
" scoring_fn=scoring_fn,\n",
" observation_extractor_fn=observation_extractor_fn,\n",
")\n",
"\n",
"# Get minimum reward value to make sure qd_score are positive\n",
Expand Down Expand Up @@ -290,13 +331,6 @@
"metadata": {},
"outputs": [],
"source": [
"# Instantiate AURORA\n",
"aurora = AURORA(\n",
" scoring_function=scoring_fn,\n",
" emitter=mixing_emitter,\n",
" metrics_function=metrics_fn,\n",
")\n",
"\n",
"aurora_dims = hidden_size\n",
"centroids = jnp.zeros(shape=(num_centroids, aurora_dims))\n",
"\n",
Expand All @@ -306,23 +340,19 @@
" (\n",
" repertoire,\n",
" random_key,\n",
" model_params,\n",
" mean_observations,\n",
" std_observations,\n",
" aurora_extra_info\n",
" ) = carry\n",
"\n",
" # update\n",
" (repertoire, _, metrics, random_key,) = aurora.update(\n",
" repertoire,\n",
" None,\n",
" random_key,\n",
" model_params,\n",
" mean_observations,\n",
" std_observations,\n",
" aurora_extra_info=aurora_extra_info,\n",
" )\n",
"\n",
" return (\n",
" (repertoire, random_key, model_params, mean_observations, std_observations),\n",
" (repertoire, random_key, aurora_extra_info),\n",
" metrics,\n",
" )\n",
"\n",
Expand All @@ -344,38 +374,67 @@
"else:\n",
" ValueError(\"The chosen option is not correct.\")\n",
"\n",
"# define the seq2seq model\n",
"# Define the seq2seq model\n",
"model = train_seq2seq.get_model(\n",
" observations_dims[-1], True, hidden_size=hidden_size\n",
")\n",
"\n",
"# init the model params\n",
"# Init the model params\n",
"random_key, subkey = jax.random.split(random_key)\n",
"model_params = train_seq2seq.get_initial_params(\n",
" model, subkey, (1, *observations_dims)\n",
")\n",
"\n",
"print(jax.tree_map(lambda x: x.shape, model_params))\n",
"\n",
"# Define the encoder function\n",
"encoder_fn = jax.jit(\n",
" functools.partial(\n",
" get_aurora_encoding,\n",
" model=model,\n",
" )\n",
")\n",
"\n",
"# Define the training function\n",
"train_fn = functools.partial(\n",
" train_seq2seq.lstm_ae_train,\n",
" model=model,\n",
" batch_size=lstm_batch_size,\n",
")\n",
"\n",
"# Instantiate AURORA\n",
"aurora = AURORA(\n",
" scoring_function=aurora_scoring_fn,\n",
" emitter=mixing_emitter,\n",
" metrics_function=metrics_fn,\n",
" encoder_function=encoder_fn,\n",
" training_function=train_fn,\n",
")\n",
"\n",
"# define arbitrary observation's mean/std\n",
"mean_observations = jnp.zeros(observations_dims[-1])\n",
"std_observations = jnp.ones(observations_dims[-1])\n",
"\n",
"# init step of the aurora algorithm\n",
"repertoire, _, random_key = aurora.init(\n",
" init_variables,\n",
" centroids,\n",
" random_key,\n",
"# init all the information needed by AURORA to compute encodings\n",
"aurora_extra_info = AuroraExtraInfoNormalization.create(\n",
" model_params,\n",
" mean_observations,\n",
" std_observations,\n",
" l_value_init,\n",
")\n",
"\n",
"# init step of the aurora algorithm\n",
"repertoire, emitter_state, aurora_extra_info, random_key = aurora.init(\n",
" init_variables,\n",
" aurora_extra_info,\n",
" jnp.asarray(l_value_init),\n",
" max_observation_size,\n",
" random_key,\n",
")\n",
"\n",
"# initializing means and stds and AURORA\n",
"random_key, subkey = jax.random.split(random_key)\n",
"model_params, mean_observations, std_observations = train_seq2seq.lstm_ae_train(\n",
" subkey, repertoire, model_params, 0, hidden_size=hidden_size, batch_size=lstm_batch_size\n",
"repertoire, aurora_extra_info = aurora.train(\n",
" repertoire, model_params, iteration=0, random_key=subkey\n",
")\n",
"\n",
"# design aurora's schedule\n",
Expand Down Expand Up @@ -409,11 +468,11 @@
"while iteration < max_iterations:\n",
"\n",
" (\n",
" (repertoire, random_key, model_params, mean_observations, std_observations),\n",
" (repertoire, random_key, aurora_extra_info),\n",
" metrics,\n",
" ) = jax.lax.scan(\n",
" update_scan_fn,\n",
" (repertoire, random_key, model_params, mean_observations, std_observations),\n",
" (repertoire, random_key, aurora_extra_info),\n",
" (),\n",
" length=log_freq,\n",
" )\n",
Expand All @@ -427,60 +486,15 @@
" if (iteration + 1) in schedules:\n",
" # train the autoencoder\n",
" random_key, subkey = jax.random.split(random_key)\n",
" (\n",
" model_params,\n",
" mean_observations,\n",
" std_observations,\n",
" ) = train_seq2seq.lstm_ae_train(\n",
" subkey,\n",
" repertoire,\n",
" model_params,\n",
" iteration,\n",
" hidden_size=hidden_size,\n",
" batch_size=lstm_batch_size\n",
" repertoire, aurora_extra_info = aurora.train(\n",
" repertoire, model_params, iteration, subkey\n",
" )\n",
"\n",
" # re-addition of all the new behavioural descriotpors with the new ae\n",
" normalized_observations = (\n",
" repertoire.observations - mean_observations\n",
" ) / std_observations\n",
"\n",
" new_descriptors = model.apply(\n",
" {\"params\": model_params}, normalized_observations, method=model.encode\n",
" )\n",
" repertoire = repertoire.init(\n",
" genotypes=repertoire.genotypes,\n",
" centroids=repertoire.centroids,\n",
" fitnesses=repertoire.fitnesses,\n",
" descriptors=new_descriptors,\n",
" observations=repertoire.observations,\n",
" l_value=repertoire.l_value,\n",
" )\n",
" num_indivs = jnp.sum(repertoire.fitnesses != -jnp.inf)\n",
"\n",
" elif iteration % 2 == 0:\n",
" # update the l value\n",
" num_indivs = jnp.sum(repertoire.fitnesses != -jnp.inf)\n",
"\n",
" # CVC Implementation to keep a constant number of individuals in the archive\n",
" current_error = num_indivs - n_target\n",
" change_rate = current_error - previous_error\n",
" prop_gain = 1 * 10e-6\n",
" l_value = (\n",
" repertoire.l_value\n",
" + (prop_gain * (current_error))\n",
" + (prop_gain * change_rate)\n",
" )\n",
"\n",
" previous_error = current_error\n",
"\n",
" repertoire = repertoire.init(\n",
" genotypes=repertoire.genotypes,\n",
" centroids=repertoire.centroids,\n",
" fitnesses=repertoire.fitnesses,\n",
" descriptors=repertoire.descriptors,\n",
" observations=repertoire.observations,\n",
" l_value=l_value,\n",
" repertoire, previous_error = aurora.container_size_control(\n",
" repertoire,\n",
" target_size=n_target,\n",
" previous_error=previous_error,\n",
" )\n",
"\n",
" iteration += 1"
Expand Down
Loading

0 comments on commit 2b2dfc9

Please sign in to comment.