Initial commit

.gitignore

@@ -0,0 +1,8 @@
+*.DS_Store
+__pycache__/
+env/
+*.pyc
+*.pkl
+*.p
+mujoco/mjpro131
+*.h5

LICENSE

@@ -0,0 +1,43 @@
+deep-neuroevolution
+Copyright (c) 2018 Uber Technologies, Inc.
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in
+all copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
+THE SOFTWARE.
+
+deep-neuroevolution includes:
+
+evolution-strategies-starter
+Copyright (c) 2016 OpenAI (http://openai.com) 
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in
+all copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
+THE SOFTWARE.

README.md

@@ -0,0 +1,78 @@
+## AI Labs Neuroevolution Algorithms
+
+This repo contains distributed implementations of the algorithms described in:
+
+[1] [Deep Neuroevolution: Genetic Algorithms Are a Competitive Alternative for Training Deep Neural Networks for Reinforcement Learning](https://arxiv.org/abs/1712.06567)
+
+[2] [Improving Exploration in Evolution Strategies for Deep Reinforcement Learning via a Population of Novelty-Seeking Agents](https://arxiv.org/abs/1712.06560)
+
+Our code is based off of code from OpenAI, who we thank. The original code and related paper from OpenAI can be found here. The repo has been modified to run both ES and our algorithms, including our Deep Genetic Algorithm (DeepGA) locally and on AWS.
+
+Note: The Humanoid experiment depends on [Mujoco](http://www.mujoco.org/). Please provide your own Mujoco license and binary
+
+The blog post describing these papers can be found [here](https://eng.uber.com/deep-neuroevolution/)
+
+## How to run locally
+
+clone repo
+
+```
+git clone https://github.com/uber-common/deep-neuroevolution.git
+```
+
+create python3 virtual env
+
+```
+python3 -m venv env
+. env/bin/activate
+```
+
+install requirements
+```
+pip install -r requirements.txt
+```
+If you plan to use the mujoco env, make sure to follow [mujoco-py](https://github.com/openai/mujoco-py)'s readme about how to install mujoco correctly
+
+launch redis
+```
+. scripts/local_run_redis.sh
+```
+
+launch sample ES experiment
+```
+. scripts/local_run_exp.sh es configurations/frostbite_es.json  # For the Atari game Frostbite
+. scripts/local_run_exp.sh es configurations/humanoid.json  # For the MuJoCo Humanoid-v1 environment
+```
+
+launch sample NS-ES experiment
+```
+. scripts/local_run_exp.sh ns-es configurations/frostbite_nses.json
+. scripts/local_run_exp.sh ns-es configurations/humanoid_nses.json
+```
+
+launch sample NSR-ES experiment
+```
+. scripts/local_run_exp.sh nsr-es configurations/frostbite_nsres.json
+. scripts/local_run_exp.sh nsr-es configurations/humanoid_nsres.json
+```
+
+launch sample GA experiment
+```
+. scripts/local_run_exp.sh ga configurations/frostbite_ga.json  # For the Atari game Frostbite
+```
+
+launch sample Random Search experiment
+```
+. scripts/local_run_exp.sh rs configurations/frostbite_ga.json  # For the Atari game Frostbite
+```
+
+
+visualize results by running a policy file
+```
+python -m scripts.viz 'FrostbiteNoFrameskip-v4' <YOUR_H5_FILE>
+python -m scripts.viz 'Humanoid-v1' <YOUR_H5_FILE>
+```
+
+### extra folder
+The extra folder holds the XML specification file for the  Humanoid
+Locomotion with Deceptive Trap domain used in https://arxiv.org/abs/1712.06560. Use this XML file in gym to recreate the environment.

configurations/frostbite_es.json

@@ -0,0 +1,24 @@
+{
+  "config": {
+    "calc_obstat_prob": 0.0,
+    "episodes_per_batch": 5000,
+    "eval_prob": 0.01,
+    "l2coeff": 0.005,
+    "noise_stdev": 0.005,
+    "snapshot_freq": 20,
+    "timesteps_per_batch": 10000,
+    "return_proc_mode": "centered_rank",
+    "episode_cutoff_mode": 5000
+  },
+  "env_id": "FrostbiteNoFrameskip-v4",
+  "optimizer": {
+    "args": {
+      "stepsize": 0.01
+    },
+    "type": "adam"
+  },
+  "policy": {
+    "args" : {},
+    "type": "ESAtariPolicy"
+  }
+}

configurations/frostbite_ga.json

@@ -0,0 +1,22 @@
+{
+  "config": {
+    "calc_obstat_prob": 0.0,
+    "episodes_per_batch": 5000,
+    "eval_prob": 0.01,
+    "l2coeff": 0.005,
+    "noise_stdev": 0.005,
+    "snapshot_freq": 20,
+    "timesteps_per_batch": 10000,
+    "return_proc_mode": "centered_rank",
+    "episode_cutoff_mode": 5000
+  },
+  "population_size": 10,
+  "num_elites": 1,
+  "env_id": "FrostbiteNoFrameskip-v4",
+  "policy": {
+    "args" : {
+      "nonlin_type": "relu"
+    },
+    "type": "GAAtariPolicy"
+  }
+}

configurations/frostbite_nses.json

@@ -0,0 +1,31 @@
+{
+  "config": {
+    "calc_obstat_prob": 0.0,
+    "episodes_per_batch": 100,
+    "eval_prob": 0.03,
+    "l2coeff": 0.005,
+    "noise_stdev": 0.02,
+    "snapshot_freq": 10,
+    "timesteps_per_batch": 1000,
+    "return_proc_mode": "centered_sign_rank",
+    "episode_cutoff_mode": 5000
+  },
+  "env_id": "FrostbiteNoFrameskip-v4",
+  "algo_type": "ns",
+  "novelty_search": {
+    "k": 10,
+    "population_size": 3,
+    "num_rollouts": 1,
+    "selection_method": "novelty_prob"
+  },
+  "optimizer": {
+    "args": {
+      "stepsize": 0.01
+    },
+    "type": "adam"
+  },
+  "policy": {
+    "args": {},
+    "type": "ESAtariPolicy"
+  }
+}

configurations/frostbite_nsres.json

@@ -0,0 +1,31 @@
+{
+  "config": {
+    "calc_obstat_prob": 0.0,
+    "episodes_per_batch": 1000,
+    "eval_prob": 0.1,
+    "l2coeff": 0.005,
+    "noise_stdev": 0.02,
+    "snapshot_freq": 10,
+    "timesteps_per_batch": 10000,
+    "return_proc_mode": "centered_sign_rank",
+    "episode_cutoff_mode": 5000
+  },
+  "env_id": "FrostbiteNoFrameskip-v4",
+  "algo_type": "nsr",
+  "novelty_search": {
+    "k": 10,
+    "population_size": 3,
+    "num_rollouts": 1,
+    "selection_method": "novelty_prob"
+  },
+  "optimizer": {
+    "args": {
+      "stepsize": 0.01
+    },
+    "type": "adam"
+  },
+  "policy": {
+    "args": {},
+    "type": "ESAtariPolicy"
+  }
+}

configurations/humanoid.json

@@ -0,0 +1,34 @@
+{
+  "config": {
+    "calc_obstat_prob": 0.01,
+    "episodes_per_batch": 1000,
+    "eval_prob": 0.03,
+    "l2coeff": 0.005,
+    "noise_stdev": 0.02,
+    "snapshot_freq": 10,
+    "timesteps_per_batch": 100000,
+    "return_proc_mode": "centered_rank",
+    "episode_cutoff_mode": "env_default"
+  },
+  "env_id": "Humanoid-v1",
+  "exp_prefix": "humanoid",
+  "optimizer": {
+    "args": {
+      "stepsize": 0.01
+    },
+    "type": "adam"
+  },
+  "policy": {
+    "args": {
+      "ac_bins": "continuous:",
+      "ac_noise_std": 0.01,
+      "connection_type": "ff",
+      "hidden_dims": [
+        256,
+        256
+      ],
+      "nonlin_type": "tanh"
+    },
+    "type": "MujocoPolicy"
+  }
+}

configurations/humanoid_nses.json

@@ -0,0 +1,41 @@
+{
+  "config": {
+    "calc_obstat_prob": 0.01,
+    "episodes_per_batch": 1000,
+    "eval_prob": 0.03,
+    "l2coeff": 0.005,
+    "noise_stdev": 0.02,
+    "snapshot_freq": 10,
+    "timesteps_per_batch": 100000,
+    "return_proc_mode": "centered_sign_rank",
+    "episode_cutoff_mode": "env_default"
+  },
+  "env_id": "Humanoid-v1",
+  "algo_type": "ns",
+  "exp_prefix": "humanoid",
+  "novelty_search": {
+    "k": 10,
+    "population_size": 5,
+    "num_rollouts": 5,
+    "selection_method": "novelty_prob"
+  },
+  "optimizer": {
+    "args": {
+      "stepsize": 0.01
+    },
+    "type": "adam"
+  },
+  "policy": {
+    "args": {
+      "ac_bins": "continuous:",
+      "ac_noise_std": 0.01,
+      "connection_type": "ff",
+      "hidden_dims": [
+        256,
+        256
+      ],
+      "nonlin_type": "tanh"
+    },
+    "type": "MujocoPolicy"
+  }
+}

configurations/humanoid_nsres.json

@@ -0,0 +1,41 @@
+{
+  "config": {
+    "calc_obstat_prob": 0.01,
+    "episodes_per_batch": 1000,
+    "eval_prob": 0.03,
+    "l2coeff": 0.005,
+    "noise_stdev": 0.02,
+    "snapshot_freq": 10,
+    "timesteps_per_batch": 100000,
+    "return_proc_mode": "centered_sign_rank",
+    "episode_cutoff_mode": "env_default"
+  },
+  "env_id": "Humanoid-v1",
+  "algo_type": "nsr",
+  "exp_prefix": "humanoid",
+  "novelty_search": {
+    "k": 10,
+    "population_size": 5,
+    "num_rollouts": 5,
+    "selection_method": "novelty_prob"
+  },
+  "optimizer": {
+    "args": {
+      "stepsize": 0.01
+    },
+    "type": "adam"
+  },
+  "policy": {
+    "args": {
+      "ac_bins": "continuous:",
+      "ac_noise_std": 0.01,
+      "connection_type": "ff",
+      "hidden_dims": [
+        256,
+        256
+      ],
+      "nonlin_type": "tanh"
+    },
+    "type": "MujocoPolicy"
+  }
+}