VedantDave/DB
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity
6 Commits 1 Branch 0 Tags 522 KiB
Python 100%
55fa55fdff
Go to file
VedantDave 55fa55fdff Adding Files
2023-05-29 14:26:35 +02:00
baselines@ea25b9e8b2 Adding file 2023-05-29 13:32:06 +02:00
ROMS Adding ROMS 2023-05-29 13:40:45 +02:00
__init__.py DB 2021-10-08 19:33:47 -05:00
cnn_policy.py DB 2021-10-08 19:33:47 -05:00
cppo_agent.py DB 2021-10-08 19:33:47 -05:00
dynamic_bottleneck.py Adding Files 2023-05-29 13:37:22 +02:00
environment.yml Adding Files 2023-05-29 14:26:35 +02:00
mpi_utils.py DB 2021-10-08 19:33:47 -05:00
para.json DB 2021-10-08 19:33:47 -05:00
README.md DB 2021-10-08 19:33:47 -05:00
recorder.py DB 2021-10-08 19:33:47 -05:00
rollouts.py Adding Files 2023-05-29 13:37:22 +02:00
run.py Adding Files 2023-05-29 14:26:35 +02:00
utils.py Adding Files 2023-05-29 13:34:41 +02:00
vec_env.py DB 2021-10-08 19:33:47 -05:00
wrappers.py DB 2021-10-08 19:33:47 -05:00

README.md

Dynamic Bottleneck

Introduction

This is a TensorFlow based implementation for our paper on

"Dynamic Bottleneck for Robust Self-Supervised Exploration". NeurIPS 2021

Prerequisites

python3.6 or 3.7, tensorflow-gpu 1.x, tensorflow-probability, openAI baselines, openAI Gym

Installation and Usage

Atari games

The following command should train a pure exploration agent on "Breakout" with default experiment parameters.

python run.py --env BreakoutNoFrameskip-v4

Atari games with Random-Box noise

The following command should train a pure exploration agent on "Breakout" with randomBox noise.

python run.py --env BreakoutNoFrameskip-v4 --randomBoxNoise

Atari games with Gaussian noise

The following command should train a pure exploration agent on "Breakout" with Gaussian noise.

python run.py --env BreakoutNoFrameskip-v4 --pixelNoise

Atari games with sticky actions

The following command should train a pure exploration agent on "sticky Breakout" with a probability of 0.25

python run.py --env BreakoutNoFrameskip-v4 --stickyAtari

Baselines

  • ICM: We use the official code of "Curiosity-driven Exploration by Self-supervised Prediction, ICML 2017" and "Large-Scale Study of Curiosity-Driven Learning, ICLR 2019".
  • Disagreement: We use the official code of "Self-Supervised Exploration via Disagreement, ICML 2019".
  • CB: We use the official code of "Curiosity-Bottleneck: Exploration by Distilling Task-Specific Novelty, ICML 2019".