TianhongDai / esil-hindsight

This is the official code of our paper "Episodic Self-Imitation Learning with Hindsight" [Electronics 2020].
MIT License
7 stars 2 forks source link
reinforcement-learning

Episodic Self-Imitation Learning with Hindsight (ESIL)

This is the official code of our paper "Episodic Self-Imitation Learning with Hindsight" [Electronics, Special Issue on Deep Reinforcement Learning: Methods and Applications 2020].

FetchPush-v1 FetchPickAndPlace-v1 FetchSlide-v1

Requirements

Installation:

Install the OpenMPI on the machine.

sudo apt-get install openmpi-bin openmpi-doc libopenmpi-dev

Instructions

- run the **FetchPush-v1**:
```bash
mpirun -np 16 python train.py --env-name='FetchPush-v1' --adaptive-beta --display-interval=1 --total-frames=2500000
- run the **FetchSlide-v1**:
```bash
mpirun -np 32 python train.py --env-name='FetchSlide-v1' --adaptive-beta --display-interval=1 --batch-size=40 --ncycles=100 --total-frames=5000000
## Plot Curves
The scirpt to plot learning curves in the paper is provided in `plot_curves.py`.

## BibTex
To cite this code for publications - please use:

@article{dai2020episodic, title={Episodic Self-Imitation Learning with Hindsight}, author={Dai, Tianhong and Liu, Hengyan and Anthony Bharath, Anil}, journal={Electronics}, volume={9}, number={10}, pages={1742}, year={2020}, publisher={Multidisciplinary Digital Publishing Institute} }