Skip to content

Latest commit

 

History

History
14 lines (10 loc) · 798 Bytes

README.md

File metadata and controls

14 lines (10 loc) · 798 Bytes

Temporal-Difference-Models-with-Value-Function-Based-sampling

Keras implementation of Temporal Difference Models by V. Pong et al.(2018) + Value based sampling

Requirements:

  • Keras
  • Tensorflow or Theano as backend
  • Gym - Robotics environment along with Mujoco environment

The structure of this code is built on Keras-rl

A few tweaks that we did -

  1. Relabelling goals based on expected reward henceforth, with some probability. We found that it lead to faster convergence in the FetchReach environment.
  2. Decayed the 'goal reached' condition radius gradually. It lead to faster convergence as well