search

IntelLabs / coach

Reinforcement Learning Coach by Intel AI Lab enables easy experimentation with state of the art Reinforcement Learning algorithms
https://intellabs.github.io/coach/
Apache License 2.0
2.32k stars 459 forks source link

Removed unnecessary observation #241

Closed allan-avatar1 closed 5 years ago

allan-avatar1 commented 5 years ago

unless i am missing something, 'achieved_goal' is the exact same thing as 'state' ... so there is probably no need to duplicate it in the observations. referencing this file shows that originally there was only two variables in the observation space 1) the current state 2) the goal state https://github.com/NervanaSystems/gym-bit-flip/blob/master/gym_bit_flip/bit_flip.py

gal-leibovich commented 5 years ago

Although in some environments (bitflip included), these seem to be the same, achieved_goal is not necessarily the same as state. For instance, in the 'Fetch' environment (which is used in the Fetch_DDPG_HER_baselines preset), these are not the same.

You can also take a look at the Gym API and the corresponding blog-post for more details.