QDax

QDax copied to clipboard

Default Scoring Functions for Sphere, Rastrigin, Arm and Brax environments

1

- adding `sphere`, `rastrigin`, `arm`, and `noisy_arm` scoring functions - adding possibility to create default scoring_function for brax environments. - Changing type of `EnvState` to `brax.envs.State` - Fix some quick...

Lookatator

With the latest version of Jax (obtained automatically with pip install qdax) we get a lot of FutureWarnings: ``` FutureWarning: jax.tree_map is deprecated, and will be removed in a future...

Aneoshun

Add DvD

4

It would be nice to add Diversity via Determinants (DvD) to the repo. It is possible some of the ideas here could combine well with others to create exciting new...

jparkerholder

Re-write the `insert()` method of `Replay Buffer` to fix #71 and #70. This fix also avoids the error raised in #67. I tested this new buffer on the HalfCheetah with...

manon-but-yes

Hi :) The current PGAME Replay Buffer is using [jax.lax.dynamic_update_slice](https://jax.readthedocs.io/en/latest/_autosummary/jax.lax.dynamic_update_slice.html) to add new transition to the replay buffer. However, this is not acting like a circular buffer, meaning that if...

manon-but-yes

PGAME add dead transition to Replay Buffer

2

Hi :) It seems that the current implementation of PGAME is adding to the Replay-Buffer all the transitions that are collected in the environment, meaning also transitions that occur after...

manon-but-yes

Hello :), When running more memory intensive QD algorithms, it is possible to be out of memory when filling up the repertoire. I found a [comment](https://github.com/google/jax/issues/1222#issuecomment-622919670) on the jax repo,...

maxiallard

The function `_sample_in_masked_pareto_front` does not return an updated `random_key` (contrary to all other functions in the project). https://github.com/adaptive-intelligent-robotics/QDax/blob/50b997fd5795e0df2b80234ec9fc6772bce0b6a8/qdax/core/containers/mome_repertoire.py#L49-L61 This can probably be solved by using the `in_axis` and `out_axis` arguments...

Lookatator

Hi all, At the moment, PointMaze's action space is [-0.1, 0.1]² whereas all the other environments have an action space of [-1, 1]**n Wouldn't it be better to have the...

felixchalumeau

PGAME replay buffer addition fail for large batch-size

2

Hello :) The current implementation of the PGAME emitter is failing for large batch-size (here for example 32768) with the following error: ``` Traceback (most recent call last): File "main.py",...

manon-but-yes