fuyuan-li / tensorflow-RIMs

A tensorflow implementation of Recurrent Independent Mechanisms Goyal et al. 2019
Other
7 stars 3 forks source link

Compatibility with dynamic_unroll_layer #2

Open dvVenuto opened 3 years ago

dvVenuto commented 3 years ago

I was wondering if you would be interested in working on the TF-RIM compatibility with tf-agents, specifically the dynamic unroll function for use in RL described here: https://www.tensorflow.org/agents/api_docs/python/tf_agents/keras_layers/dynamic_unroll_layer

fuyuan-li commented 3 years ago

Hey David, I can't find any reason to say no to such an exciting project! Happy to follow up!

dvVenuto commented 3 years ago

Hello, I am at david.venuto@mail.mcgill.ca I can explain what my goals are and what i have done so far. The implementation shouldn't be hard. I think it may make a really nice addition to tf-agents and some research im working on since, as you know RIMs have been useful in RL but haven't really been implemented with an RL framework outside of pytorch.

dvVenuto commented 3 years ago

Hello Fuyuan, I have the implementation partially functional but with a few issues. I was wondering if you are still interested as it may make a nice addition to Tensorflow (especially in the tf-agents RL framework) in the future.