Support new Avalanche version

#8 · closed · 3 comments

NickLucche

This new version (https://github.com/ContinualAI/avalanche/commit/4380ebc1b87b6034120c915249cde660e6d432e4) includes a refactoring of the RLScenario and RLExperience using the newly defined benchmark stack components. In order to support it, we must also modify our `RLStrategy` to inherit from the "BaseTemplate" (https://github.com/ContinualAI/avalanche/tree/master/avalanche/training/templates), this is the newest "hierarchy in town" that just joined the Avalanche training module. This integration should represent a significant step toward an eventual merge of Avalanche RL into Avalanche.

Comments

NickLucche

Other significant changes in avl since version 0.0.1 include Plugins (TODO: BasePlugin->RLPlugin) and loggers

NickLucche

also we should be installing it from the master branch, the beta version is the one you get when you run pip

NickLucche

A solution to this is currently being implemented here https://github.com/ContinualAI/avalanche-rl/tree/new-avalanche-version. It involves a workaround to keep the old avalanche `BaseStrategy` [around](https://github.com/ContinualAI/avalanche-rl/blob/new-avalanche-version/avalanche_rl/compat.py) before fully switching to the Templates model.