[RLlib] Update bandit_envs_recommender_system #22421
Conversation
|
|
||
|
|
||
| class ParametricItemRecoEnv(gym.Env): | ||
| """A recommendation environment which generates items with visible features |
There was a problem hiding this comment.
Why did we remove this? I feel like we should explain, why it's called parametric. Or re-name it, as any of our recomm envs is somewhat parametric, no?
There was a problem hiding this comment.
oh good catch. yeah, I should have copied over the description.
done.
| def render(self, mode="human"): | ||
| raise NotImplementedError | ||
|
|
||
| # Because ParametricRecSys follows RecSim's API, we have to wrap it before |
There was a problem hiding this comment.
Should we do this here?
Maybe just move this into the block below and register it under "ParametricRecSysEnvForBandits" or something like this.
Also, in the example script that actually uses this env, we can register it to show, how this is done.
There was a problem hiding this comment.
I really like the idea of registering the env in example script. good demo.
thanks.
sven1977
added
the
tests-ok
|
|
||
|
|
||
| class ParametricItemRecoEnv(gym.Env): | ||
| """A recommendation environment which generates items with visible features |
There was a problem hiding this comment.
oh good catch. yeah, I should have copied over the description.
done.
| def render(self, mode="human"): | ||
| raise NotImplementedError | ||
|
|
||
| # Because ParametricRecSys follows RecSim's API, we have to wrap it before |
There was a problem hiding this comment.
I really like the idea of registering the env in example script. good demo.
thanks.
|
Hey @gjoliver , could you re-merge? Lots of tests failing, I'm thinking b/c we are too far behind master. |
gjoliver
force-pushed
the
update_rec_env_for_bandits
branch
from
dac3892
to
d0a1bdf
Compare
| ( | ||
| "item", | ||
| gym.spaces.Box( | ||
| low=-np.ones((num_items, embedding_dim)), | ||
| high=np.ones((num_items, embedding_dim)), | ||
| low=-1.0, high=1.0, shape=(num_items, embedding_dim) |
Why are these changes needed?
Update bandit_envs_recommender_system with Sven's in-house implementa…, which supports multiple users and slate sizes.
Make sure example tune_lin_ucb_train_recommendation.py nails this environment.
Related issue number
Checks
scripts/format.shto lint the changes in this PR.