Bernoulli Target and Bernoulli Multi-Armed Bandit Surrogate #231
Conversation
julianStreibel
commented
May 4, 2024
julianStreibel
commented
- Added a bernoulli target class with just an identity transformation for now
- Added a bernoulli multi-armed bandit surrogate with a beta prior
- Added an example optimizing a bernoulli multi-armed bandit
julianStreibel
requested review from
Scienfitz,
AdrianSosic and
AVHopp
as code owners
|
|
|
Hi @julianStreibel, awesome that you already drafted it 🥇 Let's sit together on Monday and talk about what needs to be changed and remaining things to be modified 👌 @AVHopp @Scienfitz: I'be talked to Julian about this feature last week already. Let me have a first iteration with him before we go into the regular review since I would like to discuss the general direction of this PR with him first ✌️ |
AdrianSosic
force-pushed
the
feature/bernoulli_target
branch
from
37da056
to
03be257
Compare
|
@AdrianSosic @julianStreibel would you mind making this a Draft PR until the two of you are done with the first iteration? Would make it a bit easier for me and @Scienfitz to check when "proper" PR reviews should start :) |
julianStreibel
force-pushed
the
feature/bernoulli_target
branch
from
d79d7e6
to
1389b1d
Compare
julianStreibel
force-pushed
the
feature/bernoulli_target
branch
from
c75bf8d
to
328aa7f
Compare
| @@ -186,6 +191,9 @@ def add_fake_results( | |||
|
|
|||
| # Add the fake data for each target | |||
| for target in campaign.targets: | |||
| if isinstance(target, BinaryTarget): | |||
| data[target.name] = np.random.choice([0, 1]) | |||
There was a problem hiding this comment.
use target.accepted_values
This PR moves our priors from `baybe.kernels.priors` to `baybe.priors` since they can be used in general non-kernel contexts, for instance, as in #231.
AdrianSosic
added
the
on hold
| @@ -0,0 +1 @@ | |||
| """Multi-armed bandit examples.""" | |||
There was a problem hiding this comment.
@julianStreibel @AdrianSosic just sharing the following ICML paper here with you in case this is interesting for this or a future PR in this direction: https://proceedings.mlr.press/v235/jun24a.html
|
Closed in favor of #343 |
