alpha_ and lambda_ are not necessary for contextual linear bandit algorithms
Kurorororo opened this issue · comments
Currently, contextual linear and logistic bandit algorithms share the same superclass BaseContextualPolicy
.
The constructor of BaseContextualPolicy
has alpha_ and lambda_ as arguments:
Lines 93 to 129 in c9ad20c
These arguments are used to initialize self.alpha_list
and self.lambda_list
, which are used by LogisticEpsilonGreedy
, LogisticTS
, and LogisticUCB
but not used by LinearEpsilonGreedy
, LinTS
, and LinUCB
.
I suggest moving alpha_
, lambda_
, self.alpha_list
, and self.lambda_list
to another class, BaseLogisticPolicy
for example, and making logistic policies inherit this new class.
@Kurorororo Can you fix this issue by yourself after finishing #78 ?
Sure!