Need more understanding for the method add_arm about contextualbandits HOT 1 CLOSED

It's used when you want to add a new arm to an object which has already been fit to data which didn't initially include it, and it works in the same way as with the initial arms: that is, if you used beta_prior, it will use it for the predictions; if you used smoothing, will use it for the predictions while assuming the regular outputs from the model are zero or some small number; and if you use neither, will assume the arm always gets a prediction of zero until you fit it to data from that arm (so in this last case it's not going to be chosen until you later fit it to data from that arm collected externally).

The rest is as explained in the documentation - you can pass counts which will be used for beta_prior and smoothing, can pass data to use with refit_buffer next time you call partial_fit, or can pass a fitted classifier, or all of them together.

from contextualbandits.