Random MAB

class gobrec.mabs.random_mab.RandomMAB(seed: int | None = None)

Random Multi-Armed Bandit (MAB) Algorithm.

This class implements a random MAB algorithm that generates random scores for each arm. It can be used as a baseline for comparison with more sophisticated MAB algorithms.

Examples

A simple example using RandomMAB to generate random scores for items.

>>> import numpy as np
>>> from gobrec.mabs import RandomMAB
>>> random_mab = RandomMAB(seed=2)
>>> random_mab.fit(
...    contexts=np.array([[0, 1],[0, 1], [1, 0], [1, 0]]),
...    decisions=np.array([0, 1, 0, 1]),
...    rewards=np.array([1, 0, 0, 1])
... )
>>> random_mab.predict(np.array([[0, 1],[0, 1], [1, 0], [1, 0]]))
tensor([[0.2616, 0.2985],
        [0.8142, 0.0919],
        [0.6001, 0.7286],
        [0.1879, 0.0551]], dtype=torch.float64)

Methods

fit(contexts, decisions, rewards)	Fit the RandomMAB algorithm with contexts, decisions, and rewards.
predict(contexts)	Predict random scores for each arm given the contexts.
reset()	Reset the RandomMAB algorithm to its initial state.

fit(contexts: ndarray, decisions: ndarray, rewards: ndarray)

Fit the RandomMAB algorithm with contexts, decisions, and rewards.

This method just updates the label encoder and internal state. There is no actual learning in this random algorithm.

Parameters:

contextsnp.ndarray

A 2D array of shape (n_samples, n_features) representing the context arrays

decisionsnp.ndarray

A 1D array of item IDs (arms or decisions) of shape (n_samples,) where each element can be strings or integers.

rewardsnp.ndarray

A 1D array of rewards (ratings) of shape (n_samples,). It can be integers or floats.

predict(contexts: ndarray)

Predict random scores for each arm given the contexts.

Parameters:

contextsnp.ndarray

A 2D array where each row represents the context features for which predictions are to be made.

Returns:

scorestorch.Tensor

A 2D tensor of shape (n_samples, n_arms) where each element is a random score for the corresponding context-arm pair.

reset()

Reset the RandomMAB algorithm to its initial state.

This method clears the labeled encoder and reinitializes the random number generator with the original seed.