Recommender

class gobrec.recommender.Recommender(mab_algo: MABAlgo, top_k: int)

GOBRec recommender class.

Recommender that uses a MAB algorithm to generate top-K recommendations based on contexts.

Examples

A simple example using LinUCB as the MAB algorithm to recommend items.

>>> import numpy as np
>>> import gobrec
>>> from gobrec.mabs.lin_mabs import LinUCB
>>> contexts = np.array([[1, 0, 0], [0, 1, 0], [0, 0, 1],
...                      [1, 0, 0], [0, 1, 0], [0, 0, 1],
...                      [1, 0, 0], [0, 1, 0], [0, 0, 1]])
>>> decisions = np.array(['a', 'a', 'a', 
...                       'b', 'b', 'b',
...                       'c', 'c', 'c'])
>>> rewards = np.array([10, 0 , 1 , 
...                     1 , 10, 0 ,
...                     0 , 1 , 10])
>>> recommender = gobrec.Recommender(mab_algo=LinUCB(seed=42, use_gpu=True),top_k=2)
>>> recommender.fit(contexts, decisions, rewards)
>>> recommender.recommend(np.array([[1, 0, 0], [0, 1, 0], [0, 0, 1]]))
(array([['a', 'b'],  # Top-2 recommendations for each context. Each value is an item ID.
        ['b', 'c'],
        ['c', 'a']]),
 array([[5.70710678, 1.20710678],   # Corresponding scores for each recommended item.
        [5.70710678, 1.20710678],
        [5.70710678, 1.20710678]]))

Example using LinTS as the MAB algorithm to recommend items with item filtering.

>>> import numpy as np
>>> import gobrec
>>> from gobrec.mabs.lin_mabs import LinTS
>>> contexts = np.array([[1, 0, 0], [0, 1, 0], [0, 0, 1],
...                      [1, 0, 0], [0, 1, 0], [0, 0, 1],
...                      [1, 0, 0], [0, 1, 0], [0, 0, 1]])
>>> decisions = np.array(['a', 'a', 'a', 
...                       'b', 'b', 'b',
...                       'c', 'c', 'c'])
>>> rewards = np.array([10, 0 , 1 , 
...                     1 , 10, 0 ,
...                     0 , 1 , 10])
>>> recommender = gobrec.Recommender(mab_algo=LinTS(seed=42, use_gpu=True),top_k=2)
>>> recommender.fit(contexts, decisions, rewards)
>>> recommender.recommend(np.array([[1, 0, 0], [0, 1, 0], [0, 0, 1]]),
...                       decisions_filter=[[0, 1, 2], ['a', 'b', 'c']])
...                       # The first element is the indices of contexts to filter
...                       # The second element is the list of item IDs to filter
...                       # In this example, it will filter item a from context 0, 
...                       # item b from context 1 and item c from context 2
(array([['b', 'c'],  # Top-2 recommendations for each context. Each value is an item ID.
        ['c', 'a'],
        ['a', 'b']]),
 array([[ 0.71546751,  0.21546751],    # Corresponding scores for each recommended item.
        [-0.87959021, -1.37959021],
        [ 0.48811979, -0.01188021]]))

Attributes:

mab_algoMABAlgo

The multi-armed bandit algorithm used to generate item scores.

top_kint

The number of items to be recommended. It will return the K highest scored items.

Methods

fit(contexts, decisions, rewards)	Trains the MAB algorithm with the provided data.
recommend(contexts[, decisions_filter])	Predicts top-K recommendations for the given contexts.
reset()	Reset the internal state of the MAB algorithm.

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

Trains the MAB algorithm with the provided data.

Parameters:

contextsnp.ndarray

A 2D array where each row represents the context features.

decisionsnp.ndarray

A 1D array where each element is the item ID (decision) taken for the corresponding context.

rewardsnp.ndarray

A 1D array where each element is the reward (rating) received for the corresponding context-decision pair.

recommend(contexts: ndarray, decisions_filter: list[np.ndarray, np.ndarray] = None)

Predicts top-K recommendations for the given contexts.

Parameters:

contextsnp.ndarray

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

decisions_filterlist[np.ndarray, np.ndarray], optional

A list containing two arrays:

1. contexts_indexes: A 1D array of indices representing the contexts for which certain items should be filtered out. 2. items_ids: A 1D array of item IDs to be filtered out for the corresponding contexts.

For example, if contexts_indexes is [0, 2] and items_ids is [‘itemA’, ‘itemB’], then ‘itemA’ will be filtered out from the recommendations for the context at index 0, and ‘itemB’ will be filtered out for the context at index 2. contexts_indexes can contain repeated indices if multiple items need to be filtered for the same context.

If decisions_filter is None, no items will be filtered from the recommendations.

Returns:

recommendationsnp.ndarray

A 2D array where each row contains the top-K recommended item IDs for the corresponding context.

scoresnp.ndarray

A 2D array where each row contains the scores of the top-K recommended items for the corresponding context.

reset()

Reset the internal state of the MAB algorithm.

This method is useful when you want to clear the learned parameters of the MAB algorithm and start fresh without any prior knowledge.

This will make the recommender behave as if it was just initialized.