recpack.postprocessing

Module with classes for postprocessing recommendations

Postprocessor

The post processor provides all functionality to bundle postprocessing in one step. This makes it less prone to error, when applying the same processing to different recommendation outputs. It also makes initialisation more declarative, rather than having to chain outputs yourself.

Postprocessor()

Class to postprocess a CSR Response Matrix.

Filters

Filters provide functionality to retain only a part of the generated recommendations, setting the rest to 0.

PostFilter()	Abstract baseclass for postprocessing filter implementations
ExcludeItems(items)	Remove the recommendations of specified items.
SelectItems(items)	Select only the recommendations of specified items.

Filters can be applied manually, simply pass the csr_matrix to be processed to the apply function.:

import numpy as np
from scipy.sparse import csr_matrix

from recpack.postprocessing.filters import ExcludeItems

# Generate random recommendations
AMOUNT_OF_USERS = 20
AMOUNT_OF_ITEMS = 5
recommendations = csr_matrix(np.random.random_sample(size=(AMOUNT_OF_USERS, AMOUNT_OF_ITEMS)))

# Remove all recommendations for items 1 and 3
items_to_remove = [1, 3]
filter = ExcludeItems(items_to_remove)
processed = filter.apply(recommendations)

The preferred way to use filters though is through the recpack.postprocessing.postprocessors.Postprocessor. That way all postprocessing happens in a more controlled way, leaving less room for error.:

import numpy as np
from scipy.sparse import csr_matrix

from recpack.postprocessing.filters import ExcludeItems, RemovePreviousInteractions
from recpack.postprocessing.postprocessors import Postprocessor

# Generate random recommendations
AMOUNT_OF_USERS = 20
AMOUNT_OF_ITEMS = 5
recommendations_1 = csr_matrix(np.random.random_sample(size=(AMOUNT_OF_USERS, AMOUNT_OF_ITEMS)))
recommendations_2 = csr_matrix(np.random.random_sample(size=(AMOUNT_OF_USERS, AMOUNT_OF_ITEMS)))

# Construct processor which removes items 1 and 3.
processor = Postprocessor()
processor.add_filter(ExcludeItems([1, 3]))

# Apply processing step to both recommendation matrices.
processed_1, processed_2 = process.process_many(recommendations_1, recommendations_2)