Recommender System

You are looking at the old Evidently documentation: this API is available with versions 0.6.7 or lower. Check the newer version here.

TL;DR: You can monitor and analyze the performance of a recommender system.

• Report: for visual analysis or metrics export, use the RecsysPreset.

• Test Suite: for pipeline checks, use the RecsysTestPreset.

Recommendation Quality Report

If you want to visually explore the model performance, create a new Report object and include the RecsysPreset.

Code example

report = Report(metrics=[
    RecsysPreset(k=5),
])
column_mapping = ColumnMapping(recommendations_type='rank', target='rating', prediction='rank', item_id='title', user_id='user_id')
report.run(
    reference_data=reference,
    current_data=current,
    column_mapping=column_mapping,
    additional_data={'current_train_data': train_data}
  )
report

Check the How-to example for an end-to-end example on a sample dataset.

How it works

The Recommender System Performance report evaluates the quality of a recommender system. It can also compare the performance against the past, or the performance of an alternative model.

• Works for a single model or helps compare the two.

• Displays 10+ plots related to the predictive quality, ranking quality and recommendation diversity.

• Helps explore recommendation bias and examples of recommendations.

Check the All Metrics Table for a complete list of metrics and parameters. Check the Ranking Metrics Guide for an explanation of each metric.

Data Requirements

• To run this report, you need to have both target (true relevance) and prediction (rank or score) columns available. Additional training data and input features are required for some of the metrics.

• To generate a comparative report, you will need two datasets. The reference dataset serves as a benchmark, e.g. past model performance or a performance of a different model. Evidently analyzes the change by comparing the current production data to the reference data.

• You can also run this report for a single dataset, with no comparison performed.

• You must perform column mapping to map the data inputs correctly.

Check the Column Mapping guide and instructions on using Additional Data in recommendations.

Report customization

• You can pass user_bias_columns columns or item_bias_columns you want to evaluate for bias.

• You can pass display_features you want to display in the recommendation cases table.

• You can pass user_ids: List to specify which users you want to display in the recommendation cases table.

• You can use a different color schema for the report.

• If you want to exclude some metrics, you can create a custom report by combining the chosen metrics. See the complete list here

Check the All Metrics Table for a complete list of metrics and parameters.

Recommender System Test Suite

If you want to run recommender system performance checks as part of a pipeline, you can create a Test Suite and include the RecsysTestPreset.

Code example

tests = TestSuite(tests=[
    RecsysTestPreset(k=5)
])

column_mapping = ColumnMapping(recommendations_type='rank', target='rating', prediction='rank', item_id='title', user_id='user_id')
tests.run(
    reference_data=reference,
    current_data=current,
    column_mapping=column_mapping,
    additional_data={'current_train_data': train_data}
  )
tests

How it works

You can use the RecsysTestPreset to evaluate the quality of a recommender or ranking system, when you have the ground truth data (relevance scores).

• It compares ranking quality metrics against the defined expectation.

• For Evidently to generate the test conditions automatically, you should pass the reference dataset (e.g., performance during the previous period). You can also set the performance expectations manually by passing custom test conditions.

Head here to the All tests table to see the composition of each preset and default parameters.

Test Suite customization

• You can set custom test conditions.

• If you want to exclude some tests or add additional ones, you can create a custom test suite by combining the chosen tests. See the complete list here.