Skip to content

Track

A track evaluate a model's performance.

The following metrics are recorded:

- FLOPS: floating point operations per second. - Time, which should be interpreted with wisdom. Indeed time can depend on the architecture and local resource situations. Comparison via FLOPS should be preferred. - The model's memory footprint. - The model's predictive performance on the track's dataset.

Parameters

  • name (str)

    The name of the track.

  • dataset

    The dataset from which samples will be retrieved. A slice must be used if the dataset is a data generator.

  • metric

    The metric(s) used to track performance.

  • n_samples (int) – defaults to None

    The number of samples that are going to be processed by the track.

Methods

run