MLAlgo

class lightautoml.ml_algo.base.MLAlgo(default_params=None, freeze_defaults=True, timer=None, optimization_search_space={})[source]

Bases: ABC

Abstract class for machine learning algorithm.

Assume that features are already selected, but parameters my be tuned and set before training.

Parameters:

  • default_params (Optional[dict]) – Algo hyperparams.

  • freeze_defaults (bool) –

    • True : params may be rewrited depending on dataset.

    • False: params may be changed only manually

      or with tuning.

  • timer (Optional[TaskTimer]) – Timer for Algo.

property name

Get model name.

property features

Get list of features.

property is_fitted

Get flag is the model fitted or not.

property params

Get model’s params dict.

init_params_on_input(train_valid_iterator)[source]

Init params depending on input data.

Parameters:

train_valid_iterator (TrainValidIterator) – Classic cv-iterator.

Return type:

dict

Returns:

Dict with model hyperparameters.

abstract fit_predict(train_valid_iterator)[source]

Abstract method.

Fit new algo on iterated datasets and predict on valid parts.

Parameters:

train_valid_iterator (TrainValidIterator) – Classic cv-iterator.

Return type:

LAMLDataset

abstract predict(test)[source]

Predict target for input data.

Parameters:

test (LAMLDataset) – Dataset on test.

Return type:

LAMLDataset

Returns:

Dataset with predicted values.

score(dataset)[source]

Score prediction on dataset with defined metric.

Parameters:

dataset (LAMLDataset) – Dataset with ground truth and predictions.

Return type:

float

Returns:

Metric value.

set_prefix(prefix)[source]

Set prefix to separate models from different levels/pipelines.

Parameters:

prefix (str) – String with prefix.

set_timer(timer)[source]

Set timer.

Return type:

MLAlgo