module

The four main modules for auto graph learning are listed here.

feature

class autogl.module.feature.selectors.BaseSelector(data_t='np', multigraph=False, **kwargs)[source]
class autogl.module.feature.selectors.SeFilterConstant(data_t='np', multigraph=False, **kwargs)[source]

drop constant features

class autogl.module.feature.selectors.SeGBDT(fixlen=10, *args, **kwargs)[source]

simple wrapper of lightgbm , using importance ranking to select top-k features.

Parameters

fixlen (int) – K for top-K important features.

model

train

hyper parameter optimization

ensemble

class autogl.module.ensemble.Stacking(meta_model='gbm', meta_params={}, *args, **kwargs)[source]

A stacking ensembler. Currently we support gradient boosting as the meta-algorithm.

Parameters
  • meta_model ('gbm' or 'glm' (Optional)) –

    Type of the stacker:

    ’gbm’ : Gradient boosting model. This is the default. ‘glm’ : Generalized linear model.

  • meta_params (a dict (Optional)) – When meta_model is specified, you can customize the parameters of the stacker. If this argument is not provided, the stacker will be configurated with default parameters. Default {}.

ensemble(predictions, identifiers, *args, **kwargs)[source]

Ensemble the predictions of base models.

Parameters
  • predictions (a list of np.ndarray) – Predictions of base learners (corresponding to the elements in identifiers).

  • identifiers (a list of str) – The names of base models.

Returns

The ensembled predictions.

Return type

np.ndarray

fit(predictions, label, identifiers, feval, n_classes='auto', *args, **kwargs)[source]

Fit the ensembler to the given data using Stacking method.

Parameters
  • predictions (a list of np.ndarray) – Predictions of base learners (corresponding to the elements in identifiers).

  • label (a list of int) – Class labels of instances.

  • identifiers (a list of str) – The names of base models.

  • feval ((a list of) autogl.module.train.evaluate) – Performance evaluation metrices.

  • n_classes (int or str (Optional)) – The number of classes. Default as 'auto', which will use maximum label.

Returns

The validation performance of the final stacker.

Return type

(a list of) float

class autogl.module.ensemble.Voting(ensemble_size=10, *args, **kwargs)[source]

An ensembler using the voting method.

Parameters

ensemble_size (int) – The number of base models selected by the voter. These selected models can be redundant. Default as 10.

ensemble(predictions, identifiers, *args, **kwargs)[source]

Ensemble the predictions of base models.

Parameters
  • predictions (a list of np.ndarray) – Predictions of base learners (corresponding to the elements in identifiers).

  • identifiers (a list of str) – The names of base models.

Returns

The ensembled predictions.

Return type

np.ndarray

fit(predictions, label, identifiers, feval, *args, **kwargs)[source]

Fit the ensembler to the given data using Rich Caruana’s ensemble selection method.

Parameters
  • predictions (a list of np.ndarray) – Predictions of base learners (corresponding to the elements in identifiers).

  • labels (a list of int) – Class labels of instances.

  • identifiers (a list of str) – The names of base models.

  • feval ((a list of) instances in autogl.module.train.evaluate) – Performance evaluation metrices.

Returns

The validation performance of the final voter.

Return type

(a list of) float

autogl.module.ensemble.build_ensembler_from_name(name: str) → autogl.module.ensemble.base.BaseEnsembler[source]
Parameters

name (str) – the name of ensemble module.

Returns

the ensembler built using default parameters

Return type

BaseEnsembler

Raises

AssertionError – If an invalid name is passed in