rfgb.boosting module

Core methods for performing learning and inference, such as computing gradients, updating gradients, and performing inference.

Documentation

rfgb.boosting.computeAdviceGradient(example)

Proves each clause (Prover.prove()) and computes the advice gradient as NumberTrue - NumberFalse.

Parameters:example –

rfgb.boosting.computeSumOfGradients(example, trees, data)

Computes new gradients for an example.

Parameters:

• example –
• trees –
• data –

rfgb.boosting.inferTreeValue(clauses, query, data)

Returns the probability of query given data and the clauses learned.

Parameters:

• clauses –
• query –
• data –

rfgb.boosting.performInference(testData, trees)

Computes the probabilities for test examples.

Parameters:

• testData (utils.Data object.) – Data for testing.
• trees (list.) – List of strings representing learned decision trees.

Example:

from rfgb.boosting import performInference

rfgb.boosting.updateGradients(data, trees, loss='LS', delta=None)

Update gradients of the data.

Parameters:

• data (utils.Data object.) – Training or testing data (with parameters).
• trees (list.) – List of strings representing trees.
• loss (str.) – Loss function for regression (currently implemented: ‘LS’, ‘LAD’, ‘Huber’).
• delta (float) – Delta value for Huber loss.

Example:

from rfgb.boosting import updateGradients