Step¶

class baikal.Step(*args, name=None, n_outputs=1, **kwargs)¶

Bases: baikal._core.step._StepBase

Mixin class to endow scikit-learn classes with Step capabilities.

Steps are defined by combining any class we would like to make a step from with this mixin class. This mixin, among other things, endows the class of interest with a __call__ method, making the class callable on the outputs (DataPlaceholder objects) of previous steps and optional targets (also DataPlaceholder objects). You can make a step from any class you like, so long that class implements the scikit-learn API.

Instructions:

Define a class that inherits from both this mixin and the class you wish to make a step of (in that order!).
In the class __init__, call super().__init__(...) and pass the appropriate step parameters.

The base class may implement a predict/transform method (the compute function) that take multiple inputs and returns multiple outputs, and a fit method that takes multiple inputs and targets. In this case, the input/target arguments are expected to be a list of (typically) array-like objects, and the compute function is expected to return a list of array-like objects.

Parameters

name – Name of the step (optional). If no name is passed, a name will be automatically generated.
n_outputs – The number of outputs of the step’s function (predict, transform, or any other callable passed in the compute_func argument).

Examples

import sklearn.linear_model
# The order of inheritance is important!
class LogisticRegression(Step, sklearn.linear_model.LogisticRegression):
    def __init__(self, *args, name=None, **kwargs):
        super().__init__(*args, name=name, **kwargs)

logreg = LogisticRegression(C=2.0)

Methods

`get_compute_func_at`(port)	Get compute function at the specified port.
`get_fit_compute_func_at`(port)	Get fit-compute function at the specified port.
`get_inputs_at`(port)	Get inputs at the specified port.
`get_outputs_at`(port)	Get outputs at the specified port.
`get_targets_at`(port)	Get targets at the specified port.
`get_trainable_at`(port)	Get trainable flag at the specified port.
`set_compute_func_at`(port, value)	Set compute function at the specified port.
`set_fit_compute_func_at`(port, value)	Set fit-compute function at the specified port.
`set_trainable_at`(port, value)	Set trainable flag at the specified port.
`__call__`(inputs[, targets, compute_func, …])	Call the step on input(s) (from previous steps) and generates the output(s) to be used in further steps.

Attributes

`compute_func`	Get the compute function of the step.
`fit_compute_func`	Get the fit-compute function of the step.
`inputs`	Get the inputs of the step.
`n_outputs`	Get the number of outputs the step produces.
`name`	Get the name of the step.
`outputs`	Get the outputs of the step.
`targets`	Get the targets of the step.
`trainable`	Get trainable flag of the step.