Regressor

In this module regressor_obj class is presented, which is related to machine learning regression. This class doesn’t have an implementation of any supervised learning regression algorithm, but is an interface between a regressor object implemented elsewhere and ESCAPE model and data objects. In other words this class allows one to use any 3rd party or self-implemented regressor class.

One of those well-known 3rd party regressors is a Multilayer Perceptor Regressor from scikit-learn package [1].

Machine learning regressors after training can predict a solution for a model for a given experimental dataset. This solution can be optimized further using any optimizer from ESCAPE.

Regressors can be quite useful for predictions of solutions for large number of experimental data, so-called batch processing, giving quite close to actual initial solutions for further optimization.

[1]

https://scikit-learn.org/stable/modules/neural_networks_supervised.html

escape.core.regressor.regressor(name, stack, nsamples=1000, ntests=100, impl=None, norm_method=None, noisify_method=None, fit_method=None, score_method=None, predict_method=None, stop_method=None)

Given an instance of a regressor class implemented in python impl returns regressor_obj object. Regressor represents a supervised leaerning regression algorithm which training can predict model parameters for a given experimental data.

Parameters:

stack: modelstack_obj, list of models or model_obj

Models to be optimized.

nsamples: positive integer

Number of training datasets to be generated.

ntests: non-negative integer value

Number of tests to use for calculation of optimization error.

impl: object

Instance of regressor implementation.

norm_method: callable

Normalization method applied to generated training data. If None nor normalization occured.

noisify_method: callable

Method to noisify generated training data. If None nor normalization occured.

fit_method: callable

Custom fit method of regressor. If None uses ‘fit’ method of implementation.

score_method: callable

Custom score method of regressor. If None uses ‘score’ method of implementation.

predict_method: callable

Custom predict method of regressor. If None uses ‘predict’ method of implementation.

stop_method: callable

Custom stop method of regressor. If None uses ‘stop’ method of implementation.

Returns:

instance of regressor_obj

class escape.core.regressor.regressor_obj

Wrapper class for regressor

fit_method

Returns:

Fit method of regressor instance

load_parameters(self, fn)

Load parameters from ASCII file with name ‘fn’.

modelstack

Returns:

Modestack used to generate training data

name

Returns:

Name of regressor object

next_test(self)

Predicts parameter values for the next test set and calls model for the found solution.

noisify_method

Returns:

Noisify method for generated training data

norm_method

Returns:

Normalization method, applied to generated training data.

nsamples

Returns:

Number of training samples

ntests

Returns:

Number of tests

num_of_params

Returns:

Number of parameters.

parameter(self, size_t i)

predict_method

Returns:

Predict method

prev_test(self)

Predicts parameter values for the previous test set and calls model for the found solution.

regressor_impl

Returns:

Regressor instance

reset(self)

Resets parameters values.

save_parameters(self, fn)

Saves parameters to ASCII file with name ‘fn’.

score_method

Returns:

Score method of regressor instance

shake(self)

Sets parameter values randomly.

stop(self)

Stops training process.

stop_method

Returns:

Stop method for stopping training process

train(self, bool reset_data)

Trains regressor.

Parameters:

reset_data: bool

If True resets previously generated training set