Fitting models · Feyn Documentation

by: Chris Cave & Jaan Kasak
(Feyn version 3.0 or newer)

Each Model is a mathematical function that can be optimised to a loss function in order to fit to a dataset.

Example

Building on from the previous section, here's an example of how to sample and fit models.

import feyn
from feyn.datasets import make_classification

train, test = make_classification()

ql = feyn.QLattice()

models = ql.sample_models(train.columns, 'y', 'classification', max_complexity=10)

models = feyn.fit_models(
    models=models,
    data=train,
    loss_function='binary_cross_entropy',
    criterion='bic',
    threads=4
)

Parameters of `fit_models`

models

This is the list of Models you want to fit the data to. This will return the same list but sorted by ascending loss.

data

This is the data that the Models will be fitted to. This needs to be a pandas.DataFrame object where a sample is represented by a single row and the variable names are the names of the columns.

loss_function

This is the loss function that the Models are optimised for. If no criterion is given, the returned list of Models is sorted by this loss function in ascending order. It takes any loss function in the feyn.losses module. Popular choices of loss functions are:

binary_cross_entropy
squared_error

criterion

The criterion takes into account the complexity of the Model when comparing losses. Higher complexity is penalized. Choices of criterion are:

bic (Bayesion Information Criterion)
aic (Akaike Information Criterion)

bic penalises complex models more than aic.

If a criterion is given, the returned list of Models are sorted by that instead of the loss.

threads

Determines how many concurrent threads to use during fitting. For best effect, set the threads parameter to match the number of threads available on your computer.