`discretefirstorder`.DFORegressor¶

class discretefirstorder.DFORegressor(loss='mse', learning_rate='auto', k=1, polish=True, n_runs=50, max_iter=100, tol=0.001, fit_intercept=True, normalize=True, random_state=None)[source]¶

Discrete first-order regressor.

Parameters:

lossstr: type of loss to be minimized. One of ‘mse’ or ‘mae’.
learning_ratestr or float: learning rate to be used.
kint: number of non-zero features to keep.
polishbool: whether to polish coefficients by running least squares on the active set.
n_runsint: number of runs of the discrete first order optimization procedure.
max_iterint: maximum number of steps to take during one run of the discrete first order optimization algorithm.
tolfloat: tolerance below which the optimization algorithm stops.
fit_interceptbool: whether to fit an intercept term
normalizebool: whether to normalize the input data.

Examples

>>> from discretefirstorder import DFORegressor
>>> import numpy as np
>>> X = np.arange(100).reshape(100, 1)
>>> y = np.random.normal(size=(100, ))
>>> estimator = DFORegressor()
>>> estimator.fit(X, y)
DFORegressor()

Attributes:

coef_ndarray, shape (n_features,): coefficient vector.
intercept_float: intercept.

__init__(loss='mse', learning_rate='auto', k=1, polish=True, n_runs=50, max_iter=100, tol=0.001, fit_intercept=True, normalize=True, random_state=None)[source]¶

fit(X, y, coef_init=None)[source]¶

Implementation of the fit method for the discrete first-order regressor.

Parameters:

Xarray-like of shape (n_samples, n_features): the training input samples.
yarray-like of shape (n_samples,): the target values.
coef_init(optional) array-like of shape (n_features,): initial value of regression coefficients

Returns: