- model : instance of sklearn.linear_model.LinearModel
The linear model to alter.
- cov : instance of CovEstimator
The method used to estimate the covariance. Can either be one of the
predefined CovEstimator objects, or a function. If a function is used,
it must have the signature: def cov_modifier(cov, X, y), take the
training data as input and return a matrix that will be added to the
emperical covariance matrix. Defaults to None, which means the
default empirical estimator of the covariance matrix is used.
- cov_param_x0 : tuple | None
The initial parameters for the covariance estimator. These parameters
will be optimized. Defaults to None, which means the settings
defined in the CovEstimator object will be used.
- cov_param_bounds : list of tuple | None
For each parameter for the covariance estimator, the (min, max) bounds.
You can set a boundary to None to indicate the parameter is
unbounded in that direction. By default, the settings defined in the
CovEstimator object will be used.
- pattern_modifier : function | None
Function that takes a pattern (an ndarray of shape (n_features,
n_targets)) and modifies it. Must have the signature:
def pattern_modifier(pattern, X, y)
and return the modified pattern. Defaults to None, which means no
modification of the pattern.
- pattern_param_x0 : tuple | None
The initial parameters for the pattern modifier function. These
parameters will be optimized. Defaults to None, which means no
parameters of the pattern modifier function will be optimized.
- pattern_param_bounds : list of tuple | None
For each parameter for the pattern modifier function, the (min, max)
bounds. You can set a boundary to None to indicate the parameter
is unbounded in that direction. By default, all parameters are
considered unbounded.
- normalizer_modifier : function | None
Function that takes a normalizer (an ndarray of shape (n_targets,
n_targets)) and modifies it. Must have the signature:
def normalizer_modifier(normalizer, X, y, pattern, coef)
and return the modified normalizer. Defaults to None, which means
no modification of the normalizer.
- normalizer_param_x0 : tuple | None
The initial parameters for the normalizer modifier function. These
parameters will be optimized. Defaults to None, which means no
parameters of the normalizer modifier function will be optimized.
- normalizer_param_bounds : list of tuple | None
For each parameter for the normalizer modifier function, the (min, max)
bounds. You can set a boundary to None to indicate the parameter
is unbounded in that direction. By default, all parameters are
considered unbounded.
- random_search : int
Number of random sets of parameters to try, before picking the best set
to use as starting point for the L-BFGS-S algorithm. Set to 0 to
disable the random search and use the specified x0 instead.
Defaults to 0.
- optimizer_options : dict | None
A dictionary with extra options to supply to the L-BFGS-S algorithm. See
https://docs.scipy.org/doc/scipy/reference/optimize.minimize-lbfgsb.html.http://scikit-learn.org/stable/modules/model_evaluation.html
for a list of parameters.
- loo_patterns_method : ‘auto’ | ‘traditional’ | ‘kernel’
Method for computing the pattern matrix for the leave-one-out inner
crossvalidation loop. The ‘traditional’ setting will compute it by
performing a linear regression between the output of the initial model
and the weight matrix, as described in [1]. The ‘kernel’ setting
will use the matrix inversion lemma to compute the patterns, which can
be much faster when the number of features is larger than the number of
training samples, at the possible loss of some precision. The default
‘auto’ setting will switch between the ‘traditional’ and ‘kernel’
approaches based on the ratio between the number of features and
training samples.
- scoring : str | function
The scoring function to use while optimizating the parameters. Can be
any scikit-learn scorer, see:
http://scikit-learn.org/stable/modules/model_evaluation.html.
Defaults to 'neg_mean_squared_error' which causes the optimization
algorithm to attempt to minimize the mean-squared-error (MSE) between
the model output and training labels.
- random_state :
numpy.random.RandomState | int | None
Random state to use. Can either be a NumPy
numpy.random.RandomState object, an integer seed or None
to use a different seed every time. Defaults to None.
- verbose : bool
Whether to be verbose or not. Defaults to True.