Note

This page is a reference documentation. It only explains the class signature, and not how to use it. Please refer to the user guide for the big picture.

nilearn.decoding.FREMClassifier

class nilearn.decoding.FREMClassifier(estimator='svc', mask=None, cv=30, param_grid=None, clustering_percentile=10, screening_percentile=20, scoring='roc_auc', smoothing_fwhm=None, standardize=True, target_affine=None, target_shape=None, mask_strategy='background', low_pass=None, high_pass=None, t_r=None, memory=None, memory_level=0, n_jobs=1, verbose=0)

State of the art decoding scheme applied to usual classifiers.

FREM uses an implicit spatial regularization through fast clustering and aggregates a high number of estimators trained on various splits of the training set, thus returning a very robust decoder at a lower computational cost than other spatially regularized methods [1].

Parameters:

estimatorstr, default ‘svc’)

The estimator to choose among:

• "svc": Linear support vector classifier with L2 penalty.

svc = LinearSVC(penalty="l2", max_iter=1e4)

• "svc_l2": Linear support vector classifier with L2 penalty.

Note

Same as option svc.

• "svc_l1": Linear support vector classifier with L1 penalty.

svc_l1 = LinearSVC(penalty="l1", dual=False, max_iter=1e4)

• "logistic": Logistic regression with L2 penalty.

logistic = LogisticRegressionCV(penalty="l2", solver="liblinear")

• "logistic_l1": Logistic regression with L1 penalty.

logistic_l1 = LogisticRegressionCV(penalty="l1", solver="liblinear")

• "logistic_l2": Logistic regression with L2 penalty

Note

Same as option logistic.

• "ridge_classifier": Ridge classifier.

ridge_classifier = RidgeClassifierCV()

• "dummy_classifier": Dummy classifier with stratified strategy.

dummy = DummyClassifier(strategy="stratified", random_state=0)

maskfilename, Nifti1Image, NiftiMasker, or MultiNiftiMasker, optional, default=None

Mask to be used on data. If an instance of masker is passed, then its mask and parameters will be used. If no mask is given, mask will be computed automatically from provided images by an inbuilt masker with default parameters. Refer to NiftiMasker or MultiNiftiMasker to check for default parameters.

cvcross-validation generator, int or None, default=30

A cross-validation generator. See: https://scikit-learn.org/stable/modules/cross_validation.html. If None is passed, cv=30 will be used. It can be an integer, in which case it is the number of folds in a KFold using StratifiedKFold when groups is None in the fit method for this class. If groups is specified but cv is not set to custom CV splitter, default is LeaveOneGroupOut.

param_griddict of str to sequence, or sequence of such. default=None

The parameter grid to explore, as a dictionary mapping estimator parameters to sequences of allowed values.

None or an empty dict signifies default parameters.

A sequence of dicts signifies a sequence of grids to search, and is useful to avoid exploring parameter combinations that make no sense or have no effect. See scikit-learn documentation for more information, for example: https://scikit-learn.org/stable/modules/grid_search.html

clustering_percentileint, float, in closed interval [0, 100], default=10

Used to perform a fast ReNA clustering on input data as a first step of fit. It agglomerates similar features together to reduce their number down to this percentile. ReNA is typically efficient for cluster_percentile equal to 10.

screening_percentileint, float, in closed interval [0, 100], default=20

The percentage of brain volume that will be kept with respect to a full MNI template. In particular, if it is lower than 100, a univariate feature selection based on the Anova F-value for the input data will be performed. A float according to a percentile of the highest scores.

scoringstr, callable or None, optional. default=’roc_auc’

The scoring strategy to use. See the scikit-learn documentation at https://scikit-learn.org/stable/modules/model_evaluation.html#the-scoring-parameter-defining-model-evaluation-rules If callable, takes as arguments the fitted estimator, the test data (X_test) and the test target (y_test) if y is not None. e.g. scorer(estimator, X_test, y_test)

For classification, valid entries are: ‘accuracy’, ‘f1’, ‘precision’, ‘recall’ or ‘roc_auc’; default=’roc_auc’

smoothing_fwhmfloat or int or None, optional.

If smoothing_fwhm is not None, it gives the full-width at half maximum in millimeters of the spatial smoothing to apply to the signal.

standardizeany of: ‘zscore_sample’, ‘zscore’, ‘psc’, True, False or None; default=True

Strategy to standardize the signal:

• 'zscore_sample': The signal is z-scored. Timeseries are shifted to zero mean and scaled to unit variance. Uses sample std.

• 'zscore': The signal is z-scored. Timeseries are shifted to zero mean and scaled to unit variance. Uses population std by calling default numpy.std with N - ddof=0.

  Deprecated since Nilearn 0.10.1: This option will be removed in Nilearn version 0.14.0. Use zscore_sample instead.

• 'psc': Timeseries are shifted to zero mean value and scaled to percent signal change (as compared to original mean signal).

• True: The signal is z-scored (same as option zscore). Timeseries are shifted to zero mean and scaled to unit variance.

  Deprecated since Nilearn 0.13.0dev: In nilearn version 0.15.0, True will be replaced by 'zscore_sample'.

• False: Do not standardize the data.

  Deprecated since Nilearn 0.13.0dev: In nilearn version 0.15.0, False will be replaced by None.

Deprecated since Nilearn 0.10.1: The default will be changed to 'zscore_sample' and 'zscore' will be removed in in version 0.14.0.

Deprecated since Nilearn 0.15.0dev: The default will be changed to 'zscore_sample' in version 0.15.0.

target_affine3x3 or a 4x4 array-like, or None, default=None

If specified, the image is resampled corresponding to this new affine.

target_shapetuple or list or None, default=None

If specified, the image will be resized to match this new shape. len(target_shape) must be equal to 3.

Note

If target_shape is specified, a target_affine of shape (4, 4) must also be given.

mask_strategy{“background”, “epi”, “whole-brain-template”,”gm-template”, “wm-template”}, optional

The strategy used to compute the mask:

• "background": Use this option if your images present a clear homogeneous background. Uses nilearn.masking.compute_background_mask under the hood.

• "epi": Use this option if your images are raw EPI images. Uses nilearn.masking.compute_epi_mask.

• "whole-brain-template": This will extract the whole-brain part of your data by resampling the MNI152 brain mask for your data’s field of view. Uses nilearn.masking.compute_brain_mask with mask_type="whole-brain".

  Note

  This option is equivalent to the previous ‘template’ option which is now deprecated.

• "gm-template": This will extract the gray matter part of your data by resampling the corresponding MNI152 template for your data’s field of view. Uses nilearn.masking.compute_brain_mask with mask_type="gm".

  Added in Nilearn 0.8.1.

• "wm-template": This will extract the white matter part of your data by resampling the corresponding MNI152 template for your data’s field of view. Uses nilearn.masking.compute_brain_mask with mask_type="wm".

  Added in Nilearn 0.8.1.

Note

This parameter will be ignored if a mask image is provided.

Note

Depending on this value, the mask will be computed from nilearn.masking.compute_background_mask, nilearn.masking.compute_epi_mask, or nilearn.masking.compute_brain_mask.

Default=’background’.

low_passfloat or int or None, default=None

Low cutoff frequency in Hertz. If specified, signals above this frequency will be filtered out. If None, no low-pass filtering will be performed.

high_passfloat or int or None, default=None

High cutoff frequency in Hertz. If specified, signals below this frequency will be filtered out.

t_rfloat or int or None, default=None

Repetition time, in seconds (sampling period). Set to None if not provided.

memoryNone, instance of joblib.Memory, str, or pathlib.Path

Used to cache the masking process. By default, no caching is done. If a str is given, it is the path to the caching directory.

memory_levelint, default=0

Rough estimator of the amount of memory used by caching. Higher value means more memory for caching. Zero means no caching.

n_jobsint, default=1

The number of CPUs to use to do the computation. -1 means ‘all CPUs’.

verboseint, default=0

Verbosity level (0 means no message).

Attributes:

coef_numpy.ndarray, shape=(n_classes, n_features)

Contains the mean of the models weight vector across fold for each class. Returns None for Dummy estimators.

coef_img_dict of Nifti1Image

Dictionary containing coef_ with class names as keys, and coef_ transformed in Nifti1Images as values. In the case of a regression, it contains a single Nifti1Image at the key ‘beta’. Ignored if Dummy estimators are provided.

cv_list of pairs of lists

List of the (n_folds,) folds. For the corresponding fold, each pair is composed of two lists of indices, one for the train samples and one for the test samples.

cv_params_dict of list

Best point in the parameter grid for each tested fold in the inner cross validation loop. The grid is empty when Dummy estimators are provided.

Note

If the estimator used its built-in cross-validation, this will include an additional key for the single best value estimated by the built-in cross-validation (‘best_C’ for LogisticRegressionCV and ‘best_alpha’ for RidgeCV/RidgeClassifierCV/LassoCV), in addition to the input list of values.

cv_scores_dict, (classes, n_folds)

Scores (misclassification) for each parameter, and on each fold

dummy_output_ndarray, shape=(n_classes, 2) or shape=(1, 1) for regression

Contains dummy estimator attributes after class predictions using strategies of sklearn.dummy.DummyClassifier (class_prior) and sklearn.dummy.DummyRegressor (constant) from scikit-learn. This attribute is necessary for estimating class predictions after fit. Returns None if non-dummy estimators are provided.

estimator_Estimator object used during decoding.

intercept_ndarray, shape (nclasses,)

Intercept (also known as bias) added to the decision function. Ignored if Dummy estimators are provided.

mask_img_Nifti1Image or SurfaceImage

Mask computed by the masker object.

masker_instance of NiftiMasker, MultiNiftiMasker, or SurfaceMasker

The masker used to mask the data.

memory_joblib memory cache

n_elements_int

The number of voxels or vertices in the mask.

Added in Nilearn 0.12.1.

n_outputs_int

Number of outputs (column-wise)

scorer_function

Scorer function used on the held out data to choose the best parameters for the model.

screening_percentile_float

Percentile value for ANOVA univariate feature selection. A value of 100 means ‘keep all features’. This percentile is expressed with respect to the volume of either a standard (MNI152) brain (if mask_img is a 3D volume) or a the number of vertices in the mask mesh (if mask_img is a SurfaceImage).

std_coef_numpy.ndarray, shape=(n_classes, n_features)

Contains the standard deviation of the models weight vector across fold for each class. Note that folds are not independent, see https://scikit-learn.org/stable/modules/cross_validation.html#cross-validation-iterators-for-grouped-data Ignored if Dummy estimators are provided.

std_coef_img_dict of Nifti1Image

Dictionary containing std_coef_ with class names as keys, and coef_ transformed in Nifti1Image as values. In the case of a regression, it contains a single Nifti1Image at the key ‘beta’. Ignored if Dummy estimators are provided.

classes_ndarray of labels (n_classes_)

Labels of the classes

n_classes_int

number of classes

See also

nilearn.decoding.Decoder

Classification strategies for Neuroimaging,

nilearn.decoding.FREMRegressor

State of the art regression pipeline for Neuroimaging

References

[1]

Andrés Hoyos-Idrobo, Gaël Varoquaux, Yannick Schwartz, and Bertrand Thirion. Frem – scalable and stable decoding with fast regularized ensemble of models. NeuroImage, 180:160–172, 2018. New advances in encoding and decoding of brain signals. doi:10.1016/j.neuroimage.2017.10.005.

__init__(estimator='svc', mask=None, cv=30, param_grid=None, clustering_percentile=10, screening_percentile=20, scoring='roc_auc', smoothing_fwhm=None, standardize=True, target_affine=None, target_shape=None, mask_strategy='background', low_pass=None, high_pass=None, t_r=None, memory=None, memory_level=0, n_jobs=1, verbose=0)

decision_function(X)

Predict class labels for samples in X.

Parameters:

XNiimg-like, list of either Niimg-like objects or str or path-like

See Input and output: neuroimaging data representation. Data on prediction is to be made. If this is a list, the affine is considered the same for all.

Returns:

y_prednumpy.ndarray, shape (n_samples,)

Predicted class label per sample.

fit(X, y, groups=None)

Fit the decoder (learner).

Parameters:

Xlist of Niimg-like or SurfaceImage objects

See Input and output: neuroimaging data representation. Data on which model is to be fitted. If this is a list, the affine is considered the same for all.

ynumpy.ndarray of shape=(n_samples) or list of length n_samples

The dependent variable (age, sex, IQ, yes/no, etc.). Target variable to predict. Must have exactly as many samples as the input images.

groupsNone, default=None

Group labels for the samples used while splitting the dataset into train/test set.

Note that this parameter must be specified in some scikit-learn cross-validation generators to calculate the number of splits, for example sklearn.model_selection.LeaveOneGroupOut or sklearn.model_selection.LeavePGroupsOut.

For more details see https://scikit-learn.org/stable/modules/cross_validation.html#cross-validation-iterators-for-grouped-data

get_metadata_routing()

Get metadata routing of this object.

Please check User Guide on how the routing mechanism works.

Returns:

routingMetadataRequest

A MetadataRequest encapsulating routing information.

get_params(deep=True)

Get parameters for this estimator.

Parameters:

deepbool, default=True

If True, will return the parameters for this estimator and contained subobjects that are estimators.

Returns:

paramsdict

Parameter names mapped to their values.

predict(X)

Predict a label for all X vectors indexed by the first axis.

Parameters:

XNiimg-like, SurfaceImage, list of Niimg-like objects or list of SurfaceImage, or {array-like, sparse matrix}, shape = (n_samples, n_features)

See Input and output: neuroimaging data representation. Data on which prediction is to be made.

Returns:

array, shape=(n_samples,) if n_classes == 2 else (n_samples, n_classes)

Confidence scores per (sample, class) combination. In the binary case, confidence score for self.classes_[1] where >0 means this class would be predicted.

score(X, y, *args)

Compute the prediction score using the scoring metric defined by the scoring attribute.

Parameters:

XNiimg-like, SurfaceImage, list of Niimg-like objects or list of SurfaceImage, or {array-like, sparse matrix}, shape = (n_samples, n_features)

See Input and output: neuroimaging data representation. Data on which prediction is to be made.

ynumpy.ndarray

Target values.

argsOptional arguments that can be passed to

scoring metrics. Example: sample_weight.

Returns:

scorefloat

Prediction score.

set_fit_request(*, groups='$UNCHANGED$')

Configure whether metadata should be requested to be passed to the fit method.

Note that this method is only relevant when this estimator is used as a sub-estimator within a meta-estimator and metadata routing is enabled with enable_metadata_routing=True (see sklearn.set_config). Please check the User Guide on how the routing mechanism works.

The options for each parameter are:

• True: metadata is requested, and passed to fit if provided. The request is ignored if metadata is not provided.

• False: metadata is not requested and the meta-estimator will not pass it to fit.

• None: metadata is not requested, and the meta-estimator will raise an error if the user provides it.

• str: metadata should be passed to the meta-estimator with this given alias instead of the original name.

The default (sklearn.utils.metadata_routing.UNCHANGED) retains the existing request. This allows you to change the request for some parameters and not others.

Added in version 1.3.

Parameters:

groupsstr, True, False, or None, default=sklearn.utils.metadata_routing.UNCHANGED

Metadata routing for groups parameter in fit.

Returns:

selfobject

The updated object.

set_params(**params)

Set the parameters of this estimator.

The method works on simple estimators as well as on nested objects (such as Pipeline). The latter have parameters of the form <component>__<parameter> so that it’s possible to update each component of a nested object.

Parameters:

**paramsdict

Estimator parameters.

Returns:

selfestimator instance

Estimator instance.

Examples using nilearn.decoding.FREMClassifier

Decoding with FREM: face vs house vs chair object recognition

Decoding with FREM: face vs house vs chair object recognition