Sparse Multinomial Logistic Regression (SMLR) classifier, developed by Kei Majima at Kyoto Univ.

Sparse regularization by automatic relevance determination (ARD) prior was introduced to the linear multinomial logistic regression algorithm (Yamashita et al., 2008). This regularization process estimates the importance of each voxel (feature) and prunes away voxels that are not useful for prediction.

Original SLR toolbox for Matlab is available at http:https://www.cns.atr.jp/%7Eoyamashi/SLR_WEB.html.

Run the following command:

$ pip install git+https://github.com/KamitaniLab/smlr.git

import smlr

model = smlr.SMLR(max_iter=1000, tol=1e-5, verbose=1)
model.fit(x, y)
model.predict(x_test)

• x, x_test: numpy array of input features (# of samples x # of features)
• y: label vector consisting of integers (len(y) = # of samples; please use integers 0, 1, 2, ..., K-1 when K-class classification)

The API of this function is compatible with the logistic regression in scikit-learn.

• max_iter: The number of iterations in training (default 1000).
• tol: The tolerance value of stopping criteria (default 1e-5; positive value)
• verbose: If 1, print verbose information (default).

• coef_: array, shape = [n_classes, n_features]
  • Coefficient of the features in the decision function.
• intercept_: array, shape = [n_classes]
  • Intercept (a.k.a. bias) added to the decision function.

For demonstration, try demoSMLR_20170617.py.

Yamashita O, Sato MA, Yoshioka T, Tong F, Kamitani Y. (2008) Sparse estimation automatically selects voxels relevant for the decoding of fMRI activity patterns. NeuroImage. doi: 10.1016/j.neuroimage.2008.05.050. http:https://www.sciencedirect.com/science/article/pii/S1053811908006940

The scripts provided here are released under the MIT license (http:https://opensource.org/licenses/mit-license.php).