NeuroMiner 1.2 (09/2023)

What is new

Check out the new features!

• Synthetic data modue

• Different label module

• Modality manager

• New preprocessing options

  • Dimensionality reduction: fastICA, orthogonal NNMF

  • Extract variance components with fastICA

  • sequential kNN for imputation (faster than kNN!)

• Model significance with permutations in external validation mode

• Export your models

• Constrain Cross-validation structure

• Subgroups for external validation

• Improvements to multiclass module

• resolved bugs and improved usability!

Release notes

• One-class support vector machine (classification): To correctly use the one-class SVM (within the LIBSVM package), the reference group needs to be labeled as -1 or 2. One class classification is currently still in development, thus, errors might arise. Please send an email to the developers should you encounter some bugs.

• Default parameters of matLearn algorithms: Some of the algoirthms from the matLearn package currently come with no default parameters set. Make sure to choose these before initializing and running a model to avoid issues and errors. You can set the parameters when navigating to the Learning algorithm parameters (more information) option from the Parameter Template menu. To get an overview of the parameters and what values they can take, have a look at the matLearn documentation.

  • Algorithms that are missing/ have currently wrong default parameters**

    • Local regression

    • IMRelief needs different default parameters

• Python functions Some of the learnign algorithms and preprocessing steps call Python functions. In order to use these, please follow the steps on how to configure Python in Matlab before.

• Classification with MCRVM and polynomial kernel The polynomial kernel currently does not work for the MCRVM algorithm. This is a known issue and we are working on resolving it.

• Alternative label for multiclass problems is not yet implemented.