Vladislavs Dovgalecs

Original CRFSuite has optimization flags to use SSE/SSE2 intrinsics to speed up numerical computation. Is this enabled in "python-crfsuite"? Does it make sense to build this package locally and install...

Get loss value(s) during training

1

I would like to use a hyperparameter search algorithm to find the best hyperparameters (L1 and L2). For this I think the last loss value reported by the training algorithm...

Using features as extra evidence for preference selection

2

Thanks for the great package! Is it possible in choix for each observation (or pair if you wish) to attribute and use some feature vector? The feature vectors, not outcomes...

Current implementation should be called "Linear ELM" since only linear relationships are used in the learning process. An option or even a code branch could implement kernelized version of ELM...

The code allows to include the training/prediction functions easily in any C++ project. Currently the only interface to this library is implemented for MATLAB. It would be nice to develop...

This is embarrassing.

Current code implements sigmoid activation function. It could be nice to let the user select which function to select and/or allow easily use his own function.

Occasionally on relatively small datasets the program crashes. This may be due to the matrix which _may_ be get singular, hence no solution is found. Some checking and/or validation is...

Current implementation works nice and as expected if class labels are natural numbers, and start from 1 up to maximal value. Implementation is not very nice if labels are say:...

Does the package return (or write in the log file) the perplexity or any other goodness of fit measure? If yes, would it be a good idea to run a...