h2o4gpu

h2o4gpu copied to clipboard

Currently, we only support dense matrices.

mdymczyk

Intel DAAL implementation for tSVD

1

Research if we can gain anything from it and if so implement our tSVD algo using Intel DAAL. Things to look into: 1) how long it would take 2) can...

mdymczyk

KMeans DAAL implementation.

2

Implement KMeans on DAAL, preferably in C++ so it's accessible in Python and R.

mdymczyk

Let's figure out if we can use [scikit-learn's CV functions](http://scikit-learn.org/stable/modules/cross_validation.html) to perform cross-validation of the estimators, while still training on GPUs. We need a generic way to do CV on...

ledell

Add the Stacked Ensemble algorithm

2

We should implement stacking, as there are not many (if any) stacking implementations for GPUs. This will require writing wrappers in Python and then adding some code to the R...

ledell

*) compute capability from xgboost code *) deviceQuery *) malloc (for testing) https://stackoverflow.com/questions/5730010/pycuda-querying-device-status-memory-specifically

pseudotensor

When running KMeans and GPU GLM benchmarks the performance on a volta machine is twice as slow when compared to a pascal machine. The benchmark scripts I am running are...

CreatureDev

* Implementation should be in cpp/cu files mostly * DRY the code so we don't duplicate it

mdymczyk

In both Python and C/C++ code we could use a simple annotation/preprocessor directive that would time the running time of each method. For Python we could do something similar to...

mdymczyk

Multi GPU XGBoost performance improvement

1

mdymczyk