Moved metric-tabular code and notebooks into main branch of metric-tabular repo. Consolidated notebooks for datasets in notebooks/ leaving only dataset files in top level dir.
Added logic for converting to llvm using scipy. This can surely be done faster outside of python, but don't overoptimize. Strangely, the libsvm, svmlite file for BRCA is larger than the dense csv. This may be because we're storing large integers. May be less after normalization. It may simplly be that the count data is not as sparse as imagined. We really need to reduce to the subset of genes available for gene2vec intersection HiG2Vec, and that will greatly reduce the size of both. The libsvm format is good for text processing, I think but csr matrices or even dense matrices could be just as good if the sparsity is not so bad.
