tree d780d7fa0b13a862fd6d94b32df533f1b0ccc885
parent 5596fcf0d77946cf11ff34085e8879552dd852be
author Sebastian <baunsgaard@tugraz.at> 1590948578 +0200
committer baunsgaard <baunsgaard@tugraz.at> 1599387528 +0200

[SYSTEMDS-371] ColGroup Quantization

New ColGroupQuan smiley that internally changes all values to Byte.
It is compressing by using the highest value (or Math.abs(lowest))
in the matrix block, and therefore the bucket sizes are dependent
on that value.

This implementation that does not exploit sparsity but it gives an
alternative col group for columns that are previously uncompressed.

The testing associated with this new lossy representation is based
around either percentage deviation from ideal results (double precision)
or loss threshold calculated from the worst rounding of values in that
input matrix. (they currently show only few percent deviation in average
results)