creating consistency is great, and very welcome. However, the concrete
proposal is not ideal as it does not follow the common convention of
capital letters for matrices, small letters for vectors and scalars. In
papers you often see capital bold for matrices, small bold for vectors,
and small regular for scalars. This convention really helps to quickly
understand the characteristics of matrix-vector multiplications etc.
So I would recommend to change the algorithms like multiLogReg and l2svm
where this was missed initially. Right now, multiLogReg takes a vector
(should be y) and internally does one-hot encoding to get Y (reassigning
one Y variable is easy to get missed but completely changes shapes and
characteristics).
Regards,
Matthias
On 5/16/2022 6:43 PM, Janardhan Pulivarthi wrote:
Thanks a lot, for starting this much needed discussion.
Best Regards,
Janardhan
On Mon, May 16, 2022 at 8:04 PM Baunsgaard, Sebastian
<[email protected]> wrote:
Hi All,
I would like to suggest that we change all builtin functions to use capital
letters for single letter arguments.
Functions currently use small and large letters a randomly,
and it is confusing when changing algorithm you have to sometimes change "Y" to
"y".
examples that would break this:
https://github.com/apache/systemds/blob/main/scripts/builtin/arima.dml
https://github.com/apache/systemds/blob/main/scripts/builtin/cvlm.dml
https://github.com/apache/systemds/blob/main/scripts/builtin/deepWalk.dml
https://github.com/apache/systemds/blob/main/scripts/builtin/getAccuracy.dml
https://github.com/apache/systemds/blob/main/scripts/builtin/gridSearch.dml
(All Img builtin)
https://github.com/apache/systemds/blob/main/scripts/builtin/kmeans.dml
https://github.com/apache/systemds/blob/main/scripts/builtin/lasso.dml
https://github.com/apache/systemds/blob/main/scripts/builtin/lm.dml
https://github.com/apache/systemds/blob/main/scripts/builtin/lmCG.dml
https://github.com/apache/systemds/blob/main/scripts/builtin/lmDS.dml
(All outlier)
etc..
I would also like to suggest that we change all instances where we call X and Y
for X_Train or Y_Train to just X and Y.
example that break:
https://github.com/apache/systemds/blob/main/scripts/builtin/hyperband.dml
https://github.com/apache/systemds/blob/main/scripts/builtin/knn.dml
https://github.com/apache/systemds/blob/main/scripts/builtin/lasso.dml
https://github.com/apache/systemds/blob/main/scripts/builtin/sherlock.dml
All verbose booleans should be changed to "verbose"
https://github.com/apache/systemds/blob/main/scripts/builtin/matrixProfile.dml
best regards
Sebastian
