DavidSandruTU edited a comment on pull request #1357: URL: https://github.com/apache/systemds/pull/1357#issuecomment-890359063
Some additional notes: - The MatrixMult and MatrixTranspose microbenchmarks were not written with SPARK in mind. They both call systemds with perf tools, for which I'm unsure if it is available on the cluster? For now they are commented in the runAll script, but should be executable locally. Should I adapt them so they call ./SparkDML.sh if specified? - The built-in MSVM (via scripts/perftest/scripts/m-svm.dml) is ~80% slower than the fully implemented MSVM algorithm in scripts/algorithms/m-svm.dml. I'm unsure why. This was tested with the Intel MKL. - I've subtracted ".4" for all the time calculations in the bash scripts, to account for bash script and JVM overhead. I'm assuming this number is dependent on the machine it's ran on, so I'm unsure if this was a good choice. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
