h2o4gpu
h2o4gpu copied to clipboard
h2oai
Logging framework
Add a proper logging framework for C++ and Python enabling debug prints and (maybe) performance timings.
@navdeep-G for C/C++ yes but we still need something in Python (and possibly R?).
I'm not sure what's the best practice for R but in Python we probably can just use the standard import logging. We just need to map the verbosity number to a String level representation (INFO, DEBUG etc.). Ideally we'd add a Python/R call to our C/C++ backend which would map int -> String according to what is hardcoded in the C code (there are definitions for those in the logger files) so we don't have to hardcode that for example 500 is DEBUG several times in C, Python and R code.
We just need to expose the verbosity in Python API and then we can modify it as an option in R. We can also use futile.logger package if needed.
Yes, only needs to be in python and we can use reticulate to invoke it in R
python already has logging, used in DAI alot: https://docs.python.org/3/library/logging.html
@pseudotensor yes, we just need to replace the print statements in our code for those logger calls and set the log level appropriately based on the verbosity level.
Copy/paste the logger setup arno did for DAI. It's non-trivial to set it up and now it's very nice for DAI.
This isn't really very high priority, we can do this after we have tSVD in DAI and are done with preparations for GTC I think.
Agreed on priority, just laying out what would be done to someone doesn't do crazy amounts of independent work.
We should try cusolver in DAI, easy to try, just busy getting release out.