kaknikhil opened a new pull request #375: DL: Add GPU support for evaluate URL: https://github.com/apache/madlib/pull/375 In our experiments on a multi host gpdb cluster, we noticed that using CPU and compiling the model everytime caused the fit query to fail during the validate code with the error `Dst tensor is not initialized`. Adding GPU and caching the compiled model fixed this issue. 1. Add GPU support to the evaluate code called by the fit function. 2. The evaluate code previously compiled the model for each row in the validation table. This was unnecessary and takes more resources in terms of time and memory. This commit avoids this by using SD to cache the compiled model. 3. Also refactored the code by creating a function to get the rows per segment
---------------------------------------------------------------- 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. For queries about this service, please contact Infrastructure at: [email protected] With regards, Apache Git Services
