khannaekta opened a new pull request #378: DL: Update madlib_keras_fit code to weight by images instead of buffers URL: https://github.com/apache/madlib/pull/378 JIRA: MADLIB-1310 Previously, while merging the weights, accuracy and loss, we assume that all the buffers are of the same size. This may not be always the case. Ideally we should use the total no of images per segment to average the weights, accuracy and loss. This commit, updates the transition, merge and final function for `madlib_keras_fit()` to use the number of images on a segment instead of the buffer count for averaging weights, accuracy and loss. Additionally, the function `deserialize_weights()` when deserializing loss and accuracy from state, type-casted it to an int, which resulted in computing lower loss and accuracy while training. This PR also addresses this issue.
---------------------------------------------------------------- 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
