[MADLIB-1330] DL: Do not compile model for predict and add GPU support

• Type: Improvement
• Status: Closed
• Priority: Major
• Resolution: Fixed
• Fix Version/s: v1.16
• Component/s: Deep Learning
• Assignee: Nandish Jayaram
• Reporter: Nandish Jayaram
• Created: 25/Apr/19 18:22
• Updated: 06/Jun/19 00:40
• Resolved: 06/Jun/19 00:40

Description

In internal_keras_predict, we compile the model for every row. Starting from keras 1.0.3, keras predict does not need a compiled model. This will make our current predict code(which does not use SD) easier and hopefully slightly faster even with CPU. See more details here https://github.com/keras-team/keras/issues/3074

Acceptance:
1. Validate that removing the compile code does not impact the accuracy of the results
2. Add new optional param named use_gpu in predict interface.