[MADLIB-1309] Change the way we pass compile_params and fit_params to madlib_keras.fit and madlib_keras.predict

• Type: Improvement
• Status: Closed
• Priority: Major
• Resolution: Fixed
• Fix Version/s: v1.16
• Component/s: Deep Learning
• Assignee: Frank McQuillan
• Reporter: Nandish Jayaram
• Created: 20/Mar/19 17:36
• Updated: 06/Jun/19 00:33
• Resolved: 06/Jun/19 00:33

Description

Based on the code in PR https://github.com/apache/madlib/pull/355:

Currently we have to pass in the compile_params and fit_params as follows

SELECT madlib.madlib_keras_fit(... , '''optimizer''=SGD(lr=0.01, decay=1e-6, nesterov=True), ''loss''=''categorical_crossentropy'', ''metrics''=[''accuracy'']'::text,
 ,'''batch_size''=2, ''epochs''=1, ''verbose''=0'::text ...)

Internally in the code we use a function called `convert_string_of_args_to_dict` which does some string parsing and then calls eval to create a dictionary. This is bug prone and unsafe. We should look into using the function `extract_keyvalue_params` (might have to change it. We can easily write a unit test for `extract_keyvalue_params`)

We should also think about how complicated compile_params can get and how can we generalize the code.