reductionista opened a new pull request #356: Keras model arch table helper 
functions for keras_fit()
URL: https://github.com/apache/madlib/pull/356
 
 
   This PR introduces two new helper functions, `load_keras_model()` and 
`delete_keras_model()`.  It is a part of the new deep learning functionality 
being added to madlib.
   
   These functions will help prepare the input to the `keras_fit()` function 
being introduced in PR [#355](https://github.com/apache/madlib/pull/355).
   
   `keras_fit()` takes a model_arch_table param and a model_id param.  This 
refers to a row in a "keras model arch table" which has to have a particular 
format.  These functions will help the user to create and manage that table, 
without worrying about the details of the format.
   
   Each row of the keras model arch table represents a keras model 
architecture, and is identified by a model_id.  `load_keras_model()` will add a 
new model architecture to the table, creating the table if it doesn't exist and 
then inserting the appropriate row into it.  The newly assigned model_id will 
be printed for the user, who can then pass that as the model_id parameter to 
`keras_fit()`.
   
   The model architectures themselves are expected to be in JSON format, which 
can be achieved by calling `model.to_json()` on any keras model after it is 
created (in python).
   
   The keras model arch table also has a column to store weights.  This is 
initialized to `null` by `load_keras_model()` but we plan to add another helper 
function in the near future to be able to set the weights (useful for doing a 
warm start, instead of letting keras randomly initialize the weights when 
training begins).
   
   The keras model arch table itself will be automatically dropped by 
`delete_keras_model()` after all of the models in it are deleted and the table 
is empty.
   
   To help avoid mistakes, if either of these functions is run on a table that 
doesn't match the expected format, and exception will be thrown and the table 
will be left untouched.

----------------------------------------------------------------
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

Reply via email to