thunterdb
commented
6 years ago Keras is meant to train deep learning models while MLlib is a much more general framework that includes algorithms such as random forests, recommendation systems, etc. The function that was used contained some export logic that was specific to logistic and linear regression (which map well to keras equivalents). It has not been released because we feel it does not cover enough use cases to be useful at this stage.
Can you tell us more about your use cases?
I am new to SparkDL and basically following available resources. One of the video shows (https://databricks.com/blog/2017/06/06/databricks-vision-simplify-large-scale-deep-learning.html) , a Spark DL model got converted into Keras model before registering as an UDF. Check the "mllib_to_keras_model" function used by the presenter. Could not find that function anywhere ?