Add predict api to estimator

[buaazhwb]

For now, user can only obtain predictions by getting the model and then predicting

[ConeyLiu]

what input and output should we want? And for predicting, does it should be the model serving part?

[buaazhwb]

I'm not sure whom raydp's target user is. Looks like the evaluate function is enough.
If I'm a kaggler, I need the prediction result of test dataset which is without label and submit the result to the competition. For now, I need to get the model and predict by myself.
The purpose of predicting is different from serving. We just want to obtain the prediction in a python script

Input should support both RayDp dataset and spark dataframe. For output, should it be consistent with input or just numpy array/pandas dataframe?

[kira-lin]

close as stale