Support to take Python object directly for scoring

[yueguoguo]

If I did not miss anything, currently, the scoring operation of a model takes a txt file. It would be great that the model can directly take a Python object like pandas dataframe, numpy array, or dict, which can make the model object more useful for scoring.

Imaging a scenario where a query is made to use the model to score a given input, one needs to convert it to an txt file before the scoring can happen. This will induce overhead.