Packing estimator and mapper together about sklearn2pmml

[vruusmann]

I've been thinking about this problem lately in the context of an online conversion service - it would be much more easier/intuitive to build an UI for a single file upload vs. two file uploads.

The answer to this question is to add support for sklearn.pipeline.Pipeline class (see jpmml/jpmml-sklearn#3). Basically, you would be creating a two-step pipeline, where the first step is a DataFrameMapper, and the second step is an Estimator of your choice:

pipeline = Pipeline([
  ("mapper", iris_mapper),
  ("estimator", iris_classifier)
])

joblib.dump(pipeline, "iris_pipeline.pkl")

Also, Pipelines are needed anyway in order to support Scikit-Learn feature extraction and selection transformations (see #15; you could introduce any number of steps between the initial "mapper" step and the final "estimator" step).