Classifier serial release in scikit learn Pipeline

Question

I use Pipelinein scikit to learn to group some preprocessing together with OneClassSVMas a final classifier. To calculate reasonable metrics, I need a post-processing that converts the -1.1 output OneClassSVMto 0 and 1. Is there any structured way to add such post-processing to Pipeline? Transformers cannot be used after final evaluation.

Answer 1

We developed PipeGraph, an extension for Scikit-Learn Pipeline, which allows you to receive intermediate data, build a graph as workflows and, in particular, solve this problem (see examples in the gallery http://mcasl.imtqy.com/PipeGraph )