When you use a minibatch data source, you need a slightly different setup for the loss and metric. Let's go back and review how you can set up training using a minibatch source and extend it with metrics to validate the model. First, we need to set up a way to feed data to the trainer of the model:
from cntk.io import StreamDef, StreamDefs, MinibatchSource, CTFDeserializer, INFINITELY_REPEATdef create_datasource(filename, limit=INFINITELY_REPEAT): labels_stream = StreamDef(field='labels', shape=3, is_sparse=False) features_stream = StreamDef(field='features', shape=4, is_sparse=False) deserializer = CTFDeserializer(filename, StreamDefs(labels=labels_stream, features=features_stream)) ...
