Support for multiple training and testing files

[hdidwania]

Is your feature request related to a problem? Please describe.
The path to file containing test data is fixed, making it clumsy to evaluate multiple slices (e.g. split by language).
Similarly the train data has to be merged into a single file. Both of these is opposed to the nature of data we get i.e. different files from different sources.

Describe the solution you'd like
Code would be able to:

• take multiple csv splits for training data and merge together to train
• take multiple csv splits for testing data and report individually on those

Describe alternatives you've considered
Manually doing these:

• Merging the training data in a large file before training
• Replacing test.csv with suitable split for evaluating