Feature request: CSV Reader

[tomvars]

This issue is to track progress on new features for the CSVReader

The current version is in the refactoring-data-app branch. Its intended use is to be called within any Sampler such that the user can load CSV data as easily as they can load images. This data could be used as input to a network or as the target.

Features

• Read in CSV as network target
• Read in CSV as arbitrary features
• Write a Sampler which uses information from a CSV to specify which patches to sample
• Documentation
• Tests

[wyli]

perhaps leave a flexible interface for the advanced features as well:

• Read in CSV as network targets of variable length sequences
• Support preprocessing/data augmentation layers e.g. read in CSV network targets as texts and encode them as numerical vectors, like layer/discrete_label_normalisation.py

--
relevant issues:

• https://cmiclab.cs.ucl.ac.uk/CMIC/NiftyNet/issues/88:
  Allow for patch sampling from GZip images without loading the full dataset
  Use this package: https://github.com/pauldmccarthy/indexed_gzip

• Label issue for Classification task #42