Input Converters and General Pipeline units

[sjperkins]

Presently the units of the internal tensorflow graph are fixed. For e.g.

• lm coordinates in radians
• linear (XX, XY, YX, YY) correlations

and the user must convert their data to/from this format in the Source/Sink Providers. However, it may be useful to allow the user to specify conversion functions that convert data into these formats. For e.g.

• lm coordinates in radec to radians
• circular (LL, RL, LR, RR) correlations to (XX, XY, YX, YY)

This involves writing custom CPU/GPU operators for each conversion function which would then be invoked internally within the tensorflow graph. This is a nice to have, but not critical feature.