Implement numOutputs for minimum

[nathantalewis]

This seems very similar to #129.

The code in question:

let y = (TF.minimum 0 (TF.expm1 x)) `TF.add` (TF.relu x)

The error message: numOuputs not implemented for "Minimum"

I'm using commit 79d8d7e.

Alternatively, if there is an existing implementation of elu that I could use, how would I use it?

[fkm3]

Yeah, it is the same general problem, someone needs to add the gradient implementation for minimum. You will also hit the same issue for expm1 I'd guess.

There is already an implementation for maximum. The minimum version should be exactly the same, but instead of greaterEqual, it should use lessEqual here https://github.com/tensorflow/haskell/blob/master/tensorflow-ops/src/TensorFlow/Gradient.hs#L540

[fkm3]

There is a builtin implementation of elu https://tensorflow.github.io/haskell/haddock/tensorflow-core-ops-0.1.0.0/TensorFlow-GenOps-Core.html#v:elu and its gradient function https://tensorflow.github.io/haskell/haddock/tensorflow-core-ops-0.1.0.0/TensorFlow-GenOps-Core.html#v:eluGrad, so that might be the easier route. (though you'd still need to plug that gradient function into the Gradient.hs framework)

[nathantalewis]

I must have missed the built-in version. Thanks for pointing it out. If I can get the gradient function plugged in I'll make a PR for it.