Add gradients for tf.min and tf.max

[caisq]

Fixes tensorflow/tfjs#677

This change is 

[nsthorat]

Reviewed 2 of 2 files at r1.
Reviewable status: 0 of 1 approvals obtained (waiting on @caisq and @nsthorat)

---

src/ops/reduction_ops.ts, line 223 at r1 (raw file):

  return {
    $x: () => {
      const dx = dy.mul(xOrig.equal(y).cast(dy.dtype));

dy.dtype is always float32

---

src/ops/reduction_ops_test.ts, line 200 at r1 (raw file):

      [[[0, 20], [20, 20]], [[-10, -30], [-10, -30]]]
    ]);
    const dy = tf.tensor4d([[[[-1]], [[-1]]], [[[-1]], [[-1]]]]);

small nit, can you make the values of dy be -1, -2, -3, -4 etc here and above? this will ensure that the first value isn't being placed everywhere

[caisq]

Reviewable status: complete! 1 of 1 approvals obtained (waiting on @caisq)

---

src/ops/reduction_ops.ts, line 223 at r1 (raw file):

Previously, nsthorat (Nikhil Thorat) wrote…

dy.dtype is always float32

It seems to be less magical and potentially more maintainable to use dy.dtype here. Also, I don't think that call is particularly expensive.

---

src/ops/reduction_ops_test.ts, line 200 at r1 (raw file):

Previously, nsthorat (Nikhil Thorat) wrote…

small nit, can you make the values of dy be -1, -2, -3, -4 etc here and above? this will ensure that the first value isn't being placed everywhere

Done.