I dont think the gradients wrt to `points` is right? about tensorflow-nufft

[jmontalt]

I'm glad you're finding it useful! I have quickly tested your suggested change against the naive NUDFT implementation currently used for testing TensorFlow NUFFT.

Testing passes with or without your change, which suggests the tests themselves may be incomplete. If you are able to create a PR that includes a test that passes with your change but fails without it, that would be amazing. Thank you!

[chaithyagr]

Hmm that's weird, I remember physically checking that the values were off just taking NDFT from your library.
Anyway I will make a PR tomorrow.

[jmontalt]

Hi @chaithyagr, the problem is that the current test does not include any further operations originating from the NUFFT. As a result, the upstream gradient is 1, and taking its conjugate simply has no effect. However, when adding any operation that results in a complex-valued upstream gradient, your fix does indeed become necessary to obtain the correct values.

Since it's such a simple fix, there is no need for you to submit a PR, although by all means do send it if you've already worked on it. Otherwise, I'll sort this out later today and release a new version that incorporates the fix. Thank you for pointing out this issue!

[chaithyagr]

That's wonderful that you have figured this out. I'm sorry that there is some maintainance at my end and hence I can't send in the PR today. Feel free to do so yourself and close this issue.

[jmontalt]

This issue has been addressed with Release 0.7.1. Thank you @chaithyagr !