Tensor.clipByValue does not work for tensors of dtype int32 in Node backend

[arcman7]

TensorFlow.js version 2.0.1

node version 12.13.0

backend: "tensorflow"

OS: Windows 10

Issue: Tensor.clipByValue does not work for tensors of dtype int32.

Code to reproduce issue:

const x = tf.tensor([-1, 2, -3, 4], undefined, 'int32');
x.clipByValue(-2, 4).print();

This throws an error with the stack trace:

Thrown:
Error: Invalid TF_Status: 3
Message: cannot compute Minimum as input #1(zero-based) was expected to be a int32 tensor but is a float tensor
    at NodeJSKernelBackend.executeSingleOutput (C:\Users\Ryan\projects\pysc2\node_modules\@tensorflow\tfjs-node\dist\nodejs_kernel_backend.js:192:43)
    at NodeJSKernelBackend.minimum (C:\Users\Ryan\projects\pysc2\node_modules\@tensorflow\tfjs-node\dist\nodejs_kernel_backend.js:558:21)
    at NodeJSKernelBackend.clip (C:\Users\Ryan\projects\pysc2\node_modules\@tensorflow\tfjs-node\dist\nodejs_kernel_backend.js:636:25)
    at ENGINE.runKernelFunc.x (C:\Users\Ryan\projects\pysc2\node_modules\@tensorflow\tfjs-node\node_modules\@tensorflow\tfjs-core\dist\tf-core.node.js:6882:27)
    at C:\Users\Ryan\projects\pysc2\node_modules\@tensorflow\tfjs-node\node_modules\@tensorflow\tfjs-core\dist\tf-core.node.js:3229:55
    at C:\Users\Ryan\projects\pysc2\node_modules\@tensorflow\tfjs-node\node_modules\@tensorflow\tfjs-core\dist\tf-core.node.js:3075:22
    at Engine.scopedRun (C:\Users\Ryan\projects\pysc2\node_modules\@tensorflow\tfjs-node\node_modules\@tensorflow\tfjs-core\dist\tf-core.node.js:3085:23)
    at Engine.tidy (C:\Users\Ryan\projects\pysc2\node_modules\@tensorflow\tfjs-node\node_modules\@tensorflow\tfjs-core\dist\tf-core.node.js:3074:21)
    at kernelFunc (C:\Users\Ryan\projects\pysc2\node_modules\@tensorflow\tfjs-node\node_modules\@tensorflow\tfjs-core\dist\tf-core.node.js:3229:29)

Please note that it appears to work just fine in the browser using the webgl backend.

[arcman7]

I'm not sure exactly where the issue is in the code base, but just writing this helper method gets me around the issue for now-

function clipByValue(x, min, max) {
  const minT = tf.fill(x.shape, min, x.dtype);
  const maxT = tf.fill(x.shape, max, x.dtype);
  return x.where(x.greaterEqual(min), minT).where(x.lessEqual(max), maxT);
}

I noticed that when you rely on broadcasting to add a scalar value to an existing tensor, the resulting tensor dtype can change. For example:

const x = tf.tensor([-1, 2, -3, 4], undefined, 'int32');
const y = x.add(4); // tensor y has dtype float32

I think this may have something to do with the issue I'm experiencing.

[google-ml-butler]

Are you satisfied with the resolution of your issue?
Yes
No