When in webgl backend, tf.isNaN() produces incorrect results #5800
Description
Please make sure that this is a bug. As per our
GitHub Policy,
we only address code/doc bugs, performance issues, feature requests and
build/installation issues on GitHub. tag:bug_template
System information
- Have I written custom code (as opposed to using a stock example script provided in TensorFlow.js):yes
- OS Platform and Distribution (e.g., Linux Ubuntu 16.04): Ubuntu 20.04.2 LTS
- Mobile device (e.g. iPhone 8, Pixel 2, Samsung Galaxy) if the issue happens on mobile device:
- TensorFlow.js installed from (npm or script link): script link
- TensorFlow.js version (use command below): 3.11.0
- Browser version: Google Chrome 93.0.4577.63
- Tensorflow.js Converter Version:
Describe the current behavior
tf.isNaN() has different results when running on different backends. For example, for the following code:
const y = tf.tensor1d([Infinity, .5, 4, -38.8, NaN]);
y.isNaN().print();
In webgl backend, the output is
Tensor
[false, false, false, false, false]
In cpu backend, the output is
Tensor
[false, false, false, false, true]
In wasm backend, get the error is
Uncaught (in promise) Error: Kernel 'IsNan' not registered for backend 'wasm'
These results indicate that tf.isNaN()cannot correctly recognize NaN in webgl backend.
Describe the expected behavior
According to the official documentation，tf.isNaN(x) RReturns which elements of x are NaN.
Therefore, the correct result should be
Tensor
[false, false, false, false, true]
Standalone code to reproduce the issue
Provide a reproducible test case that is the bare minimum necessary to generate
the problem. If possible, please share a link to Colab/CodePen/any notebook.
the code to reproduce the problem is
<!DOCTYPE html>
<html lang="en">
<head>
<meta charset="UTF-8">
<title>Title</title>
<script src="https://cdn.jsdelivr.net/npm/@tensorflow/tfjs/dist/tf.min.js"> </script>
<script src="https://cdn.jsdelivr.net/npm/@tensorflow/tfjs-core"></script>
<script src="https://cdn.jsdelivr.net/npm/@tensorflow/tfjs-backend-webgl"></script>
<script src="https://cdn.jsdelivr.net/npm/@tensorflow/tfjs-backend-wasm/dist/tf-backend-wasm.js"></script>
</head>
<body>
<script>
async function nan_test() {
const y = tf.tensor1d([Infinity, .5, 4, -38.8, NaN]);
tf.setBackend("webgl");
await tf.ready();
console.log("webgl backend result:");
y.isNaN().print();
tf.setBackend("cpu");
await tf.ready();
console.log("cpu backend result:");
y.isNaN().print();
tf.setBackend("wasm");
await tf.ready();
console.log("wasm backend result:");
y.isNaN().print();
}
nan_test();
</script>
</body>
</html>
Other info / logs Include any logs or source code that would be helpful to
diagnose the problem. If including tracebacks, please include the full
traceback. Large logs and files should be attached
the code to reproduce the problem
nan_test.zip
.