sdfdsf

[rajesht2418]

Issue Type

Bug

Source

source

Tensorflow Version

2.8

Custom Code

Yes

OS Platform and Distribution

No response

Mobile device

No response

Python version

No response

Bazel version

No response

GCC/Compiler version

No response

CUDA/cuDNN version

No response

GPU model and memory

No response

Current Behaviour?

Hi! I noticed that `tf.ForwardAccumulator` is currently issuing a warning when it is used to compute a jvp of a real-valued function for complex input. I'm not sure if this is a historic artifact since the result seems correct despite the warning as the code example below demonstrates. I would appreciate if someone could clarify why the warning exists.

The example below implements two Hessian-vector product (hvp) functions, one based on reverse-over-reverse mode using two `tf.GradientTape` instances (i.e., the gradient of the directional derivative), and one based on forward-over-reverse mode using `tf.ForwardAccumulator` and `tf.GradientTape`. The latter issues the warning `WARNING:tensorflow:The dtype of the watched primal must be floating (e.g. tf.float32), got tf.complex64` which originates here: https://github.com/tensorflow/tensorflow/blame/d8ce9f9c301d021a69953134185ab728c1c248d3/tensorflow/python/eager/forwardprop.py#L399-L402.

Standlone code to reproduce the issue

Hi! I noticed that `tf.ForwardAccumulator` is currently issuing a warning when it is used to compute a jvp of a real-valued function for complex input. I'm not sure if this is a historic artifact since the result seems correct despite the warning as the code example below demonstrates. I would appreciate if someone could clarify why the warning exists.

The example below implements two Hessian-vector product (hvp) functions, one based on reverse-over-reverse mode using two `tf.GradientTape` instances (i.e., the gradient of the directional derivative), and one based on forward-over-reverse mode using `tf.ForwardAccumulator` and `tf.GradientTape`. The latter issues the warning `WARNING:tensorflow:The dtype of the watched primal must be floating (e.g. tf.float32), got tf.complex64` which originates here: https://github.com/tensorflow/tensorflow/blame/d8ce9f9c301d021a69953134185ab728c1c248d3/tensorflow/python/eager/forwardprop.py#L399-L402.

Relevant log output

Hi! I noticed that `tf.ForwardAccumulator` is currently issuing a warning when it is used to compute a jvp of a real-valued function for complex input. I'm not sure if this is a historic artifact since the result seems correct despite the warning as the code example below demonstrates. I would appreciate if someone could clarify why the warning exists.

The example below implements two Hessian-vector product (hvp) functions, one based on reverse-over-reverse mode using two `tf.GradientTape` instances (i.e., the gradient of the directional derivative), and one based on forward-over-reverse mode using `tf.ForwardAccumulator` and `tf.GradientTape`. The latter issues the warning `WARNING:tensorflow:The dtype of the watched primal must be floating (e.g. tf.float32), got tf.complex64` which originates here: https://github.com/tensorflow/tensorflow/blame/d8ce9f9c301d021a69953134185ab728c1c248d3/tensorflow/python/eager/forwardprop.py#L399-L402.

[csat-bot[bot]]

Thanks for opening this issue!

[csat-bot[bot]]

Thanks for opening this issue!