Fix histogram op for symbolic inputs

[abheesht17]

Resolves #21708

On the TensorFlow backend, ops.histogram converts a tensor to a Python list [here], which won't work in graph mode with symbolic tensors. The current piece of code uses tf.raw_ops.Bucketize which necessarily requires boundaries to be a Python list [see documentation].

This PR:

• Replaces tf.raw_ops.Bucketize with tf.searchsorted;
• Adds a unit test to make sure the op runs with predict;
• Also adds a missing test (unrelated to the specific issue linked above), which ensures we are assigning values which are "on the boundaries" to the correct bucket. This is an important unit test to have.

We should probably add a unit test which tests model.predict for every op at some point.

Note: Skipping the NumPy test for predict. Seems to be an unrelated issue. We can solve it separately.

[codecov-commenter]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 82.59%. Comparing base (4e1edcf) to head (ec9fa13).

Additional details and impacted files

@@           Coverage Diff           @@
##           master   #21729   +/-   ##
=======================================
  Coverage   82.59%   82.59%           
=======================================
  Files         572      572           
  Lines       58535    58535           
  Branches     9158     9158           
=======================================
  Hits        48345    48345           
  Misses       7853     7853           
  Partials     2337     2337           

Flag	Coverage Δ
keras	82.39% <100.00%> (ø)
keras-jax	63.19% <0.00%> (ø)
keras-numpy	57.56% <0.00%> (+<0.01%)	⬆️
keras-openvino	34.33% <0.00%> (ø)
keras-tensorflow	63.94% <100.00%> (ø)
keras-torch	63.49% <0.00%> (ø)

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[abheesht17]

Works with jit_compile = False, but does not work with jit_compile = True. Trying to figure out another way

FAILED keras/src/ops/numpy_test.py::HistogramTest::test_histogram_predict - tensorflow.python.framework.errors_impl.InvalidArgumentError: Graph execution error:

Detected at node functional_1/histogram_layer_1/bincount/DenseBincount defined at (most recent call last):
<stack traces unavailable>
Input 1 to node `functional_1/histogram_layer_1/bincount/DenseBincount` with op DenseBincount must be a compile-time constant.

XLA compilation requires that operator arguments that represent shapes or dimensions be evaluated to concrete values at compile time. This error means that a shape or dimension argument could not be evaluated at compile time, usually because the value of the argument depends on a parameter to the computation, on a variable, or on a stateful operation such as a random number generator.

[abheesht17]

Works with jit_compile = False, but does not work with jit_compile = True. Trying to figure out another way

FAILED keras/src/ops/numpy_test.py::HistogramTest::test_histogram_predict - tensorflow.python.framework.errors_impl.InvalidArgumentError: Graph execution error:

Detected at node functional_1/histogram_layer_1/bincount/DenseBincount defined at (most recent call last):
<stack traces unavailable>
Input 1 to node `functional_1/histogram_layer_1/bincount/DenseBincount` with op DenseBincount must be a compile-time constant.

XLA compilation requires that operator arguments that represent shapes or dimensions be evaluated to concrete values at compile time. This error means that a shape or dimension argument could not be evaluated at compile time, usually because the value of the argument depends on a parameter to the computation, on a variable, or on a stateful operation such as a random number generator.

Alright, played some ops golf here, and replaced bincount with scatter_nd_add. Seems to work now

[hertschuh]

It looks like you can just use tf.scatter_nd:

• it creates a new tensor with zeros
• it also sums in the case of duplicate indices

[abheesht17]

Thanks, done!