v3.0.0

v3.0.0 (2026-06-11)

Breaking

• feat: Release/v3.0.0 (#51)

BREAKING CHANGE: * feat: Allow layer and output names to be equal (#42)

• Previously we did not allow this and added an identity layer to all model outputs

• feat: Update hash indexer null behaviour (#41)

• feat: Update hash indexer null behaviour

• Previously nulls were passed through Spark as nulls
• This can create issues with training data
• Instead we mimic the string indexer behaviour by keeping 0 as the null/mask index.
• And shifting everything else up by one index.

• fix: Ensure num bins > 1

• 1 bin doesn't really make sense and could error as nulls always go to bin 0.

• feat: Bring release v3 up to date with latest main (#45)

• feat: pairwise sim and array reduce max (#44)

• adding modules for pairwise similarity

• tests for pairwise similarity

• adding the new modules on README and tests of serialisation

• formatting issues

• fix header

• 2.40.0

Automatically generated by python-semantic-release

---

Co-authored-by: Youssef Achenchabe <youssef.achenchabe@gmail.com>
Co-authored-by: semantic-release <semantic-release>

• feat: Keras3 migration (#40)

• docs: README restructured

• Installation moved further up, above the massive table
• Quick start shows a quick example
• Sklearn removed as a proper usage pattern
• Reduced size by ~40%
• Much more readable

• feat: add Keras 3 multi-backend support with portable layers

• Remove Keras 2 version detection and TypeSpec support
• Update dependencies: keras>=3.0.0, tensorflow>=2.16.0
• Add keras.core package with BaseLayer for multi-backend layers
• Add keras.tensorflow package with TfBaseLayer for TF-specific layers
• Port 5 MVP layers to multi-backend: identity, absolute_value, multiply, exp, log
• Add 34 passing tests for portable layer infrastructure

• feat: migrate TF-only layers to keras.tensorflow package

Move 35 TensorFlow-specific layers from kamae.tensorflow.layers to kamae.keras.tensorflow.layers as part of Keras 3 multi-backend migration.

These layers require TensorFlow backend and cannot be made portable:

• 5 hash/encoding layers (BloomEncode, Bucketize, HashIndex, etc.)
• 8 datetime layers (CurrentDate, DateParse, UnixTimestampToDateTime, etc.)
• 7 list operations (ListMax, ListMean, ListMedian, etc.)
• 14 string layers (StringConcatenate, StringIndex, StringContains, etc.)
• 1 lambda layer (LambdaFunction)

Also migrate TF-specific utilities to kamae.keras.tensorflow.utils:

• date_utils.py: 18 datetime functions (unix_timestamp_to_datetime, etc.)
• list_utils.py: 6 list operations (get_top_n, segmented_operation, etc.)
• transform_utils.py: 4 map_fn functions
• typing.py: TF-specific Tensor type (includes SparseTensor, RaggedTensor)

All TensorFlow operations remain byte-identical to originals. Only changes:

• Base class: BaseLayer → TfBaseLayer (adds require_tensorflow() check)
• Import paths updated to new package structure
• Input decorators now use portable keras.core.utils.input_utils

Numeric layers (divide, subtract, sum, etc.) remain in old location to be
properly ported to multi-backend in next commits.

• feat: add 5 portable numeric layers (divide, subtract, round, modulo)

Migrate divide, subtract, round, round_to_decimal, and modulo layers from kamae.tensorflow.layers to kamae.keras.core.layers. .

Changes:

• divide.py: Implemented divide_no_nan using ops.where to
  handle division by zero (returns 0 instead of NaN/Inf)
• subtract.py: Direct port using ops.subtract
• round.py: Direct port using ops.ceil/floor/round
• round_to_decimal.py: Uses numpy.finfo/iinfo for dtype max values
  instead of TF-specific tensor.dtype.max
• modulo.py: Port using ops.mod (equivalent to tf.math.floormod)

All layers:

• Use keras.ops instead of tf.math operations
• Import from keras.core.layers.base (BaseLayer)
• Use portable decorators from keras.core.utils.input_utils
• Use keras.saving.register_keras_serializable (not tf.keras.utils)
• Return string dtype names (not tf.dtypes.DType objects)

• feat: add 5 portable numeric layers (sum, max, min, mean, exponent)

Migrate sum, max, min, mean, and exponent layers from kamae.tensorflow.layers to kamae.keras.core.layers.

New layers:

• SumLayer: Element-wise addition with addend constant or reduce multiple tensors
• MaxLayer: Element-wise maximum with max_constant or reduce multiple tensors
• MinLayer: Element-wise minimum with min_constant or reduce multiple tensors
• MeanLayer: Element-wise mean with mean_constant or reduce multiple tensors
• ExponentLayer: Raise tensor to power (x^exponent)

Implementation:

• sum.py: Uses ops.add with functools.reduce for multiple inputs
• max.py: Uses ops.maximum with functools.reduce
• min.py: Uses ops.minimum with functools.reduce
• mean.py: Uses ops.add + ops.true_divide(result, len(inputs))
• exponent.py: Uses ops.power for x^y operation

All layers follow portable patterns:

• keras.ops instead of tf.math operations
• keras.core.layers.base.BaseLayer as parent
• keras.core.utils.input_utils decorators
• keras.saving.register_keras_serializable
• String dtype names (not tf.dtypes.DType objects)

• feat: add 3 portable logical layers (and, or, not)

Migrate logical_and, logical_or, and logical_not layers from
kamae.tensorflow.layers to kamae.keras.core.layers.

These layers are now backend-agnostic and work with TensorFlow, JAX,
and PyTorch.

New layers:

• LogicalAndLayer: Element-wise AND operation on multiple boolean tensors
• LogicalOrLayer: Element-wise OR operation on multiple boolean tensors
• LogicalNotLayer: Element-wise NOT operation on a single boolean tensor

Implementation:

• logical_and.py: Uses ops.logical_and with functools.reduce
• logical_or.py: Uses ops.logical_or with functools.reduce
• logical_not.py: Uses ops.logical_not for single tensor

All layers:

• Only support "bool" dtype
• Use enforce_multiple_tensor_input (and/or) or enforce_single_tensor_input (not)
• Use keras.ops instead of tf.math operations
• Follow portable layer patterns

• feat: add portable numerical_if_statement, move if_statement to TF-only

Migrate numerical_if_statement to kamae.keras.core.layers (portable) and
if_statement to kamae.keras.tensorflow.layers (TF-only).

Decision rationale:

• NumericalIfStatementLayer: Numeric-only, fully portable
• IfStatementLayer: Supports strings, requires TensorFlow backend

NumericalIfStatementLayer (portable):

• Conditional element-wise selection for numeric tensors only
• Uses ops.where for conditional selection
• Uses Python's operator module via get_condition_operator
• Replaced tf.constant with ops.convert_to_tensor
• Only supports numeric dtypes: bfloat16, float16, float32, float64
• Removed deprecation TODO (serves different purpose than IfStatementLayer)
• Works on TensorFlow, JAX, and PyTorch

IfStatementLayer (TF-only):

• Conditional element-wise selection for any dtype including strings
• Supports string comparisons (eq, neq) and numeric comparisons (all operators)
• Inherits from TfBaseLayer with updated imports
• Keeps all TensorFlow operations (tf.where, tf.constant, dtype checks)
• Requires TensorFlow backend for string operations

Both layers support:

• Constants or tensor inputs for value_to_compare, result_if_true, result_if_false
• Six comparison operators: eq, neq, lt, leq, gt, geq
• Dynamic input construction pattern

• fix: Add check in base layer for string inputs

• Some layers can accept any type. These will be created as multi-backend layers but must fail for string inputs if the backend is not tensorflow

• feat: add multi-backend array operation layers

Migrate 4 array operation layers from kamae.tensorflow.layers to portable
kamae.keras.core.layers with backend-agnostic operations.

ArrayConcatenateLayer (portable):

• Concatenates multiple input tensors along specified axis
• Supports auto_broadcast feature to match tensor ranks before concatenation
• Uses ops.concatenate, ops.shape, ops.broadcast_to, ops.stack, ops.max
• compatible_dtypes = None (accepts any backend-supported dtype)
• Key change: tf.reduce_max(list) → ops.max(ops.stack(list))

ArraySplitLayer (portable):

• Splits single tensor into list of tensors along specified axis
• Expands dimensions to preserve shape consistency
• Uses ops.unstack, ops.expand_dims
• compatible_dtypes = None (accepts any backend-supported dtype)
• Direct 1:1 operation replacement

ArrayCropLayer (portable):

• Crops or pads tensor final dimension to fixed length
• Uses ops.minimum, ops.maximum, ops.pad, ops.reshape
• compatible_dtypes = None (accepts any backend-supported dtype)
• Key changes:
  • inputs_shape.shape[0] → len(inputs.shape) for rank calculation
  • Added static vs dynamic shape handling for efficiency
  • Build reshape target using mix of static/dynamic dimensions

ArraySubtractMinimumLayer (portable):

• Computes difference from minimum non-padded value along axis
• Supports optional pad_value to exclude from minimum calculation
• Uses ops.min, ops.subtract, ops.expand_dims, ops.where, ops.equal
• compatible_dtypes = explicit numeric list
• Key change: inputs.dtype.max → numpy.finfo/iinfo portable introspection

Supporting changes:

Created portable shape_utils.py:

• New module: kamae/keras/core/utils/shape_utils.py
• Added reshape_to_equal_rank() function as portable equivalent
• Uses ops.concatenate, ops.shape, ops.ones, ops.reshape

All changes are mechanical API replacements:

• tensorflow as tf → keras, from keras import ops
• @tf.keras.utils.register_keras_serializable → @keras.saving.register_keras_serializable
• kamae.tensorflow.* → kamae.keras.core.*
• tf.operation → ops.operation
• List[tf.dtypes.DType] → List[str]
• Zero algorithmic changes, only API-level conversions

...

v2.40.1

v2.40.1 (2026-06-02)

Fix

• fix: standard scaling do not replace 0 variance with epsilon (#53)

• fix: in standard scaling do not replace 0 variance with epsilon - output 0 - same as in spark.

• chore: add tests for conditional standard scaler as well

---

Co-authored-by: Marian Andrecki <t-mandreki@expediagroup.com> (66b4bd6)

v2.40.0

v2.40.0 (2026-05-06)

Documentation

• docs: README restructured (#39)

• Installation moved further up, above the massive table
• Quick start shows a quick example
• Sklearn removed as a proper usage pattern
• Reduced size by ~40%
• Much more readable (28c70ba)

Feature

• feat: pairwise sim and array reduce max (#44)

• adding modules for pairwise similarity

• tests for pairwise similarity

• adding the new modules on README and tests of serialisation

• formatting issues

• fix header (579612b)

v2.39.1

v2.39.1 (2026-04-14)

Fix

• fix: add Python 3.12 support with version-specific pandas constraints (#38)

• chore: Update Python 3.12 CI

• chore: uv add setuptools

• chore: restrict setuptools version

• chore: revert to pip

• chore: move to run

• chore: add with lt bound

• chore: change build system deps

• chore: fix syntax

• chore: extra deps

• chore: constraint deps

• chore: try --with

• chore: try --with somewhere else

• chore: try add one last time...

• chore: try add one last time again...

• fix: fix python 3.12 pandas incompatibility

• chore: drop unneeded line

• chore: try expanding pandas up to <3.0.0

---

Co-authored-by: James Shinner <jshinner@expediagroup.com>
Co-authored-by: Marian Andrecki <t-mandreki@expediagroup.com> (627be01)

v2.39.0

v2.39.0 (2026-03-26)

Feature

• feat: Add sampling mixin to expensive preprocessing fit functions (#37)

• Add mixin to scaling

• Move to mixin consumption

• Update standard_scale.py for mixin

• Update standard_scale.py for mixin

• Add sample fraction init

• Move to generic import

• Standard scale clean

• Add sampling to min max

• Update conditional_standard_scale to sample

• Update single_feature_array_standard_scale to sample

• Update impute to sample

• Update default values for standard scale

• Missing comma

• Update initial values for sampling

• Updated initial values

• Update initial values for min max

• Linting

• linting

• linting

• Added initial values

• Add test case for standard scale

• Add test cases for sampling impute

• Add test cases for min max scaling

• Test standard scale

• Add test case for single feat sampling

• Add warning

• Move to hasParam

• Made unable to handle 0 and 1 to ensure sampling

• Update conditional_standard_scale.py

• Update impute.py

• Update min_max_scale.py

• Update single_feature_array_standard_scale.py

• Update standard_scale.py

• Update test_conditional_standard_scale.py

• Update test_impute.py

• Update test_min_max_scale.py

• Update test_single_feature_array_standard_scale.py

• Update test_standard_scale.py

• Update min_max_scale.py

• Black base handling

• Black impute handling

• Black min max scale

• Black single feat

• Black standard scale

• Update for better setting

• Bump for larger sampling

• Bump for larger scaling

• Bump for larger scaling

• Bump for larger scaling

• Bump for larger scaling

• Bump for larger scaling

• Fix for black (1670c9a)

v2.38.1

v2.38.1 (2026-01-26)

Chore

• chore: Fix examples (#32)

• Since moving to keras 3 by default in the library, most of the examples are broken.
• Use dict not tf typespecs and if we are in keras 3 we add .keras to the file path (0004cc1)

Documentation

• docs: Use proper Sphinx docs (#31)

• Set the docstring style to Sphinx, this gives us proper tables for func params
• Removed/renamed param docstrings that needed it
• Ensured we don't have any malformed :param blocks (6a1cfe4)

Fix

• fix: tf listwise segmented ops (#35)

Co-authored-by: danzamora <danzamora@expediagroup.com> (b047638)

v2.38.0

v2.38.0 (2025-10-13)

Feature

• feat: Adding segmented mean/min/max in TF and Pypark (#20)

• Adding segmented mean/min/max in TF and Pypark

• Delete output

• reduce line length

• fix: Listwise statistic transforms don't support integers

If you try and use integers with these layers they error due to the float(nan) and is_finite checks. Therefore we remove support for int here so we get a better error message.

• Add segmentation to existing listwise ops

• fix test

• remove unwanted model file

• remove repitition, remove casting

• adding typing and clean up mean

• handle edge case, remove redundant test with ints

• Update doc string for segmented op fn

• typo in some docstrings

• Remove repetition from Spark side

• Typo in doc string

• Fix wrong types, add type hint to segment function, fix examples.

• Correct some doc string issues

• remove comemnt from example

---

Co-authored-by: Andrew Woods <anwoods@expediagroup.com>
Co-authored-by: George Barrowclough <george.d.b@hotmail.com> (35ad82b)

Refactor

• refactor: Add type hints everywhere and force type hints going forward (#27)

• docs: Add type hints everywhere and force type hints going forward

• Add flake8 config to ensure type hints are enforced via linting.

• docs: Remove unneeded self typehint

• tests: Add more layer serialisation tests (#30)

• fix: add serialisation wrapper to OneHotLayer alias

• chore: Update typehints for list rank

• chore: Fix linting (80d0bc7)

v2.37.0

v2.37.0 (2025-08-28)

Feature

• feat: Add rank transformer (#26)

• feat: Add transformer and layer

• refactor: Restructure listwise params

• refactor: Formatting

• Add tests

• docs: Update README

• fix: Add top-level imports

• fix: Add top-level imports

• fix: Update tests

• fix: Update tests

• feat: Add test cases

• fix: Update tests

• feat: Add serialisation test

• fix: Add default layer name

• docs: Tidy up docstrings

• feat: Add sort order option

• fix: Make desc default

• docs: Tidy up type hints/defaults

• test: Add test cases for more axes

• fix: Fix type hint

• fix: Add sort order to layer config

• tests: Expand serialisation test

• fix: Remove unused methods

---

Co-authored-by: James Shinner <jshinner@expediagroup.com> (a8af8dc)

v2.36.0

v2.36.0 (2025-08-18)

Feature

• feat: Add MinHashIndex transform & layer (#25)

• feat: Add MinHashIndex transform & layer

Adds a transformer that takes an array of strings and returns a integer bit representation using the MinHash algorithm: https://en.wikipedia.org/wiki/MinHash. This can be used to approximate Jaccard similarity between sets.

• docs: Update readme and fix linting

• refactor: Add check on numPermutations and docstring changes

• fix: add default to num permutations (f83c51d)

v2.35.0

v2.35.0 (2025-08-14)

Documentation

• docs: Update PULL_REQUEST_TEMPLATE.md (#23) (510e70f)

Feature

• feat: Add MinMaxScale estimator, transformer & layer (#21)

• feat: Add MinMaxScale estimator, transformer & layer

Adds a min max scaling op in similar vein to the standard scaler

• docs: Add missing warnings and docstrings

• refactor: Align subtract calls

• tests: Add tests for None min/max values

• chore: Align both to math.

• docs: Improve docstrings and typos (0aebfd4)

Unknown

• tests: Remove show commands in tests (#24) (8005828)