v0.2.1

What's Changed

• Fix weights sharing for Dense with kernel being an unsettable property by @nhuet in #141

Full Changelog: v0.2.0...v0.2.1

v0.2.0

What's Changed

• Improve release process by @nhuet in #121
• Squash doc into a single commit on gh-pages by @nhuet in #133
• Migration from Keras2 to Keras3 by @ducoffeM in #138
• Fix dependencies conflict between tensorflow and decomon in binder env by @nhuet in #139
• Fix interactive plots on colab by @nhuet in #140

Full Changelog: v0.1.1...v0.2.0

v0.1.1

What's Changed

• Update url to tringer binder build on gesis server by @nhuet in #112
• Fix tutorials by @nhuet in #113
• Keep direct call to get_upper_box() in backward_add() by @nhuet in #114

Full Changelog: v0.1.0...v0.1.1

v0.1.0

What's Changed

• Update pyproject.toml to switch from poetry to setuptools by @nhuet in #5
• Infer version number from git tags thanks to setuptools_scm by @nhuet in #6
• Use pre-commit for linters by @nhuet in #7
• Update test workflow by @nhuet in #8
• Fix pre-commit settings for tutorial notebooks by @nhuet in #9
• Rename test files to be found by pytest and stay consistent by @nhuet in #10
• Add a test importing each submodule of decomon by @nhuet in #11
• Make ortools dependency optional by @nhuet in #12
• Update python badge from readme with >=3.7 by @nhuet in #4
• Fix default arguments that were mutable by @nhuet in #16
• Clean code with changes proposed by code.review.doctor by @nhuet in #17
• Relax condition on tensorflow dependency by @nhuet in #13
• Fix pycharm inspections by @nhuet in #18
• Launch test suite also on windows and macos by @nhuet in #21
• Clean comments and prints by @nhuet in #19
• Clean imports by @nhuet in #20
• Move source code in src/ subdirectory by @nhuet in #23
• Apply pytest best practices by @nhuet in #22
• Create a documentation by @nhuet in #24
• Generate and deploy dev doc via github actions by @nhuet in #25
• Update publish workflow by @nhuet in #26
• Add tutorials on colab by @nhuet in #27
• Test code blocks from doc during test suite by @nhuet in #28
• Add coverage in test suite by @nhuet in #29
• Remove fully commented tests by @nhuet in #30
• Remove numpy and gradient subpackages by @nhuet in #31
• Remove insecable spaces in markdown files by @nhuet in #33
• Use x.dtype instead of K.floatx() in K.cast() by @nhuet in #34
• Remove a try...except that was there for debugging purpose by @nhuet in #35
• Bump isort pre-commit hook by @nhuet in #37
• CROWN optimisation time by @ducoffeM in #41
• Lighten parametrization of tests by @nhuet in #36
• Add a binder environment to launch notebooks on binderhub by @nhuet in #38
• Pass linters on all files after bumping pre-commit hooks by @nhuet in #39
• Pass linters on all files by @nhuet in #42
• Make README point to online documentation by @nhuet in #40
• Update license to MIT by @nhuet in #43
• Use generic email address in project metadata by @nhuet in #44
• Remove deprecated DecomonSequential models by @nhuet in #45
• Unify decomon syntax throughout the project by @nhuet in #46
• Remove unused fast argument by @nhuet in #47
• Replace np.int64 by np.int_ by @nhuet in #48
• Update notebooks by @nhuet in #49
• Remove support for python 3.7 by @nhuet in #50
• Enable visualization of a decomon model graph with tensorboard by @nhuet in #52
• Upgrade python version to 3.8 for binder environment by @nhuet in #53
• Fix minor issues in tests by @nhuet in #57
• Remove deprecated code by @nhuet in #58
• Rename to_monotonic() as to_decomon() by @nhuet in #61
• Annotate and fix mypy issues in high-level functions by @nhuet in #59
• new notebook on decomon for overestimation by @ducoffeM in #54
• Remove code not ready for official library by @nhuet in #64
• Improve "tutorials" page of online doc by @nhuet in #62
• Fix mypy issues by @nhuet in #63
• Add a static type checking step in github workflow by @nhuet in #68
• Remove previous attribute from backward layers by @nhuet in #67
• Add slope argument to all activations, layers, convert and wrapping functions by @nhuet in #69
• Use "affine" instead of "forward" in code for bounds type by @nhuet in #74
• Rename DecomonModel main args "inputs" and "outputs" by @nhuet in #72
• Add information about mypy in contributing doc page by @nhuet in #71
• Cleaning existing notebooks and add a new header by @ducoffeM in #55
• Pooling hull by @ducoffeM in #65
• Add tests for slope argument by @nhuet in #73
• Convert layers embedding activation into layer without activation + activation layer by @nhuet in #75
• Prefix the name of backward layer by the name of the original layer + "_" by @nhuet in #79
• Remove duplicate code initializing input_tensors for convert/clone by @nhuet in #77
• Refactor to_decomon(), to_decomon_merge(), and to_backward() by @nhuet in #80
• Convert deel-lip layers to keras layers before converting to decomon model by @nhuet in #81
• Use pytest.skip when skipping a test because some condition is not met by @nhuet in #82
• Fix convex_domain use by converting fields to proper enumeration type by @nhuet in #83
• Update tutos for last code version by @nhuet in #84
• Fix issues in decomon layers build() by @nhuet in #88
• Fix reset_layer() in decomon layers by @nhuet in #87
• Simplify to_decomon() under the hypothesis "deel-lip layers are already converted" by @nhuet in #85
• Remove references to self.linear_layer and self.init_layer by @nhuet in #86
• Fix typo on subtract by @nhuet in #89
• Remove finetune parameters for DecomonDense and DecomonConv2D by @nhuet in #90
• Fix minor issues in notebooks texts by @nhuet in #91
• Implement DecomonBatchNormalization.call with center or scale = False by @nhuet in #92
• Fix inputs to use for decomon models in tests by @nhuet in #96
• Improve and simplify tests by @nhuet in #93
• Relax required precision for numpy asserts in tests by @nhuet in #94
• Add tests for clone on more toy models by @nhuet in #95
• Make variable/arguments names consistent across the library by @nhuet in #97
• Replace convex domain dictionariess by a specific class PerturbationDomain and subclasses according to its type by @nhuet in #98
• Fix issue in get_linear_softplus_hull() when finetuning by @nhuet in #99
• Fix BackwardBatchNormalization.call() by @nhuet in #100
• backward relaxation of convolution with toeplitz expression by @ducoffeM in #102
• Pooling collapse by @ducoffeM in #103
• Stop triggering binder build on gke by @nhuet in #104
• Create InputsOutputsSpec class to manage inputs and outputs of deomon layers and functions by @nhuet in #105
• Add methods get_lower/get_upper/get_x_input_shape to PerturbationDomain by @nhuet in #106
• Robustify tutorials page generation and fix minor issues in tutorials by @nhuet in #107
• Adapt code to new version of Keras (>= 2.13) by @nhuet in #108
• Adapt tutorials to newer version of code by @nhuet in #109
• Remove unused constant by @nhuet in #101
• Pass linters by @nhuet in #110
• Force using tf<2.13...