Sourced from nltk's\r\nchangelog.
\r\n\r\n\r\nVersion 3.9.1 2024-08-19
\r\n\r\n
\r\n- Fixed bug that prevented wordnet from loading
\r\nVersion 3.9 2024-08-18
\r\n\r\n
\r\n- Avoid need for pickled models, resolves security vulnerability\r\nCVE-2024-39705
\r\n- No longer sort WordNet synsets and relations (sort in calling\r\nfunction when required)
\r\n- Add Python 3.12 support
\r\n- Many other minor fixes
\r\nThanks to the following contributors to 3.8.2:\r\nTom Aarsen, Cat Lee Ball, Veralara Bernhard, Carlos Brandt, Konstantin\r\nChernyshev, Michael Higgins,\r\nEric Kafe, Vivek Kalyan, David Lukes, Rob Malouf, purificant, Alex\r\nRudnick, Liling Tan, Akihiro Yamazaki.
\r\nVersion 3.8.1 2023-01-02
\r\n\r\n
\r\n- Resolve RCE vulnerability in localhost WordNet Browser (#3100)
\r\n- Remove unused tool scripts (#3099)
\r\n- Resolve XSS vulnerability in localhost WordNet Browser (#3096)
\r\n- Add Python 3.11 support (#3090)
\r\nThanks to the following contributors to 3.8.1:\r\nFrancis Bond, John Vandenberg, Tom Aarsen
\r\nVersion 3.8 2022-12-12
\r\n\r\n
\r\n\r\n- Refactor dispersion plot (#3082)
\r\n- Provide type hints for LazyCorpusLoader variables (#3081)
\r\n- Throw warning when LanguageModel is initialized with incorrect\r\nvocabulary (#3080)
\r\n- Fix WordNet's all_synsets() function (#3078)
\r\n- Resolve TreebankWordDetokenizer inconsistency with end-of-string\r\ncontractions (#3070)
\r\n- Support both iso639-3 codes and BCP-47 language tags (#3060)
\r\n- Avoid DeprecationWarning in Regexp tokenizer (#3055)
\r\n- Fix many doctests, add doctests to CI (#3054, #3050, #3048)
\r\n- Fix bool field not being read in VerbNet (#3044)
\r\n- Greatly improve time efficiency of SyllableTokenizer when tokenizing\r\nnumbers (#3042)
\r\n- Fix encodings of Polish udhr corpus reader (#3038)
\r\n- Allow TweetTokenizer to tokenize emoji flag sequences (#3034)
\r\n- Prevent LazyModule from increasing the size of\r\nnltk.dict (#3033)
\r\n- Fix CoreNLPServer non-default port issue (#3031)
\r\n- Add "acion" suffix to the Spanish SnowballStemmer (#3030)
\r\n- Allow loading WordNet without OMW (#3026)
\r\n- Use input() in nltk.chat.chatbot() for Jupyter support (#3022)
\r\n- Fix edit_distance_align() in distance.py (#3017)
\r\n- Tackle performance and accuracy regression of sentence tokenizer\r\nsince NLTK 3.6.6 (#3014)
\r\n- Add the Iota operator to semantic logic (#3010)
\r\n- Resolve critical errors in WordNet app (#3008)
\r\n- Resolve critical error in CHILDES Corpus (#2998)
\r\n- Make WordNet information_content() accept adjective satellites (#2995)
\r\n- Add "strict=True" parameter to CoreNLP (#2993, #3043)
\r\n- Resolve issue with WordNet's synset_from_sense_key (#2988)
\r\n
... (truncated)
\r\n24936a2
\r\nBump version to 3.9c222897
\r\nMerge branch 'develop' of https://github.com/nltk/nltk\r\ninto develop34c3a4a
\r\nMerge branch 'develop' of https://github.com/nltk/nltk\r\ninto develop253dd3a
\r\nadd blackc43727f
\r\nUpdate version7137405
\r\nMerge pull request #3066 from\r\nasishm/bugfix-lambda-closure-leak369cb9f
\r\nMerge pull request #3245 from\r\nekaf/hotfix-closuredup501c70e
\r\nMerge branch 'develop' into hotfix-closuredupbf05dc4
\r\nMerge pull request #3306 from\r\nekaf/py3_compat66539c7
\r\nSorted output in unit/test_wordnet.pySourced from tensorflow's\r\nreleases.
\r\n\r\n\r\nTensorFlow 2.12.1
\r\nRelease 2.12.1
\r\nBug Fixes and Other Changes
\r\n\r\n
\r\n- The use of the ambe config to build and test aarch64 is not needed.\r\nThe ambe config will be removed in the future. Making cpu_arm64_pip.sh\r\nand cpu_arm64_nonpip.sh more similar for easier future maintenance.
\r\nTensorFlow 2.12.0
\r\nRelease 2.12.0
\r\nTensorFlow
\r\nBreaking Changes
\r\n\r\n
\r\n- \r\n
\r\nBuild, Compilation and Packaging
\r\n\r\n
\r\n- Removed redundant packages
\r\ntensorflow-gpu
and\r\ntf-nightly-gpu
. These packages were removed and replaced\r\nwith packages that direct users to switch totensorflow
or\r\ntf-nightly
respectively. Since TensorFlow 2.1, the only\r\ndifference between these two sets of packages was their names, so there\r\nis no loss of functionality or GPU support. See https://pypi.org/project/tensorflow-gpu\r\nfor more details.- \r\n
\r\n\r\n
tf.function
:\r\n
\r\n- \r\n
tf.function
now uses the Python inspect library\r\ndirectly for parsing the signature of the Python function it is\r\ndecorated on. This change may break code where the function signature is\r\nmalformed, but was ignored previously, such as:\r\n\r\n
\r\n- Using
\r\nfunctools.wraps
on a function with different\r\nsignature- Using
\r\nfunctools.partial
with an invalid\r\ntf.function
input- \r\n
tf.function
now enforces input parameter names to be\r\nvalid Python identifiers. Incompatible names are automatically sanitized\r\nsimilarly to existing SavedModel signature behavior.- Parameterless
\r\ntf.function
s are assumed to have an empty\r\ninput_signature
instead of an undefined one even if the\r\ninput_signature
is unspecified.- \r\n
tf.types.experimental.TraceType
now requires an\r\nadditionalplaceholder_value
method to be defined.- \r\n
tf.function
now traces with placeholder values\r\ngenerated by TraceType instead of the value itself.- \r\n
\r\nExperimental APIs\r\n
\r\ntf.config.experimental.enable_mlir_graph_optimization
and\r\ntf.config.experimental.disable_mlir_graph_optimization
were\r\nremoved.Major Features and Improvements
\r\n\r\n
\r\n\r\n- \r\n
\r\nSupport for Python 3.11 has been added.
\r\n- \r\n
\r\nSupport for Python 3.7 has been removed. We are not releasing any\r\nmore patches for Python 3.7.
\r\n- \r\n
\r\n\r\n
tf.lite
:\r\n
\r\n- Add 16-bit float type support for built-in op\r\n
\r\nfill
.- Transpose now supports 6D tensors.
\r\n- Float LSTM now supports diagonal recurrent tensors: https://arxiv.org/abs/1903.08023
\r\n- \r\n
\r\n\r\n
tf.experimental.dtensor
:\r\n
\r\n- Coordination service now works with\r\n
\r\ndtensor.initialize_accelerator_system
, and enabled by\r\ndefault.- Add
\r\ntf.experimental.dtensor.is_dtensor
to check if a\r\ntensor is a DTensor instance.- \r\n
\r\n\r\n
tf.data
:\r\n
\r\n- Added support for alternative checkpointing protocol which makes it\r\npossible to checkpoint the state of the input pipeline without having to\r\nstore the contents of internal buffers. The new functionality can be\r\nenabled through the
\r\nexperimental_symbolic_checkpoint
option\r\noftf.data.Options()
.- Added a new
\r\nrerandomize_each_iteration
argument for the\r\ntf.data.Dataset.random()
operation, which controls whether\r\nthe sequence of generated random numbers should be re-randomized every\r\nepoch or not (the default behavior). Ifseed
is set and\r\nrerandomize_each_iteration=True
, therandom()
\r\noperation will produce a different (deterministic) sequence of numbers\r\nevery epoch.
... (truncated)
\r\nSourced from tensorflow's\r\nchangelog.
\r\n\r\n\r\nRelease 2.12.1
\r\nBug Fixes and Other Changes
\r\n\r\n
\r\n- The use of the ambe config to build and test aarch64 is not needed.\r\nThe ambe config will be removed in the future. Making cpu_arm64_pip.sh\r\nand cpu_arm64_nonpip.sh more similar for easier future maintenance.
\r\nRelease 2.12.0
\r\nBreaking Changes
\r\n\r\n
\r\n- \r\n
\r\nBuild, Compilation and Packaging
\r\n\r\n
\r\n- Removed redundant packages
\r\ntensorflow-gpu
and\r\ntf-nightly-gpu
. These packages were removed and replaced\r\nwith packages that direct users to switch totensorflow
or\r\ntf-nightly
respectively. Since TensorFlow 2.1, the only\r\ndifference between these two sets of packages was their names, so there\r\nis no loss of functionality or GPU support. See https://pypi.org/project/tensorflow-gpu\r\nfor more details.- \r\n
\r\n\r\n
tf.function
:\r\n
\r\n- \r\n
tf.function
now uses the Python inspect library\r\ndirectly for parsing the signature of the Python function it is\r\ndecorated on. This change may break code where the function signature is\r\nmalformed, but was ignored previously, such as:\r\n\r\n
\r\n- Using
\r\nfunctools.wraps
on a function with different\r\nsignature- Using
\r\nfunctools.partial
with an invalid\r\ntf.function
input- \r\n
tf.function
now enforces input parameter names to be\r\nvalid Python identifiers. Incompatible names are automatically sanitized\r\nsimilarly to existing SavedModel signature behavior.- Parameterless
\r\ntf.function
s are assumed to have an empty\r\ninput_signature
instead of an undefined one even if the\r\ninput_signature
is unspecified.- \r\n
tf.types.experimental.TraceType
now requires an\r\nadditionalplaceholder_value
method to be defined.- \r\n
tf.function
now traces with placeholder values\r\ngenerated by TraceType instead of the value itself.- \r\n
\r\nExperimental APIs\r\n
\r\ntf.config.experimental.enable_mlir_graph_optimization
and\r\ntf.config.experimental.disable_mlir_graph_optimization
were\r\nremoved.Major Features and Improvements
\r\n\r\n
\r\n\r\n- \r\n
\r\nSupport for Python 3.11 has been added.
\r\n- \r\n
\r\nSupport for Python 3.7 has been removed. We are not releasing any\r\nmore patches for Python 3.7.
\r\n- \r\n
\r\n\r\n
tf.lite
:\r\n
\r\n- Add 16-bit float type support for built-in op\r\n
\r\nfill
.- Transpose now supports 6D tensors.
\r\n- Float LSTM now supports diagonal recurrent tensors: https://arxiv.org/abs/1903.08023
\r\n- \r\n
\r\n\r\n
tf.experimental.dtensor
:\r\n
\r\n- Coordination service now works with\r\n
\r\ndtensor.initialize_accelerator_system
, and enabled by\r\ndefault.- Add
\r\ntf.experimental.dtensor.is_dtensor
to check if a\r\ntensor is a DTensor instance.- \r\n
\r\n\r\n
tf.data
:\r\n
\r\n- Added support for alternative checkpointing protocol which makes it\r\npossible to checkpoint the state of the input pipeline without having to\r\nstore the contents of internal buffers. The new functionality can be\r\nenabled through the
\r\nexperimental_symbolic_checkpoint
option\r\noftf.data.Options()
.- Added a new
\r\nrerandomize_each_iteration
argument for the\r\ntf.data.Dataset.random()
operation, which controls whether\r\nthe sequence of generated random numbers should be re-randomized every\r\nepoch or not (the default behavior). Ifseed
is set and\r\nrerandomize_each_iteration=True
, therandom()
\r\noperation will produce a different (deterministic) sequence of numbers\r\nevery epoch.- Added a new
\r\nrerandomize_each_iteration
argument for the\r\ntf.data.Dataset.sample_from_datasets()
operation, which\r\ncontrols whether the sequence of generated random numbers used for\r\nsampling should be re-randomized every epoch or not. If\r\nseed
is set and\r\nrerandomize_each_iteration=True
, the\r\nsample_from_datasets()
operation will use a different\r\n(deterministic) sequence of numbers every epoch.- \r\n
\r\n\r\n
tf.test
:
... (truncated)
\r\n8e2b665
\r\nMerge pull request #61094\r\nfrom tensorflow/venkat-patch-44402478f0
\r\nFix unit test failure caused by numpy update2cd9b41
\r\nMerge pull request #61082\r\nfrom tensorflow/venkat-patch-3337995c95
\r\nUpdating Simplified retry logic to DNS cache29479ed
\r\nMerge pull request #60872\r\nfrom tensorflow/r2.12-c45a6c0b1cbe76a933
\r\nSimplified retry logic to DNS cache76addf7
\r\nMerge pull request #60850\r\nfrom elfringham/non_pip_fix05987a8
\r\n[Linaro:ARM_CI] Fix permissions for running nonpip tests23724d2
\r\nMerge pull request #60842\r\nfrom elfringham/r2.12496730b
\r\nLimit typing_extensions to less than 4.6.0 until it worksSourced from keras's\r\nreleases.
\r\n\r\n\r\nKeras Release 2.13.1
\r\nWhat's Changed
\r\n\r\n
\r\n\r\n- Fix timeseries_dataset_from_array counts when sequence_stride > 1\r\nby
\r\n@​basjacobs93
in\r\nkeras-team/keras#17396- Fix doc for MultiHeadAttention's output_shape arg by
\r\n@​tsdeng
in keras-team/keras#17432- Added a link to the TextVectorization API Reference Page. by
\r\n@​Neeshamraghav012
\r\nin keras-team/keras#17540- Fix ConvNeXt classifier activation bug by
\r\n@​Frightera
in keras-team/keras#17544- Activations clean up by
\r\n@​joesho112358
\r\nin keras-team/keras#17545- Perform all ops in apply_gradient in a given tf.name_scope. by
\r\n@​foxik
in keras-team/keras#17550- Bugfix/normalization invert by
\r\n@​jerabaul29
in keras-team/keras#17501- OptimizerV2: Allow deepcopy by
\r\n@​jschueller
in keras-team/keras#17463- keras.models.clone_model creates different weight names for RNN\r\nlayers by
\r\n@​mergian
in keras-team/keras#17465- Fix spelling error in Initializer warning by
\r\n@​edknv
in keras-team/keras#17568- LayerScale fix to support pretrained ConvNeXt models with mixed\r\nprecision by
\r\n@​andreped
in keras-team/keras#17547- Softmax fix in Mixed Precision by
\r\n@​dsr-lab
in keras-team/keras#17571- Less restrictive ROCm+GRU/LSTM fallback logic by
\r\n@​ekuznetsov139
\r\nin keras-team/keras#17587- Update security section. by
\r\n@​sachinprasadhs
\r\nin keras-team/keras#17522- Fix serialization error with EfficientNet by
\r\n@​RocketDelivery
\r\nin keras-team/keras#17498- grammar corrections by
\r\n@​egurnick
in keras-team/keras#17595- Update documentation to keras reuters dataset by
\r\n@​hxy9243
in keras-team/keras#17635- Add Lion optimizer by
\r\n@​drauh
in keras-team/keras#17605- Correct the docstring of Head func in convnext.py by
\r\n@​Frightera
in keras-team/keras#17667- keras.layers.Bidirectional is missing argument "states",\r\nwhich is defined for keras.layers.RNN by
\r\n@​mergian
in keras-team/keras#17469- fix a small bug in tf.keras.layer.Layer.compute_output_signature by\r\n
\r\n@​yamanoko
in keras-team/keras#17636- Optimize mixed-precision finite check for sparse tensors. by
\r\n@​nluehr
in keras-team/keras#17638- Typo Lion docstring by
\r\n@​drauh
in keras-team/keras#17726- Raise ValueError if reducing axis dimension is 0 by
\r\n@​old-school-kid
\r\nin keras-team/keras#16755- Add CategoricalFocalCrossentropy to Losses API by
\r\n@​Frightera
in keras-team/keras#17651- Unifying interpolation selection by
\r\n@​jpodivin
in keras-team/keras#17742- Add a missing space by
\r\n@​sudoLife
in keras-team/keras#17744- [keras/initializers/initializers_v1.py] Standardise docstring usage\r\nof "Default to" by
\r\n@​SamuelMarks
in\r\nkeras-team/keras#17785- [keras/layers/convolutional/conv3d_transpose.py] Standardise\r\ndocstring usage of "Default to" by
\r\n@​SamuelMarks
in\r\nkeras-team/keras#17797- [keras/engine/training.py] Standardise docstring usage of\r\n"Default to" by
\r\n@​SamuelMarks
in\r\nkeras-team/keras#17780- [keras/feature_column/dense_features.py] Standardise docstring usage\r\nof "Default to" by
\r\n@​SamuelMarks
in\r\nkeras-team/keras#17783- [keras/datasets/imdb.py] Standardise docstring usage of\r\n"Default to" by
\r\n@​SamuelMarks
in\r\nkeras-team/keras#17771- [keras/backend.py] Standardise docstring usage of "Default\r\nto" by
\r\n@​SamuelMarks
in\r\nkeras-team/keras#17769- [keras/engine/base_layer_utils.py] Standardise docstring usage of\r\n"Default to" by
\r\n@​SamuelMarks
in\r\nkeras-team/keras#17774- [keras/layers/activation/leaky_relu.py] Standardise docstring usage\r\nof "Default to" by
\r\n@​SamuelMarks
in\r\nkeras-team/keras#17786- [keras/engine/training_v1.py] Standardise docstring usage of\r\n"Default to" by
\r\n@​SamuelMarks
in\r\nkeras-team/keras#17781- [keras/layers/activation/relu.py] Standardise docstring usage of\r\n"Default to" by
\r\n@​SamuelMarks
in\r\nkeras-team/keras#17787- [keras/layers/convolutional/conv2d_transpose.py] Standardise\r\ndocstring usage of "Default to" by
\r\n@​SamuelMarks
in\r\nkeras-team/keras#17795- [keras/layers/convolutional/conv3d.py] Standardise docstring usage\r\nof "Default to" by
\r\n@​SamuelMarks
in\r\nkeras-team/keras#17796- [keras/layers/activation/softmax.py] Standardise docstring usage of\r\n"Default to" by
\r\n@​SamuelMarks
in\r\nkeras-team/keras#17788- [keras/layers/convolutional/depthwise_conv2d.py] Standardise\r\ndocstring usage of "Default to" by
\r\n@​SamuelMarks
in\r\nkeras-team/keras#17799- [keras/layers/convolutional/separable_conv2d.py] Standardise\r\ndocstring usage of "Default to" by
\r\n@​SamuelMarks
in\r\nkeras-team/keras#17800- [keras/layers/locally_connected/locally_connected2d.py] Standardise\r\ndocstring usage of "Default to" by
\r\n@​SamuelMarks
in\r\nkeras-team/keras#17803- [keras/layers/locally_connected/locally_connected1d.py] Standardise\r\ndocstring usage of "Default to" by
\r\n@​SamuelMarks
in\r\nkeras-team/keras#17802- Fix performance issue with sample weights in model.fit() by
\r\n@​nershman
in keras-team/keras#17357- [keras/datasets/reuters.py] Standardise docstring usage of\r\n"Default to" by
\r\n@​SamuelMarks
in\r\nkeras-team/keras#17772- [keras/applications/mobilenet.py] Standardise docstring usage of\r\n"Default to" by
\r\n@​SamuelMarks
in\r\nkeras-team/keras#17763- [keras/applications/inception_v3.py] Standardise docstring usage of\r\n"Default to" by
\r\n@​SamuelMarks
in\r\nkeras-team/keras#17762
... (truncated)
\r\nb3ffea6
\r\nCherrypick Sequential serialization bug fix for r2.13 (#18258)87db506
\r\nCherrypick the release script fix for RC. (#18082)a51c89a
\r\nIncrease the version number for keras 2.13 (#18081)861ad74
\r\nAdds error for serializing metric using layer serialization.1b7c53d
\r\nAdds Keras v3 saving testing coverage to Keras layers tests.e7c4d09
\r\nExpands Keras internal testing coverage for the new v3 saving format for\r\ncomm...d72829a
\r\nChange references from distribution_strategy_context.py
to\r\n`distribute_lib....605b2d7
\r\nMerge pull request #17961\r\nfrom SamuelMarks:keras.layers.activation-defaults-toa64d0b7
\r\nMerge pull request #17955\r\nfrom SamuelMarks:keras.datasets-defaults-tocb1e1a0
\r\nMerge pull request #17967\r\nfrom SamuelMarks:keras.layers.preprocessing-default...