v0.150.0-BERT-Model

What's Changed

• Improve RaggedTensor by @Beacontownfc in #1186
• feat: add the implementation of sample_weight in model.fit by @Wanglongzhi2001 in #1187
• Allow Model to cache weights. by @hchen2020 in #1188
• feat: add the implementation of class_weight in model.fit by @Wanglongzhi2001 in #1189
• 解决keras模式下，使用GPU训练时会爆显存的bug。 by @dogvane in #1190
• fix: Implemented support for loading models with Concatenate layers by @Jucko13 in #1192
• add relu6 by @dogvane in #1193
• fix: Saving a loaded model by @GaijinOtohp in #1201
• fix: add the implementation of the tile's and GatherND's grad and add OptionalArgs by @Wanglongzhi2001 in #1202
• fix: fix the bug of boolean_mask by @Wanglongzhi2001 in #1205
• test: add the concat_v2 test by @Wanglongzhi2001 in #1207
• fix: fix the bug caused by concat_v2 by @Wanglongzhi2001 in #1208
• refactor: refacter the place of the kernel unittest folder by @Wanglongzhi2001 in #1209

New Contributors

• @hchen2020 made their first contribution in #1188
• @Jucko13 made their first contribution in #1192
• @GaijinOtohp made their first contribution in #1201

Full Changelog: v0.110.4-Transformer-Model...v0.150.0-BERT-Model

v0.110.4-Transformer-Model

What's Changed

• fix: revise earlystopping callback's min_delta parameter and add GRUCell layer by @Wanglongzhi2001 in #1120
• fix: fix the bug of repeated progress bar in Model.fit() by @Wanglongzhi2001 in #1123
• fix: A bug fix in KerasObjectLoader.cs by @RachamimYaakobov in #1124
• fix: inconsistent shape error while training embedding layer by @lingbai-kong in #1125
• Add pad preprocessing for imdb dataset by @lingbai-kong in #1126
• fix: optimize some APIs by @Beacontownfc in #1129
• fix: add EinsumGrad by @lingbai-kong in #1132
• Add a new optimizer by @Beacontownfc in #1136
• fix: Fix the problem where the model.save_weights() can't save multi-layer weight by @Beacontownfc in #1137
• fix: exp moved to tf.math.cs by @BalashovK in #1139
• fix same bug by @dogvane in #1140
• fix flip_left_right run bug by @dogvane in #1143
• fix bug by @dogvane in #1145
• fix: fix the bug of load LSTM model and add test by @Wanglongzhi2001 in #1144
• fix: ignore the LSTMLoad test by @Wanglongzhi2001 in #1146
• fix: make the initialization of the layer's name correct by @Wanglongzhi2001 in #1148
• feat: add Bidirectional layer by @Wanglongzhi2001 in #1149
• Fix: model.load_weights by @Beacontownfc in #1154
• fix: revise np.amin, np.amax and add np.argmin by @Wanglongzhi2001 in #1155
• feat: add adjust_contrast, adjust_hue, combined_non_max_suppression, crop_and_resize image oprs by @Wanglongzhi2001 in #1157
• fix: add the gradient of the tf.gradient opr by @Wanglongzhi2001 in #1159
• fix: add the momentum parameter's implemention of SGD by @Wanglongzhi2001 in #1161
• fix: remove the reflection in the implemention of Bidirectional by @Wanglongzhi2001 in #1162
• fix: partially fix the bug of load_model by @Wanglongzhi2001 in #1164
• fix: validate dataset of Imdb do not load bug & add: custom Imdb path by @lingbai-kong in #1165
• feat: implement GRU layer by @Wanglongzhi2001 in #1168
• add: loading pickled npy file for imdb dataset loader by @lingbai-kong in #1169
• fix: model does not stop on stop_training == true #1170 by @asaf92 in #1171
• cached_session for graph tests by @novikov-alexander in #1172
• Gradient tests which were dependant only on cached_session by @novikov-alexander in #1173
• optimize: temporal complexity of Imdb dataset loader by @lingbai-kong in #1175
• fix: improve unpickler speed with BufferedStream by @lingbai-kong in #1176
• fix: fix EarlyStopping by @Wanglongzhi2001 in #1180
• fix: adjust imdb dataset loader for faster loading speed by @lingbai-kong in #1182

New Contributors

• @RachamimYaakobov made their first contribution in #1124
• @asaf92 made their first contribution in #1171
• @novikov-alexander made their first contribution in #1172

Full Changelog: v0.110.0-LSTM-Model...v0.110.4-Transformer-Model

v0.110.0 LSTM

What's Changed

• fix: predict with multiple outputs by @fryguy1013 in #1065
• refactor: some gen_ops with added code generator. by @AsakusaRinne in #1063
• fix: error when using graph in multi-threads. by @AsakusaRinne in #1068
• docs: add discord info to readme. by @AsakusaRinne in #1069
• docs: update readme file. by @AsakusaRinne in #1070
• docs: add tf.keras badge to readme. by @AsakusaRinne in #1072
• docs: add vote info to readme. by @AsakusaRinne in #1075
• fix: error when set the activation of keras.layers.Conv2DTranspose to null. by @lingbai-kong in #1076
• fix: can't implement len for KerasShapesWrapper & Add bias implement to Conv2DTranspose.Call() by @lingbai-kong in #1078
• fix: status null reference of base session. by @AsakusaRinne in #1079
• build: add scripts to split large native library file. by @AsakusaRinne in #1071
• fix: temporarily fix the sequential nest error. by @AsakusaRinne in #1086
• fix: Update PredictInternational on Model.Predict.cs by @DevNullx64 in #1087
• fix: A new feature for NDArray.max by @RayWang-iat in #1091
• feat: add rnn basic modules by @AsakusaRinne in #1097
• fix: some possible errors of RNN. by @AsakusaRinne in #1098
• Add feature(not completed):add SimpleRNNCell, StackedRNNCell, RNN and test. by @Wanglongzhi2001 in #1100
• feat: support training of RNN. by @AsakusaRinne in #1109
• refactor: Model.eval by @DevNullx64 in #1092
• feat: np update square and dot product by @visagang in #1111
• feat: Support training of RNN and LSTM. by @Wanglongzhi2001 in #1110
• fix: error pf training RNN on Linux by @AsakusaRinne in #1113
• Improve the API of LayerNormalization by @Beacontownfc in #1114
• fix: type converting errors when loading imdb dataset by @lingbai-kong in #1116
• refactor: Standardize TensorFlowNET.Keras/Losses/* by @DevNullx64 in #1089
• feat: Add UpSampling1D layer and test. by @Wanglongzhi2001 in #1117

New Contributors

• @fryguy1013 made their first contribution in #1065
• @lingbai-kong made their first contribution in #1076
• @DevNullx64 made their first contribution in #1087
• @RayWang-iat made their first contribution in #1091
• @visagang made their first contribution in #1111
• @Beacontownfc made their first contribution in #1114

Full Changelog: v0.100.5...v0.110.0-LSTM-Model

v0.100.5-BERT-model

What's Changed

• Optimize the readme files by @AsakusaRinne in #994
• Fix some typo of readme. by @AsakusaRinne in #995
• Add default value of parameter of keras softmax. by @AsakusaRinne in #997
• Support the multiple inputs of keras model.fit. by @AsakusaRinne in #996
• Automatically add KerasInterface to tf by @AsakusaRinne in #998
• Reduce the time of keras unittest. by @AsakusaRinne in #999
• Support mutiple inputs of keras modek.predict. by @AsakusaRinne in #1000
• Refine keras.Activation and add tf.keras.activations. by @AsakusaRinne in #1001
• Add github issue templates. by @AsakusaRinne in #1002
• Fix the error of issue templates. by @AsakusaRinne in #1003
• Fix model.evaluate don't have output by @Wanglongzhi2001 in #1004
• Add EarlyStopping callback by @Wanglongzhi2001 in #1007
• Add a Stop_training property by @Wanglongzhi2001 in #1008
• Add shape deduce for mean square error grad. by @AsakusaRinne in #1010
• Fix validation_split has no output and add validation_data parameter to model.fit by @Wanglongzhi2001 in #1014
• Added: complex, real, imag, angle by @BalashovK in #1011
• Added: fft, ifft, fft2d, ifft2d, fft3d, ifft3d by @BalashovK in #1016
• Fix the bug of non-convergence when use SparseCategoricalCrossentropy by @Wanglongzhi2001 in #1018
• Finish EarlyStopping by @Wanglongzhi2001 in #1020
• Finish EarlyStopping by @Wanglongzhi2001 in #1021
• Fix the error when saving model with GPU. by @AsakusaRinne in #1023
• Partially Support the function loading by @AsakusaRinne in #1022
• Fix the error when loading VGG19. by @AsakusaRinne in #1027
• Fix the error of example caused by #1022 by @AsakusaRinne in #1029
• Fix the error when loading Conv1D layer with initialzier. by @AsakusaRinne in #1031
• Add set_weights and get_weighst APIs by @Wanglongzhi2001 in #1030
• Fix the error of loading model saved before tf2.5. by @AsakusaRinne in #1032
• Change type of BuildInputShape and BatchInputShape by @AsakusaRinne in #1033
• Add Tensorflow.NET.Hub by @AsakusaRinne in #1034
• Change some project information by @AsakusaRinne in #1035
• Fix (#1036) and adjust the keras unittest by @AsakusaRinne in #1039
• Partially fix the error when crop image. by @AsakusaRinne in #1043
• Add the constructor of NDArray which reuses memory by @AsakusaRinne in #1044
• ci: add ci for test and auto-release. by @AsakusaRinne in #1047
• fix: partially fix the error when saving model after loading. by @AsakusaRinne in #1049

New Contributors

• @Wanglongzhi2001 made their first contribution in #1004
• @BalashovK made their first contribution in #1011

Full Changelog: v0.100.4-load-saved-model...v0.100.5-BERT-load

v0.100.4-load-saved-model

What's Changed

• Fix IO Exception from keras.fit. by @AsakusaRinne in #959
• Bump Newtonsoft.Json from 13.0.1 to 13.0.2 in /src/TensorFlowNET.Keras by @dependabot in #963
• Removed use of tf.Status static instance by @Superpiffer in #975
• Add pb model save by @Oceania2018 in #976
• Use a local Status variable by @Superpiffer in #978
• Fix ConvTraspose2D gradient and register rsqrt gradient. by @AsakusaRinne in #980
• Reimplemented NDArray == and != operators, handling null values by @Superpiffer in #982
• Fix the keras.sparse_categorical_crossentropy. (#985) by @AsakusaRinne in #987
• Support loading of SavedModel format by @AsakusaRinne in #989
• Fix issue #760 by @AsakusaRinne in #992
• Align keras.Input with tensorflow python. by @AsakusaRinne in #993

New Contributors

• @AsakusaRinne made their first contribution in #959
• @dependabot made their first contribution in #963
• @Superpiffer made their first contribution in #975

Full Changelog: v0.70.2-NET6...v0.100.4-load-saved-model

v0.70.2-NET 6.0

v0.70.2-NET6

Fix ReadValueMatrix ReadBytes issue in .NET 6.0. #880

v0.70.0-TimeSeries

fix array_ops.squeeze

TensorFlowOpLayer

Fixed TensorFlowOpLayer memory leaking issue.

TensorFlow NumPy

v0.60-tf.numpy

released tf.net v0.60, keras v0.6.

v0.40-tf2.4-tstring

TensorFlow v0.40 only work with TensorFlow native v2.4+ due to breaking change of TF_String.