neuralnetlib 4.0.7

• fix(LSTM): huge improvements in gradient flow
• feat(gradient_norm): better batch handling
• fix(Callbacks): now fully working generically
• fix(metrics): val_loss
• fix(Embedding): bias in output
• fix(display): val_*
• feat(metric): add pearsonr
• feat(metric): add kurtosis
• feat(metric): add skew
• fix(metric): skewness and kurtosis when variance=0
• fix(gan): batch logs
• fix(activation): config loading
• feat(lstm): improved flexibility and parameterizing
• docs(notebook): fresh run
• ci: bump version to 4.0.7

neuralnetlib 4.0.6

• refactor: autopep8 format
• ci: bump version to 4.0.6

neuralnetlib 4.0.5

• fix(cosine_sim): division by zero
• refactor: remove the use of enums
• docs: huge update and new examples
• docs(readme): update hyperlinks
• feat: add SVM
• fix(Dense): bad variable init
• fix(Tokenizer): bpe tokenizing
• fix: jacobian matrix instead of approximation
• fix(AddNorm): backward pass
• ci: bump version to 4.0.4

neuralnetlib 4.0.4

• feat(optimizers): add AdaBelief
• feat(optimizers): add RAdam
• feat(loss): add focal loss
• fix(optimizers-losses-activations): from_config
• feat(ensemble): add XGBoost
• feat(layer): add ActivationFunction init as string in Activation
• ci: bump version to 4.0.4

neuralnetlib 4.0.3

• refactor: some improvements
• feat: add validation_split parameter to fit method
• fix(lstm): vanishing and exploding gradient issue
• docs: update readme
• ci: bump version to 4.0.3

neuralnetlib 4.0.2

• feat(ensemble): add AdaBoost
• feat(ensemble): add GradientBoostingMachine
• ci: bump version to 4.0.2

neuralnetlib 4.0.1

• docs: add gif into notebook
• feat(GAN): auto sizing for epoch plot
• feat: add data imputation
• feat(ensemble): add IsolationForest
• feat(ensemble): add RandomForest
• feat(cluster): add KMeans
• feat(cluster): add DBSCAN
• docs: update readme

neuralnetlib 4.0.0

• feat: add GAN
• fix(GAN): weights and backpropagation
• refactor(BatchNormalization): more stable training
• fix(GAN): some fixes and general improvements
• feat(Loss): add WGAN loss
• feat(WGAN): improve generation
• ci: bump version to 3.3.9
• fix(WGAN): discriminator training on both real and fake data
• feat(GAN): improve image generation
• fix(Transformer): yet another mode collapse fix
• feat(layers): better gradient scaling and stability
• feat(Tokenizer): add BPE tokenize mode
• docs: update readme
• ci: bump version to 4.0.0

neuralnetlib 3.3.9

• feat: add GAN
• fix(GAN): weights and backpropagation
• refactor(BatchNormalization): more stable training
• fix(GAN): some fixes and general improvements
• feat(Loss): add WGAN loss
• feat(WGAN): improve generation
• ci: bump version to 3.3.9

neuralnetlib 3.3.8

• docs: add embedding to debug notebook
• refactor: huge simplification of sce to cels which conducted to higher BLUE score
• fix(Transformer): random state
• fix: some fixes and improvements
• fix: some fixes and improvements
• fix(MultiHeadAttion): attention weights now have correct ranges [-0.7;1.0]
• fix: some fixes and improvements
• fix: some fixes and improvements
• feat: new examples
• docs: update readme
• ci: bump version to 3.3.8