neuralnetlib 3.2.0

• feat(metrics): add roc auc, pr auc, r2 and classifrep
• feat(model): add gradient clipping and matrix padding optimizations
• docs(notebook): fresh run
• feat(Model): add temperature and sequence generation with improved sampling
• ci: bump version to 3.2.0

neuralnetlib 3.1.2

• feat: add n-gram class
• ci: bump version to 3.1.2

neuralnetlib 3.1.1

• fix: Adam time increment
• ci: bump version to 3.1.1

neuralnetlib 3.1.0

• docs: update dinosaurs example
• perf(Adam/RNN): better gradient handling
• ci: bump version to 3.0.8
• ci: bump version to 3.1.0

neuralnetlib 3.0.8

• docs: update dinosaurs example
• perf(Adam/RNN): better gradient handling
• ci: bump version to 3.0.8

neuralnetlib 3.0.7

• fix(attention): correct gradient computation in dot-product attention
• fix(RNN): global clipping instead of individual
• docs: update readme
• ci: bump version to 3.0.7

neuralnetlib 3.0.6

• fix: GRU to Model exceptions
• ci: bump version to 3.0.6

neuralnetlib 3.0.5

• feat: add GRU layer
• ci: bump version to 3.0.5

neuralnetlib 3.0.4

• feat: add sparse cce
• fix: f1score, precision, recall for bc
• refactor: better weights and biases init
• fix(Adam): limit cases and gradient clipping
• docs: update examples
• feat: improve dropout and add new regularizers
• feat(LSTM): vanishing gradient + gradient clipping
• ci: bump version to 3.0.4

neuralnetlib 3.0.3

• perf: less memoryvore
• ci: bump version to 3.0.2
• Merge pull request #48 from marcpinet/perf-improve-lstm-attention
• fix: evaluate new batch handling
• docs: add method typing
• fix(Dense): correct temporal data handling in forward and backward passes
• ci: bump version to 3.0.3