v0.5.0: new hearth.text subpackage and transformers + more

What's Changed

• expand user on load and save json (#67) @leaprovenzano

New Features

• add PearsonCorrCoef metric (#73) @leaprovenzano
• Add Bertish (#72) @leaprovenzano
• add bert style WordPieceTokenizer (#69) @leaprovenzano
• add base Tokenizer (#68) @leaprovenzano
• add norm_whitespace helper (#66) @leaprovenzano
• add pad_tokens helper (#64) @leaprovenzano
• add AttentionPool1D (#63) @leaprovenzano

Bug Fixes

• fix scriptability in TimeMasked (#71) @leaprovenzano
• fix bos and eos idx names (#70) @leaprovenzano
• fix name of data section in docs (#65) @leaprovenzano

Documentation

• fix missing Bertish in docs (#74) @leaprovenzano

v0.4.0 : Angular losses and masked MSE loss

What's Changed

New Features

• Add Angular penalty soft margin losses (#62) @leaprovenzano
• add masked mse loss (#61) @leaprovenzano

v0.3.0: Transformer bits, data tidy up, better device support

What's Changed

• refactor/tidy up data related functionality (#55) @leaprovenzano

New Features

• add transformer layers (#60) @leaprovenzano
• add MaskedLayerNorm (#59) @leaprovenzano
• add SelfAttention module (#58) @leaprovenzano
• add TimeMasked module wrapper (#57) @leaprovenzano
• add AbsolutePositionalEmbedding (#56) @leaprovenzano
• add better device support to Loop (#54) @leaprovenzano

v0.3.0

What's Changed

• No changes

v0.2.0: hearth.vision and minor improvement to FineTuneCallback

What's Changed

New Features

• base vision subpackage (#51) @leaprovenzano
• add max_depth arg to FineTuneCallback (#52) @leaprovenzano

v0.1.0: add FineTuneCallback

New Features

• add FineTuneCallback (#49) @leaprovenzano