0.6.0

Merged pull requests:

• More Steps: adds additional *_step methods #163 (cgarciae)
• Better error message on elegy.States attribute access miss #162 (alexander-g)
• Improve Docs: adds more docs + an example using pmap to distribute trianing #156 (cgarciae)
• WGAN-GP low-level API example #157 (alexander-g)

0.5.0

This version simplifies parts of the low-level API in spirit of what was introduced in 0.4.0 to provide a more homogeneous and simpler experience.

Merged pull requests:

• Improve States: uses __dict__ so States works with vars #159 (cgarciae)
• Simplify API: Cleans-up some API details around Model and Module #158 (cgarciae)

0.4.1

Merged pull requests:

• fix-maybe-initialize #155 (cgarciae)
• Add simple Flax low-level API Model example to README.md #153 (sooheon)

0.4.0

Implemented enhancements:

• [Feature Request] Monitoring learning rates #124

Merged pull requests:

• Update Getting Started + README #152 (cgarciae)
• Pretrained ResNet fix after #139 #151 (alexander-g)
• Dataset: better default batch_fn and custom batch_fn #148 (alexander-g)
• Label Smoothing for Binary Crossentropy #146 (alexander-g)
• Add adapter for handling torch dataloaders #145 (charlielito)
• Feature/tf dataset adapter #144 (charlielito)
• [*.md,*.py,*.sh] Fix typos #142 (SamuelMarks)
• verbose=4 #140 (alexander-g)
• Framework Agnostic API: Introduces a new low-level API, removes the dependency between Model and Module, adds support for Flax and Haiku, simplifies hooks. #139 (cgarciae)
• DataLoader Optimizations #137 (alexander-g)
• Autodownload pretrained ResNet #136 (alexander-g)
• Add learning rate logging #135 (cgarciae)
• Adds gitpod support to be able to develop elegy on the cloud #134 (cgarciae)
• Make Models Pickleable Again #133 (alexander-g)
• SCCE fix for bug in Jax<0.2.7 #130 (alexander-g)
• table progress #127 (alexander-g)

0.3.0

Implemented enhancements:

• elegy.nn.Sequential docs not clear #107
• [Feature Request] Community example repo. #98

Fixed bugs:

• [Bug] Accuracy from Model.evaluate() is inconsistent with manually computed accuracy #109
• Exceptions in "Getting Started" colab notebook #104

Closed issues:

• l2_normalize #102
• Need some help for contributing new losses. #93
• Document Sum #62
• Binary Accuracy Metric #58
• Automate generation of API Reference folder structure #19
• Implement Model.summary #3

Merged pull requests:

• sparse\_categorical\_crossentropy should check bounds #123 (alexander-g)
• float sample_weight for precision/recall metrics #122 (alexander-g)
• Added Huber loss #121 (abhinavsp0730)
• ResNet Docs + CIFAR10 Example #119 (alexander-g)
• Dataset & DataLoader #118 (alexander-g)
• fix/docs #116 (cgarciae)
• Better save + load #114 (cgarciae)
• Examples Cleanup #113 (alexander-g)
• merge resnet into master #111 (cgarciae)
• Fix metrics error #110 (cgarciae)
• Fix colab notebook getting started #105 (charlielito)
• Added Cosine Similarity loss. #103 (abhinavsp0730)
• small change to trigger build #101 (charlielito)
• New metrics #100 (anvelezec)
• Update CONTRIBUTING.md #97 (haruiz)
• Enhance docs #96 (charlielito)
• Loss Mean Squared Logarithmic error. #95 (abhinavsp0730)
• Documentation improvements #94 (chjort)
• Module v3 #92 (cgarciae)
• Documentation fixes of module-system.md #91 (chjort)
• binary precision and recall metrics #86 (anvelezec)

0.2.2

• Fixes metrics.BinaryAccuracy and losses.BinaryCrossentropy

0.2.1

• Uses optax instead of optix.
• Implements BinaryAccuracy.

0.2.0

• Big refactor. Elegy has its own Module system independent of Haiku and its now incompatible with it. #85

0.1.5

• Mean Absolute Percentage Error Implementation @Ciroye
• Adds elegy.nn.Linear, elegy.nn.Conv2D, elegy.nn.Flatten, elegy.nn.Sequential @cgarciae
• Add Elegy hooks @cgarciae
• Improves Tensorboard support @Davidnet
• Added coverage metrics to CI @charlielito

0.1.4

• Adds elegy.metrics.BinaryCrossentropy @sebasarango1180
• Adds elegy.nn.Dropout and elegy.nn.BatchNormalization @cgarciae
• Improves documentation
• Fixes bug that cause error when using is_training via dependency injection on Model.predict.