0.6.11

What's Changed

• Support python 3.7.0 by @cgarciae in #65
• Take any version of cerifi by @jonringer in #73
• Relax dependency restrictions by @cgarciae in #77

New Contributors

• @jonringer made their first contribution in #73

Full Changelog: 0.6.10...0.6.11

0.6.10

🚀 Features

• Add ConvTranspose
  • PR: #64

0.6.9

Changes

• Improves Release CI #61 (cgarciae)
• Bumps flax to 0.4.0 #60 (ptigwe)
• Add LayerNorm and GroupNorm #58 (lkhphuc)
• Initial implementation of GRU layers #48 (ptigwe)

0.6.8

Changes

• Fix metric argument names #56 (cgarciae)
• Add Embed layer #54 (cgarciae)

0.6.7

Changes

• @cgarciae: Re-exposes tx.Treex and tx.Filters and conditionally exposes HaikuModule #53

0.6.6

Changes

• @cgarciae Fix Documentation (#51)
• @cgarciae Update Treeo (#52)

0.6.5

Changes

• Add MSE and tests #50 (jmarrietar)
• Add trainable_parameters for extracting non-frozen parameters #47 (ptigwe)
• Remove duplicate import of Haiku Module #44 (ptigwe)
• Update select_topk to not use deprecated function #43 (ptigwe)
• Fix typo #42 (BioGeek)
• Feature/metrics #17 (anvelezec)

0.6.4

Changes

• Fix import issue when dm-haiku is not available.

0.6.3

Changes

• Adds experimental axis_name argument to next_key, KeySeq, and Linear.
• Creates the preserve_state function that enables you apply a function transformation like jit or vmap to a stateful method that doesn't propagate the state through and output. preserver_state will return the first argument (usually self) and update it after the transformation.
• Fixes type issues with Filter in shortcut methods.

0.6.2

Changes

• Adds HaikuModule that can wrap any haiku.TransformedWithState.
• Adds an example using haiku.
• Improves FlaxModule