Release 3.4.0
Compatible with Nengo 3.0.0 - 3.1.0
Compatible with TensorFlow 2.2.0 - 2.4.0
Added
- Added support for KerasSpiking layers in the Converter. (#182)
- Added support for
tf.keras.layers.TimeDistributed
in the Converter. (#182) - Added support for TensorFlow 2.4. (#185)
- Added support for Nengo 3.1. (#187)
Changed
- Minor improvements to build speed by building constants outside of TensorFlow. (#173)
- Support for PES implementation changes in Nengo core (see #1627 and #1640). (#181)
Fixed
- Global default Keras dtype will now be reset correctly when an exception occurs in a Simulator method outside the
with Simulator
context. (#173) - Support new LinearFilter step type introduced in Nengo core (see #1629). (#173)
- Fixed a bug when slicing multi-dimensional Signals (e.g. Ensemble encoders). (#181)
- Fixed a bug when loading weights saved in a different Python version. (#187)