Releases: mvinyard/torch-nets
Releases · mvinyard/torch-nets
v0.2.0
v0.2.0rc0
Full Changelog: v0.1.0...v0.2.0rc0
v0.1.0
v0.1.0rc1
v0.1.0rc0
v0.0.5
pre-release: v0.0.5rc0
Full Changelog: v0.0.4...v0.0.5rc0
v0.0.4
v0.0.4rc0
Full Changelog: v0.0.3...v0.0.4rc0
v0.0.3
Changes to core
- Revised the
TorchNetsuch that it inheritstorch.nn.Sequential, which is more intuitive in creating network objects through this interface. (commit: 6fe33cb3) - Added the
potentialproperty to theTorchNetwhich serves as a boolean indicator flag (commit: 42d0940) EncoderandDecoderwere updated accordingly as they are derived from theTorchNet(commit: 1f45826)- Updated
AugmentedTorchNetclass, which is also derived from theTorchNet(commit: 2de1bde)
Added the plotting (.pl) module
torch_nets.pl.weights_and_biases- plot the weights and biases of a given network. (commit: 63e045)
Added the tools (.tl) module
torch_nets.tl.infer_network_architecture_from_state(commit: 357fc37)- This enables one to pass a
state_dict- often saved as ackpt_model.ptand return a network composed with these weights. - Along with the
state_dict,dropoutandactivationare required to be passed with each layer, since these are not saved as part of thestate_dict.
- This enables one to pass a
Fixes made during testing in pre-release:
v0.0.3rc0
- Updated
setup.pywith additional imports (vinplots, which also brings inmatplotlib)
v0.0.3rc1
- Reverted
namesas an attribute ofTorchNet(commit: c2abc11)
v0.0.3rc2
- Add example notebook of classes available in the API (commit: 841bb58)
Still to do
- Documentation! Quarto is the likely path forward.
Full Changelog: v0.0.2...v0.0.3