Releases: ChrisLisbon/TorchCNNBuilder
Releases · ChrisLisbon/TorchCNNBuilder
v0.1.5
v0.1.4
New features:
- adding the ability to work with latent space dimensions through the new
LatentSpaceModule - adding new
LatentSpaceModuleblock to theForecasterBasewith corresponding features - updating corresponding doc examples
Code refactor:
- minor code refactoring
- updating some doc examples
v0.1.3
v0.1.2
- Fixing padding bug in convolution blocks of several convolution layers
- Removing mandatory initialization of
input_sizeinBuilder(+ adding new validations for create sequence builder methods) - Improving typing of function parameters
- Improving
Builderclass encapsulation - Restructuring code namespaces
- Updating and improving doc examples
v0.1.1
- updating
ForecasterBaseinterface - fixing bug with 3d decoder output in
ForecasterBase - adding new validation functions
- updating doc
- updating linter script
- renaming channel growth param value
ratio->exponential - adding new channel growth param value
power(now by default inForecasterBase)
v0.1.0
Code refactoring:
- decomposition of functions into different files
- restructuring of namespaces during import
- adding a linter pipeline after each push action in pull-requests
- adding a linter script for the developers
- refactoring all code logic with linter
- updating doc
New features:
- new normalization in the convolution block as
InstanceNormNd - new adaptive pooling in the decoder as
AdaptiveMaxPoolNd - new channels counters like
linearandconstant ascendingparam is replaced bychannel_growth_ratewhich can beratio(same asascending = False),proportion(same asascending = True),linearandconstant- new 3d convolutions in
ForecasterBase