v0.5.1

What's Changed

• Constrained scipy to >=1.9 to avoid bug in median calculation in #154
• Switch type-checking library from torchtyping to jaxtyping in #153
• Add support for Python 3.12 in #156

Full Changelog: v0.5.0...v0.5.1

v0.5.0

What's Changed

• Added KAGRA as detector #132
• Added IMRPhenomP implementation #138
• SineGaussian returns polarizations in dtype of passed parameters #139
• Refactored TaylorF2 and IMRPhenomD into torch.nn.Modules #137
• Added dimensionality check to GroupNorm1D #134

Bug Fixes

• Fixed typehints in ml4gw.distributions module #129
• numpy version restricted to <2.0.0 #145

Full Changelog: v0.4.2...v0.5.0

v0.4.2

• InMemoryDataset now inherits from torch.utils.data.IterableDataset for compatibility with torch.utils.data.Dataloader #119
• ml4gw.distributions now inherit from torch.distributions module

Bug Fixes

• Fixed deprecation warning for use of torch.Tensor.T in ml4gw.waveforms.IMRPhenomD #114

v0.4.1

What's Changed

• Remove support for torch<2.0 #122

v0.4.0

What's Changed

• Add GroupNorm, ResNet1d and ResNet2d modules
• Adds MultiresolutionSpectrogram transforms
• Add Sphinx Docs

Bug Fixes

• handle corner case of eta=0.25 giving nan in phenom_d

v0.3.0

Feature additions

• augmentations module that adds SignalInverter and SignalReverser modules
• dataloading.ChunkedTimeSeriesDataset replaces dataloading.ChunkLoader
• dataloading.Hdf5TimeSeriesDataset for loading directly from hdf5 files on disk
• ShiftedPearsonCorrelation transform
• New nn submodule that contains:
  • autoencoder library for building arbitrary autoencoder structures
  • streaming library that contains Snapshotter and OnlineAverager modules for IaaS deployment
• Torch implementations of TaylorF2 and PhenomD waveforms

Deprecations

• dataloading..ChunkLoader removed in favor of dataloading.ChunkedTimeSeriesDataset

v0.2.0

Feature additions

• waveforms module for generating gravitational wave polarizations from source parameters
• whiten function and Whiten module for performing time-domain whitening using arbitrary background data
• ChunkedDataloader for loading time domain strain data from disk on-the-fly during training
• Modularizing RandomWaveformInjector into more functional submodules:
  • gw.compute_observed_strain wrapped into WaveformProjector module
  • SnrRescaler module for rescaling waveform amplitudes to achieve target SNR values
• InterferometerGeometry module encoding interferometer properties locally without having to rely on gwpy

Deprecations

• spectral.normalize_psd removed in favor of enforcing use of torch Tensors during transform fitting
• gwpy dependency removed