Comparison with other packages
If you are already familiar with certain packages for calculating
topological features, you might be interested in understanding in
what aspects torch_topological differs from them. This is not
meant to be a comprehensive comparison; we are aiming for a brief
overview to simplify getting acquainted with the project.
giotto-tda
giotto-tda is a flagship
package, developed by numerous members of L2F. Its primary goal is to
provide an interface consistent with scikit-learn, thus facilitating
an integration of topological features into a data science workflow.
By contrast, torch_topological is meant to simplify the development
of hybrid algorithms that can be easily integrated into deep learning
architectures. giotto-tda is developed by a large team with a much
more professional development agenda, whereas torch_topological is
geared more towards researchers that want to prototype the integration
of topological features.
Teaspoon
Teaspoon is a library that
targets topological signal processing applications, such as the analysis
of time-varying systems or complex networks. Teaspoon integrates
very nicely with scikit-learn and targets a different set of
applications than torch_topological.
TopologyLayer
TopologyLayer is a library developed by Rickard Brüel Gabrielsson and others, accompanying their AISTATS publication A Topology Layer for Machine Learning.
torch_topological subsumes the functionality of TopologyLayer,
albeit under different names:
torch_topological.nn.VietorisRipsComplexortorch_topological.nn.CubicalComplexcan be used to extract topological features from point clouds and images, respectively.The
BarcodePolyFeatureandSumBarcodeLengthsclasses are incorporated as summary statistics loss functions instead. See the following example for more details: Point cloud optimisation with summary statisticsThe
PartialSumBarcodeLengthsfunction is not implemented, mostly because a similar effect can be achieved by pruning the persistence diagram manually. This functionality might be added later on.