This repo includes implementations of various subspace clustering algorithms in Julia, tested on Julia 1.5. To run, they require several packages, which can be installed by julia requirements.jl.
This also provides an implementation for A-DSSC and J-DSSC, the two methods from our paper:
Doubly Stochastic Subspace Clustering. Derek Lim, René Vidal, and Benjamin Haeffele. arXiv:2011.14859.
The experimental setup is as described in the paper.
The main implementations for computing the self-representation matrices are in rep_solver.jl. We also call the Python implementations by Chong You for EnSC and SSC as located in selfrepresentation.py. The file main.jl gives a framework to run these methods with some grid searches over parameters. The usage is:
julia main.jl $DATASET $METHOD
where the DATASET options are: umist, scattered_coil scattered_umist, scattered_coil_40, and the METHOD options are: adssc, jdssc, jdssc_l1, tsc, lsr, lrsc, ssc_omp, ssc, ensc.
For instance, to run LSR on UMIST, do:
julia main.jl umist lsr
Data loaders are in data_gen.jl. Other datasets can be experimented on by making another function in the format as in the file, i.e., a function that returns the data matrix norm_X with points normalized to unit l2 norm, the integer nspaces with the number of (subspace) clusters, and array label where label[i] is the subspace cluster of point i.