├── README.md
├── data
    └── df_masked.csv
├── notebooks
    ├── Real Data Example.ipynb
    └── Toy Example.ipynb
└── slides
    └── SpectralClustering.pdf


/README.md:
--------------------------------------------------------------------------------
 1 | # spectral_clustering
 2 | 
 3 | In this repository we present two demo notebooks how spectral clustering can be applied to extract relevant dynamics out of a variety of monitored metrics.
 4 | 
 5 | The toy example notebook sketches how spectral clustering works on a small perturbed block matrix with three (quasi-)blocks. 
 6 | 
 7 | In the real data example the algorithm is applied to real masked data. For this purpose is specified a similarity function, then computed a cluster assignment and reordered similarity matrix according to the identified quasi-blocks.
 8 | 
 9 | Once we have identified strongly connected clusters of metrics, we can compute a representative for each cluster by averaging over all assigned metrics. 
10 | 
11 | Eventually we give an visualization that provides a better insight, which metrics enter into the particular clusters together with a confidence corridor for each represantative.
12 | 
13 | For real data example we used the scikit-learn implementation of [Spectral Clustering](http://scikit-learn.org/stable/modules/generated/sklearn.cluster.SpectralClustering.html#sklearn.cluster.SpectralClustering) 
14 | 


--------------------------------------------------------------------------------
/slides/SpectralClustering.pdf:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/metterlein/spectral_clustering/d8df48e78d4c249e7cf2568bb5778d0da67d1649/slides/SpectralClustering.pdf


--------------------------------------------------------------------------------