Unsupervised analysis of the time evolution of clusters of VLP signal waveforms (timeseries) from the Stromboli volcano.
The selected signals stored in day files can be cutted with day_to_ascii.py.
Filtering, decimating, LPC and amplitude parametrization of the created ascii files is done with the script ascii_to_param.py. A visualization for an exemplary signal can be found in parametrization.ipynb.
The script autoencoder.ipynb implements autoencoders and can be run on Google Colab.
The expensive t-SNE algorithm and the Bayesian optimization of the SOM can be run with tsne_somopt.ipynb and then saved for later usage.
An example to cluster the filtered signals can be found in cluster.ipynb. The SOM and clustering with K-Means on the weight vectors of the SOM is done in SOM.ipynb
The time evolution of the clusters of the different methods can be found in ratio.ipynb