This repository contains data and instructions for how to implement deep embedded clustering (DEC) of seismic data recorded on the Ross Ice Shelf, Antarctica from 2014-2017. This package is an accompaniment to a paper submitted to the Journal of Geophysical Research (Jenkins II et al., submitted Jan 2021).
RISCluster is in the process of being restructured so that it can be installed and run on a Mac or Linux environment.
This repository is a PyTorch implementation of DEC. The workflow is as follows:
- Load and pre-process data
- Construct a convolutional auto-encoder (AEC)
- Tune, train, and validate AEC
- Incorporate clustering layer into AEC model architecture
- Intialize clusters (K-Means, GMM available)
- Train the DEC model: clustering and model training are simultaneous.
- Once trained, infer class labels for remainder of the data set.
- Install RISCluster using instructions contained in the RISCluster repository readme.md.
- Clone this repository to your desired working directory.
- Download environmental data and unzip contents into
/Data.
The Jupyter notebook Workflow.ipynb contains an end-to-end workflow control that guides the user through all steps of the project, including downloading and pre-processing the seismic data. Required directories, configuration files, and command-line scripts are generated within the notebook. For main routine execution, commands are copied from the Jupyter notebook into a terminal window.
Downloading and processing seismic data can take a long time. For access to the pre-processed seismic data set (16 GB), please contact me and we can arrange how best to transfer the file.
Submitted: William F. Jenkins II, Peter Gerstoft, Michael J. Bianco, Peter D. Bromirski; Unsupervised Deep Clustering of Seismic Data: Monitoring the Ross Ice Shelf, Antarctica. Submitted to Journal of Geophysical Research on 20 Jan 2021; doi: https://doi.org/10.1002/essoar.10505894.2
Dylan Snover, Christopher W. Johnson, Michael J. Bianco, Peter Gerstoft; Deep Clustering to Identify Sources of Urban Seismic Noise in Long Beach, California. Seismological Research Letters 2020; doi: https://doi.org/10.1785/0220200164
Junyuan Xie, Ross Girshick, Ali Farhadi; Unsupervised Deep Embedding for Clustering Analysis. Proceedings of the 33rd International Conference on Machine Learning, New York, NY, 2016; https://arxiv.org/abs/1511.06335v2
Project assembled by William Jenkins
wjenkins [@] ucsd [dot] edu
Scripps Institution of Oceanography
University of California San Diego
La Jolla, California, USA