I have 20,000 histograms and I want to analyse them for unimodal or bimodality.
Do any one know about any R command to automatically analyse them in a single go?
You can model this as a gaussian or laplacian mixture model and extract the different modes using unsupervised clustering based on the expectation maximisation algorithm, as demonstrated in these papers:
http://www.sciencedirect.com/science/article/pii/S0167865505002813
I used a similar approach for detecting the underlying Laplacian distributions in my work on privacy metrics:
http://arxiv.org/abs/1308.5421
R has a package called mixtools that you probably can use:
http://sites.stat.psu.edu/~dhunter/papers/mixtools.pdf
Good point - to elaborate a bit: diptest: Hartigan's Dip Test Statistic for Unimodality - Corrected.
https://cran.r-project.org/web/packages/diptest/index.html
- Badges
- Science topic