Sub-Image Analysis using Topological Summary Statistics.
The purpose of this repo is to provide the exact scripts/code used to produce to the simulation and real data results in the SINATRA manuscript.
For the version of SINATRA used to generate the results, please use the command: devtools::install({directory_where_you_cloned_this_repo}/SINATRA_AOAS_results/sinatra)
The relevant locations for the scripts used to generate each figure, and process the data for each figure are located under Scripts/figure_generation. To run the Limit Shapes algorithm, we refer to Huang et. al (2019). To access the gitrepo for Limit Shapes, please use this link. The MATLAB driver scripts are provided in Scripts/LimitShape_Scripts . Please note that to use these scripts, you will need to clone the LimitShape repo using git clone https://github.com/ruqihuang/LimitShape and add the local repo/utils path at the top of your MATLAB script.
To run a demo of the ECT alignment algorithm, we have provided a tutorial in the folder Scripts/postECT_alignment. Note that the permutation to rotation step is quite time consuming, so we have provided pre-computed permutation-rotations in the dropbox link provided below in the directory named ECT_alignment_demo.
For the locations of the data used in the manuscript, please use this link.
To generate the caricatured data, run the MATLAB scripts in Scripts/Data_Generation/GHdist/, originally sourced from: https://github.com/shaharkov/GPLmkBDMatch. The scripts to generate the data in the simulations are in Scripts/Figxx.
This branch contains further inference functions for implementing the SINATRA pipeline, including the use of Deep Gaussian Process Classification as an inference tool. The model and inference algorithm are obtained from Havasi et al. 2018, Inference in Deep Gaussian Processes using Stochastic Gradient Hamiltonian Monte Carlo. Example cases using this method can be found in the script SINATRA_deepgp_ROC.R, which relies on the python file sinatra_rate.py to generate posterior samples from the Deep GP. These are then used to generate feature importance values via RATE. Code for inference in the Deep GP model is contained in Simulations\sghmc_dgp.
B. Wang*, T. Sudijono*, H. Kirveslahti*, T. Gao, D.M. Boyer, S. Mukherjee, and L. Crawford. SINATRA: a sub-image analysis pipeline for selecting features that differentiate classes of 3D shapes. Annals of Applied Statistics. 15(2): 638-661.