This is a R implementation of the structure-aware algorithm aimed to at revealing and consolidating continuous, low-dimensional and high- density structures in the underlying higher-dimensional data, while ignoring noise and outliers. The theory, proof of convergence to the exact underlying data manifolds (under Gaussian noise assumption) and a deep investigation of its performance under different scenario can be found in Wu, Shihao, et al. [Wu, Shihao, et al. "Structure-aware Data Consolidation." IEEE transactions on pattern analysis and machine intelligence (2017)]
- PrincDevelTraj.R:
Rscript. - coordinates_sorted_HSPC_dataset.txt: Force-directed layout generated by means of SPRING for sorted HSPC dataset.
- coordinates_CD34_CD164_dataset.txt: Force-directed layout generated by means of SPRING for sorted HSPC dataset.
The script requires only a standard computer with enough RAM (>4Gb) to analyze a dataset with 10K obs. in a 2-dimensional space. The runtimes on provided datsets is approximately 4 minutes.
The script had been developed and tested on R version 3.4.x
Users should install the following packages prior to run PrincDevelTraj.R, from an R terminal:
install.packages(c('RANN', 'parallel'))
To reproduce results published in our paper set parameters with the following values:
- coordinates_sorted_HSPC_dataset.txt -> radius= 0.05; mu= 0.3;
- coordinates_CD34_CD164_dataset.txt -> radius= 0.02; mu= 0.3;