This is my compact example of script and version tracking in github
You will find example code in the script folder, I have also provided a markdown file in this folder to ease code output and interpretation.
ANN_sensitivity.R -
This is a draft script of a sensitivity analysis of variable importance in feed-forward ANNs. I utilize the variable perturbation methodlogy as a means to test variable importance.
Under repeated ANN training rounds of weighting coefficients, on random subsets of the training data, we get a distribution of feature importance rankings that better approximate the stochastic range of feature importances. This approach yields a far more robust measure of feature importance than many existing methods, and subsequently better informs feature selection.
This work is drawn from a larger project that I am currently involved with. This tool is a component of work that is developing a pipeline of analysis for complex, non-linear, and interactive relationships in global biodiversity patterns. This approach being applied both on land in the ocean with a diverse array of data sources such as satellite remote sensing data and ocean observer records.
Contact me anytime at [email protected] for more info.
tyegagne.com