glvi 0.1

Globally local variable importance

Globally-local-variable-importance-algoritm-in-python

• Two methods to calculate local variable importance with global models.
  • Since the variable importance given by scikit-learn's RandomForestRegressor only shows a global “averaged” variable importance for the whole data set, it is inadequate to represent the variable importance for certain areas or episodes (local group).
  • For several months working on this issue, I finally proposed two methods and realised them in python. With these two methods, we did not need to exclusively build local models to estimate local variable importance any more especially when model with a larger space-time range has a better performance and when local models deviate because of too small dense of data.

Hopefully, I upload the codes to get some contribution from developers all over the world. Simutaneously, we wish to obtain some suggestions.

• If you have questions or suggestios, please connect to the only auther Tao Li, Sichuan Univercity, China with the email:[e-mail].