This codebase is no longer maintained; instead, use the Python implementation below.
The python implementation at https://github.com/shubhomoydas/ad_examples is current. This python implementation also supports tree-based classifiers and streaming tree-based classifiers. Also, the optimization in python implementation uses gradient descent with ideas borrowed from deep-network training such as RMSProp and ADAM which are more suited to the high-dimensional linear optimization as required for tree-based classifiers. These changes make the per-feedback optimization much faster than solving a large constrained linear programming problem while having similar detection performance.
To execute the code:
-
Make sure the file paths are correctly set in loda/alad-paths.sh
-
Make sure the 'srcfolder' correctly points to the parent folder of 'loda' directory in the files loda/R/alad.R and loda/R/ai2.R
-
To execute AAD on (say) 'toy' dataset: cd loda bash ./run-active-mult.sh toy 3 1 0.03 1
-
To execute AI2 on (say) 'toy' dataset: cd loda bash ./run-ai2-mult.sh toy 10 0.03
-
To generate summary plots:
- first check folder paths are correctly set in loda/R/prep-alad-summary.R
- next: cd loda bash ./prep-alad-summary.sh toy 3 1 0.03