Skip to content

tse-group/BanditOptimization

Repository files navigation

Notes

  • If distance between datapoints heavily depends on a few co-oridinates then 'smooth' the data using Fast Walsh-Hadamard transform before using the software (for l2 distance).
  • Dependendices : g++-4.9, gcc-4.9, cmake, libboost-all-dev, build-essential, libhdf5-serial-dev.
  • Run ./utils/build.sh to build all the binaries.
  • aws_server.ini stores all the information (such as datapath, saveResultsPath) required by the software.

Reproducibility

  • All the figures can be generated in ipython notebooks in 'figure' folder.
  • The figures are generated from experiments - stored in 'experiments' folder.
  • The stored experiments for dense datasets can be reproduced using the following lines of code
    • K-nearest neigbhours : ./build/knn aws_server.ini start-index end-index (finds the k nearest points for points from start to end index)
    • K-means : ./build/kmeans aws_server.ini
    • Heirarchical: ./build/heirarchical aws_server.ini random-seed
    • Mutual Information Feature Selection: ./build/gasmmi aws_server.ini number-features sample-size random-seed
  • The stored experiments for sparse datasets can be reproduced using the following lines of code
    • K-nearest neigbhours : ./build/knn10x aws_server.ini start-index end-index (finds the k nearest points for points from start to end index)
    • K-means : ./build/kmeans10x aws_server.ini

Empirical results on Imagenet (Main results)

Results on Different tasks

Empirical results on sparse datasets

More Results

Datasets

Hadamard Transform

Hadamard

Software Architecture

Software Architecture

About

No description, website, or topics provided.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published