Skip to content

Latest commit

 

History

History
57 lines (48 loc) · 3.15 KB

README.md

File metadata and controls

57 lines (48 loc) · 3.15 KB
Raw

Topology and Meta-Learning

Brief Intro

About Naming

The initial object is to apply topology into meta-learning as a tool for characterisation of datasets. Yet, topology is a bit hard for me at that moment (and now), and thus I switched to an easier, conventional way of doing meta-learning.

Abstract (Copied from published paper)

Meta-features describe the characteristics of the datasets to facilitate algorithm selection. This paper proposes a new set of meta-features based on clustering the instances within datasets. We propose the use of a greedy clustering algorithm, and evaluate the meta-features generated based on the learning curves produced by the Random Forest algorithm. We also compared the utility of the proposed meta-features against pre-existing meta-features described in the literature, and evaluated the applicability of these meta-features over a sample of UCI datasets. Our results show that these meta-features do indeed improve the performance when applied to the algorithm selection task.

Structure

All scripts are coded for Python3.6/3.7

  • ./requirement.txt
    • Dependencies to run scripts in this repository
    • Outdated, please refer to the import list at the top of each script
  • ./convex_hull_cluster.py
    • Greedy algorithm with high time complexity
    • Use convex hulls to cluster the instances while ensuring homogeneity
    • Dependency
      • Internal: ./meta_features.py
      • External: numpy, scipy
    • Command example: python3 convex_hull_cluster.py path/to/input/[file]
      • Output files
        • [file].clusters.json: saved clustering results. Can be used to prevent re-calculation
        • [file].output.json: saved results with all meta-features' values calculated
        • [file].log
  • ./spherical_cluster.py
    • Greedy algorithm with moderate time complexity
    • Similar to convex hull algorithm but uses spherical clusters instead
    • Dependency
      • Internal: ./meta_features.py
      • External: numpy
    • Command: (-h option for help)
      • E.g.: python3 spherical_cluster.py -r path/to/dir/with/inputs --log path/to/log
      • E.g.: python3 spherical_cluster.py -i path/to/input/file
    • Output files:
      • [input].cluaters.json: saved clustering results
        • Cannot be used to restore clustering results
        • (Too lazy to implement)
      • [input].output.json: saved results with all meta-features' values calculated
  • ./sperical_brute_force.py
    • Brute force algorithm for optimal clustering result with the constraint of spherical clusters. Extremely high time complexity
      • Enumerate and check every possible combination of splitting the dataset into subsets
    • [Same as "./spherical_cluster.py" above]
  • ./meta_features.py
    • Separate module to calculate meta-features with clustering results
    • Do not run it directly
  • ./learning_rate.py
  • utilities/
    • utilities/extract_meta_features.py:
      • Extract meta-features from .clusters.json (raw clustering results file), not .output.json (suppose to contain meta-features, but depreciated)

[WIP]