Skip to content

- Rademacher complexity - ERM algo on boolean conjunction prediction - online learning

Notifications You must be signed in to change notification settings

ndsgit01/learning-theory

Repository files navigation

learning-theory

boolean_conjunction_prediction.py

  • implements functions required to predict the formula of conjunction of boolean literals from training data, measure / estimate errors related to the same
  • Ref: https://cs.nyu.edu/~mohri/mlbook/ [second edition, Example 2.6 (Conjunction of Boolean literals)], modified slightly to also allow detecting inability to predict a well defined hypothesis from avalable samples.

rademacher_complexity.py

online_learning.py

  • implements algorithms to predict with expert advice for the on-line learning scenario
    • Halving algorithm
    • Weighted majority algorithm
    • Exponential weighted average algorithm
  • also implements a simple single thread on-line scenario simulation class that synchronizes generators - takes in generators for input and labels and ensures that labels can be consumed only after a corresponding input has been consumed.
  • Ref: https://cs.nyu.edu/~mohri/mlbook/ [second edition, Section 8.2 (On-Line Learning > Prediction with expert advice)]

About

- Rademacher complexity - ERM algo on boolean conjunction prediction - online learning

Topics

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages