Skip to content

Latest commit

 

History

History
38 lines (33 loc) · 2.21 KB

README.md

File metadata and controls

38 lines (33 loc) · 2.21 KB
Raw

RL-Algorithms-and-Optimisation

The repository contains the beginner Reinforcement Learning algorithms dealing with the Exploration vs Exploitation dilemma, which are:

  • Epsilon-Greedy Algorithm
  • UCB Algorithm
  • KL-UCB Algorithm
  • Thompson Sampling Algorithm

Problem Statement

Please head over to the assignment webpage on the instructor's website.

Running and executing the algorithms

All of the classes, algorithms and useful functions are implemented in the file submission/bandit.py

bandit.py takes 7 command line arguments:

  • --instance: the path of the the bandit instance
  • --algorithm: the algorithm out of epsilon-greedy-t1, ucb-t1, kl-ucb-t1, thompson-sampling-t1, ucb-t2, alg-t3, alg-t4
  • --randomSeed: a non-negative integer to make the process deterministic
  • --epsilon: a non-negative real number in [0, 1]; relevant to the Epsilon Greedy methods
  • --scale: a positive real number, denoting the confidence value in the UCB algorithm, c
  • --threshold: a real number in [0, 1]; relevant to the Task 4
  • --horizon: the number of pulls

and prints the 7 arguments along with the REGRET and HIGHS calculated in a comma-separated form.

The script can be run using the following command using the appropriate and customised parameters for the conditional arguments:

~RL-Algorithms-and-Optimisation/submission $ python3 bandit.py --instance {path} --algorithm {algo} --randomSeed {seed} --epsilon {eps} --scale {c} --threshold {th} --horizon {hz}

For example

~RL-Algorithms-and-Optimisation/submission $ python3 bandit.py --instance ../instances/instances-task1/i-2.txt --algorithm ucb-t1 --randomSeed 499 --epsilon 0.02 --scale 2 --threshold 0 --horizon 27

Output

../instances/instances-task1/i-2.txt, ucb-t1, 499, 0.02, 2.0, 0.0, 27, 6.5, 0

The submissions directory contains a script runner.py to run all the tasks and prints data. Run the script and pipe the output to a file (there will be 9000+ lines printed). The complete task would take almost 4-6 hours to run. The bottom line is an output example.