QueuingBandits

Code for Regret of Queuing Bandits

Paper url: https://arxiv.org/abs/1604.06377

Dependencies: pandas, numpy, matplotlib, multiprocessing

Function to simulate QB for many runs in parallel: ep_greedy_avg_wopt_par_var(mu,u,K,l,t,num_sim,nthread)

'''
   num_sim: number of experiments
   nthread: number of parallel threads (recommended to number of cores in the machine)
   mu: True service rate matrix (u * K numpy matrix : all elements less than 1.0)
   u: number of users
   K: number of servers	   
   l: service rates (listof size u)
   t: Time Horizon
   Returns: (q,q0,sq)
   q: array of average queue lengths for bandit algorithm for u queues each till time horizon t
   q0: array of average optimal queue length for u queues each till time horizon t
   sq: std. of queue regret (q - q0) over all the runs
'''
	 

Example File

1. An example run file has been provided in ./examples/ folder. To run the example files clone the repo. Go to the directory of the repo. Then run the example file:

python ./examples/Q_run_ep.py

2. The output will be stored as two .npy files in the ./results/ directory.