/sdmp

Experiments for a Q-Based Evolutionary Algorithm (QBEA)

Primary LanguageJavaMIT LicenseMIT

Sequential Decison Making Problems (SDMPs)

Code for : Aman Soni, Peter R. Lewis and Anikó Ekárt. "Offline and Online Time in Sequential Decision-Making Problems" Computational Intelligence in Dynamic and Uncertain Environments(CIDUE),2016 IEEE Symposium on. IEEE, 2016.

The hypothesis of this work is a method for performance comparison of algorithms. The motivation was to fill the need to consider resource usage when comparing evolutionary dynamic optimisation algorithms to algorithms from reinforcement learning. The higher level of abstraction allows algorithm designers a wider selection of algorithms to solve sequential decision-making problems.

This research is based on:

  • Fu, Haobo, et al. "What are dynamic optimization problems?." Evolutionary Computation (CEC), 2014 IEEE Congress on. IEEE, 2014.
  • Fu, Haobo, Peter R. Lewis, and Xin Yao. "A Q-learning Based Evolutionary Algorithm for Sequential Decision Making Problems.

Q-Learning is an implementation of Q Learning from Watkins, Christopher John Cornish Hellaby. Learning from delayed rewards. Diss. University of Cambridge, 1989. Link

Implementation of the Particle Swarm Optimisation using Restart (RPSO) from Clerc, Maurice, and James Kennedy. "The particle swarm-explosion, stability, and convergence in a multidimensional complex space." Evolutionary Computation, IEEE Transactions on 6.1 (2002): 58-73.

Outputs

Oscilliating Environment

Large Bias Small Bias
Big Bias Small Bias

Cyclical Environment

Large Bias Small Bias
Big Bias Small Bias

Settings

Q-Learning and QBEA Parameter Settings

Setting Description Value
Discount factor Credit assigment discount 0.7
Epsilon Random chance for epsilon greedy exploration strategy 0.1
Q-table Learning policy [21:21]

Experiment Settings

Setting Description Value
Steps Number of time steps per run 1000
Repeat Number of repeat run for each experiment 100

Random seeds for reproducability

1245097796 1661198952 122864260 -1728364941 -1610161142
-1553747733 -1514202174 1222408005 -1578471556 521614943
-389764704 -1649559921 1994886919 2034262993 1507881027
-545858353 -654192656 1185726362 -1836349758 -1022557879
-552194205 -1345255965 -1203435676 -1341362130 -472864820
-2051027606 -565293299 -1779308806 1373421413 -1688259451
849821684 1298182941 -2055738353 903731815 -1166050407
-1822845219 -827989914 1621600417 1009567734 1778423930
-1967909361 1444884716 -922348556 1137581970 2025531731
2003346213 -611987680 -1528167225 -526263823 920625048
-720533346 -1858766946 -133249745 581003648 -1378875043
108191402 -1846334158 -349618738 805723531 659101161
-1792384625 -577449431 552042827 2144448365 -900817631
503529526 1025615013 -1254614466 832400668 -857677544
386874046 -370551938 2094200162 -1105797830 1372670024
1966474702 1334093332 457012299 -1189370565 842873584
804861548 240927114 -1726536252 1962410925 1409611978
-910950333 -142025090 -1624917511 -1102128116 247567512
-693307425 -1201487532 299085346 -480093052 505857268
1454893426 -1866304544 -604107491 -1102114480 1978324765