Experiment 20220302-DOLA

Experiment design

Agent adapt ontologies to agree on decision taking. Introducing conformity index for agents to consider before adapting.

Date: 20220302 (Yasser Bourahla)

5 runs; 100000 games

Setting : Agents learn decision trees (transformed into ontologies); get income from environment; adapt by splitting their leaf nodes

Hypothesis: Success rate converges to 1. Improve the average accuracy at the end of the experiment.

Variation of: 20200623-DOLA

Variables

independent variables: ['Environment bias coefficient', 'Social bias coefficient', 'Frequency of adaptation']

dependent variables: ['success rate', 'accuracy']

Experiment

Date: 20220302 (Yasser Bourahla)

LazyLavender hash: 4b837b2619086143a5bbb798ac0b80b110f80342

Link to lazylavender

Parameter file: (params.sh)

Executed command (script.sh):

#!/bin/bash

. params.sh

CURRDIR=$(pwd)
OUTPUT=${CURRDIR}/${DIRPREF}
# cd ${LLPATH}
cd lazylav
# this sample runs ExperimentalPlan. It can be replaced with Monitor if parameters are not varied.
bash scripts/runexp.sh -p ${CURRDIR} -d ${DIRPREF} java -Dlog.level=INFO -cp ${JPATH} fr.inria.exmo.lazylavender.engine.ExperimentalPlan -Dexperiment=fr.inria.exmo.lazylavender.decisiontaking.Experiment ${OPT} -DresultDir=${OUTPUT}

Experimental plan

The independent variables have been varied as follows:

Environment bias coefficient: [0.0, 1.0]
Social bias coefficient: [0.0, 0.3, 0.7, 1.0]
Frequency of adaptation: [0.05, 1.0]

Hypothesis testing

Success rate converges to 1. Improve the average accuracy at the end of the experiment.

Data exploration

Summary of data with adaptation

Environment bias coefficient Social bias coefficient Frequency of adaptation
0.0 1.0 0.0 0.3 0.7 1.0 0.05 1.0
ssrate 1 0.4000 0.5000 0.5000 0.3500 0.3500 0.6000 0.4750 0.4250
1000000 0.9155 0.9412 0.8355 0.9406 0.9697 0.9675 0.8670 0.9897
accuracy 1 0.4583 0.4721 0.4495 0.4498 0.4936 0.4678 0.4625 0.4679
1000000 0.4112 0.7965 0.5814 0.5953 0.6375 0.6011 0.6045 0.6031

Results with adaptation frequency at 0.05

for environment success index weight = 0
coeffSoc
0.0000 0.3000 0.7000 1.0000
ssrate 0 0.6000 0.0000 0.4000 0.6000
99999 0.5621 0.8774 0.9747 0.9620
accuracy 0 0.4113 0.4313 0.4712 0.4900
99999 0.3162 0.3981 0.5125 0.5000
for environment success index weight = 1
coeffSoc
0.0000 0.3000 0.7000 1.0000
ssrate 0 0.8000 0.6000 0.4000 0.4000
99999 0.8153 0.9030 0.9157 0.9256
accuracy 0 0.4637 0.4600 0.5012 0.4712
99999 0.8969 0.7331 0.7875 0.6919

The success rate appears to benefit from conformity bias.

The success bias is necessary to improve accuracy.

Compare how spread the decision for each rarity bias weight when success bias weight = 1

Social bias coefficient
0.0 0.3 0.7 1.0
loss 1000000 1.6000 4.2000 3.4000 4.8000
winDspread 1 10.6807 11.3187 11.1831 12.3754

ANOVA Results when success bias weight = 1:

influence of Social bias coefficient
PR(>F) Significance
ssrate 0.097651 False
accuracy 0.112512 False 






for ssrate
group1 group2 meandiff p-adj lower upper reject
0 0 0.3 0.0877 0.2536 -0.0420 0.2174 False
1 0 0.7 0.1003 0.1619 -0.0294 0.2301 False
2 0 1 0.1102 0.1109 -0.0195 0.2400 False
3 0.3 0.7 0.0127 0.9000 -0.1171 0.1424 False
4 0.3 1 0.0226 0.9000 -0.1071 0.1523 False
5 0.7 1 0.0099 0.9000 -0.1198 0.1396 False 
for accuracy
group1 group2 meandiff p-adj lower upper reject
0 0 0.3 -0.1637 0.2316 -0.3989 0.0714 False
1 0 0.7 -0.1094 0.5532 -0.3445 0.1257 False
2 0 1 -0.2050 0.0990 -0.4401 0.0301 False
3 0.3 0.7 0.0544 0.9000 -0.1807 0.2895 False
4 0.3 1 -0.0413 0.9000 -0.2764 0.1939 False
5 0.7 1 -0.0956 0.6434 -0.3307 0.1395 False