Experiment 20210927-DOTG

Experiment design

Knowledge is transmitted between agents and across generations with different parent Selection methods.

Date: 20210927 (Yasser Bourahla)

10 runs; 200000 games

Setting : Agents start with random ontologies which are used to interact with the environment and other agents; agents adapt their knowledge to interaction failures (intra-generation knowledge transmission); agents provide training samples to their children to learn ontologies (inter-generation knowledge transmission)

Hypothesis: Intra-generation transmission can compensate to the absence of parent selection.

Variation of: 20210601-DOTG

Variables

independent variables: ['period', 'parent selection', 'adaptation']

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

Experiment

Date: 20210927 (Yasser Bourahla)

LazyLavender hash: 568897f29fce4e7148702b6475da3897eb28e360

Link to lazylavender

Parameter file: (params.sh)

Executed command (script.sh):

#!/bin/bash

. params.sh

CURRDIR=$(pwd)
OUTPUT=${CURRDIR}/${DIRPREF}
cd lazylav

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.generations.GenerationExperiment ${OPT}  -DresultDir=${OUTPUT}

Experimental plan

The independent variables have been varied as follows:

period: [5001, 10001, 20001]
parent selection: ['best', 'income', 'random']
adaptation: ['0', '1']

Raw results

Full results are available on zenodo

DOI DOI 10.5281/zenodo.5929553 10.5281/zenodo.5929553

Hypothesis testing

Intra-generation transmission can compensate to the absence of parent selection.

Confidence Intervals:

income - best [ -0.0281 , 0.054 ]
random - best [ -0.0378 , 0.0442 ]
random - income [ -0.0508 , 0.0312 ]
For period 5001
influence of parent selection
PR(>F) Significance
accuracy 0.890288 False

Hypothesis validated for period: 5001



For period 10001
influence of parent selection
PR(>F) Significance
accuracy 0.972216 False

Hypothesis validated for period: 10001



For period 20001
influence of parent selection
PR(>F) Significance
accuracy 0.340589 False

Hypothesis validated for period: 20001



Data exploration

Summary of data with adaptation

period parent selection
5001 10001 20001 best income random
genSrate 1 0.1000 0.1333 0.1333 0.1000 0.1333 0.1333
1000000 0.7802 0.8964 0.9428 0.8674 0.8758 0.8762
accuracy 1 0.1694 0.1641 0.1719 0.1695 0.1705 0.1655
1000000 0.8896 0.9279 0.9271 0.9095 0.9224 0.9127

Summary of data without adaptation

period parent selection
5001 10001 20001 best income random
genSrate 1 0.10 0.23 0.20 0.17 0.23 0.13
1000000 0.60 0.49 0.42 0.78 0.42 0.31
accuracy 1 0.17 0.16 0.17 0.16 0.17 0.17
1000000 0.26 0.25 0.27 0.33 0.29 0.16

Comparison of parent selections when there is no adaptation

Confidence Intervals:

income - best [ -0.0838 , -0.0129 ]
random - best [ -0.2046 , -0.1337 ]
random - income [ -0.1563 , -0.0854 ]
For period 5001
influence of parent selection
PR(>F) Significance
genSrate 0.000000 True
accuracy 0.000011 True


For period 10001
influence of parent selection
PR(>F) Significance
genSrate 0.000000 True
accuracy 0.000000 True


For period 20001
influence of parent selection
PR(>F) Significance
genSrate 0.000000 True
accuracy 0.000000 True