This is a general framework for carrying out evolution in a serverless framework. It runs, in principle, in a Bluemix environment, but this is actually vendor implementation of open source frameworks such as Kafka and OpenWhisk.
Please install requirements.txt to test main()
Input parameters example:
args= {'id': str(uuid.uuid1()),
'problem': {
'name': 'BBOB',
'function': 'FUNCTION' in os.environ and int(os.environ['FUNCTION']) or 3,
'instance': 'INSTANCE' in os.environ and int(os.environ['INSTANCE']) or 1,
'search_space': [-5, 5],
'dim': 'DIM' in os.environ and int(os.environ['DIM']) or 3,
'error': 1e-8
},
'population': [],
'population_size':'POPULATION_SIZE' in os.environ and int(os.environ['POPULATION_SIZE']) or 1000,
'experiment':
{
'experiment_id': 'dc74efeb-9d64-11e7-a2bd-54e43af0c111',
'owner': 'mariosky',
'type': 'benchmark'
},
'algorithm': {
'name': 'GA',
'iterations': 2000,
'selection': {
'type': 'tools.selTournament',
'tournsize': 12
},
'crossover': {'type': 'cxTwoPoint',
'CXPB': [0, .2]
},
'mutation': {'type': 'mutGaussian',
'mu': 0,
'sigma': 0.5,
'indpb' : 0.05,
'MUTPB':0.5
}
}
}Output is the same as input with a new evolved population plus:
{
'diversity' :
{
'measure':'euclidean',
'diversity':123
}
'iterations':[
(0, -425.21456490345463, [-0.4316661180759782, 3.9023599854361706, 1.4169751721735144], 258),
(1, -446.15104470081553, [-4.5856156823256145, 2.271772581290935, 1.6631995281618686], 261)
],
'best': False,
'best_individual':[-4.5856156823256145, 2.271772581290935, 1.6631995281618686],
'fopt':-462.09
}
Iterations Tuple:
- Number of iteration
- Best Fitness
- Best Individual
- Number of Function Evaluations
First you log in to your vm
vagrant sshClone the source
git clone https://github.com/mariosky/ga_actionChange directory to ga_service
cd ga_serviceGenerate the vitualenv directory
docker run --rm -v "$PWD:/tmp" openwhisk/python2action sh -c "cd tmp; virtualenv virtualenv; source virtualenv/bin/activate; pip install -r requirements.txt;"Create the zip files with files needed
zip -r ga_service.zip __main__.py virtualenv ga_service.py bbobbenchmarks.pyCreate or Update the action
wsk action create gaService --kind python:2 ga_service.zip
wsk action update gaService --kind python:2 ga_service.zipGet the url
wsk action get gaService --url
ok: got action gaService
https://192.168.33.13/api/v1/namespaces/guest/actions/gaServiceGet the auth data
wsk property get --auth
whisk auth 23bc46b1-71f6-4ed5-8c54-816aa4f8c502:123zO3xZCLrMN6v2BKK1dXYFpXlPkccOFqm12CdAsMgRU4VrNZ9lyGVCGuMDGIwPFor updates with out new virtualenv
git fetch- Again clone the code
- Create your virtualenv, pip install the requests library (only needed for client)
- Edit your ga_client
If the --only-population flag is set then it returns an array of objects which represent each individual with it's id, chromosome and fitness.
[{'id': None, 'chromosome': [-3.0208012286451758, -1.9901786287762588, 0.4746940069424692], 'fitness': {'DefaultContext': -328.99633706375687, 'score': -328.99633706375687}}, {'id': None, 'chromosome': [-3.0208012286451758, -1.9901786287762588, 0.4746940069424692], 'fitness': {'DefaultContext': -328.99633706375687, 'score': -328.99633706375687}}]If the --only-population flag is not set then it returns an array of objects which contain the following keys:
[u'fopt', u'algorithm', u'best_individual', u'experiment', u'population_size', u'iterations', u'problem', u'id', u'best', u'population']The results of each iteration look like this:
[0, -422.3189885289348, [-2.3750799310197834, 0.26363257691378017, 1.9725786083712151], 17] -462.09
[1, -450.63219848002547, [-2.3750799310197834, 2.8617458011276007, 1.9725786083712151], 16] -462.09
[2, -450.63219848002547, [-2.3750799310197834, 2.8617458011276007, 1.9725786083712151], 20] -462.09
[3, -450.63219848002547, [-2.3750799310197834, 2.8617458011276007, 1.9725786083712151], 18] -462.09
[4, -450.7095245606381, [-2.3090028418986184, 2.8617458011276007, 1.9725786083712151], 15] -462.09[iteration number, best fitness, best solution, number of FE] function optima
- Move the algorithm parameters to a different json file?
- Regis logs.
- Update documentation.
- Refactor.
bx wsk action create ga_service --kind python:2 ga_service.zip