def experiment1(self):
l = GA(self.fitnessFunction, self.myNetwork.params[self.indices])
l.minimize = False
l.verbose = True
l.maxLearningSteps = 500
params, fitness = l.learn()
self.myNetwork.params[self.indices] = params
self.metaInfo["numsteps"] = l.maxLearningSteps
self.metaInfo["fitness"] = fitness
# self.myNetwork._setParameters(self.originalWeights)
self.logNet()
def GA_minimize(function, guess):
result = GA(function,[guess, ], minimize=True) # set to minimize by default
start = timeit.default_timer()
mem = max(memory_usage((result.learn,(),)))
stop = timeit.default_timer()
print result.learn() #Comment this out for faster performance, i.e. if used purely for evaluation
exec_time = stop-start
print '{0} took {1} seconds'.format('Genetic Algorithm',exec_time)
print '{0} used {1} megabytes'.format('Genetic Algorithm',mem)
print
def GA_minimize(function, guess):
"""
This function runs the genetic algorithm from PyBrain (http://pybrain.org/docs/api/optimization/optimization.html) on a function, provided with an initial guess.
| NOTE: Format is **crucial** here or this will not work (blame the authors of PyBrain, sorry):
| Function must take a **tuple** (**Even** if of only one element (in the form (a, )) and guess **MUST BE A LIST** (even if only with one element)
"""
result = GA(function, guess, minimize=True) # set to minimize by default
start = timeit.default_timer()
mem = max(memory_usage(-1,interval=.1))
print "The result is: ", result.learn()
stop = timeit.default_timer()
exec_time = stop-start
print '{0} took {1} seconds'.format('Genetic Algorithm',exec_time)
print '{0} used {1} megabytes'.format('Genetic Algorithm',mem)
print
# -*- coding: utf-8 -*- """ Created on Thu Mar 7 20:41:24 2013 Just to try and get things working with a GA @author: david """ from pybrain.tools.shortcuts import buildNetwork from pybrain.structure.parametercontainer import ParameterContainer from pybrain.rl.environments.functions.unimodal import TabletFunction from pybrain.rl.environments.cartpole.balancetask import BalanceTask, CartPoleEnvironment from pybrain.optimization import GA from pybrain.rl.agents import LearningAgent, OptimizationAgent environment = CartPoleEnvironment() task = BalanceTask() nn = buildNetwork(task.outdim, 6, task.indim) learning_agent = OptimizationAgent(nn, GA())
def update():
"""Opens file browser and gets selection"""
name = tkFileDialog.askopenfilename()
"""Creates model for data"""
if 'xml' in name:
model = strainmodel(name, type='xml')
else:
model = strainmodel(name)
data = model.get_experimental_data()
"""Will need to be set to user-input guess"""
guess = [-150, 1]
yieldpoint = None
"""Gets radio button input as to which method to use to optimize"""
if yield_method.get() == 1:
yieldpoint = material_analytics.yield_stress_classic_unfitted(data)
elif yield_method.get() == 2:
yieldpoint = material_analytics.yield_stress_classic_fitted(data)
elif yield_method.get() == 3:
yieldpoint = material_analytics.yield_stress(data)
else:
elastic, plastic = material_analytics.kmeanssplit(data)
yieldpoint = plastic[0][None, ]
"""Displays the found yield stress"""
plot.plotmult2D(data,
yieldpoint,
title='File',
xtitle='Strain ($\epsilon$)',
ytitle='Stress ($\sigma$)')
"""[0,1] is the first row, second column, which is the stress values"""
SS_stress = yieldpoint[0, 1]
model_training_methods = [
'Nelder-Mead', 'Powell', 'CG', 'Newton-CG', 'BFGS', 'L-BFGS-B',
'SLSQP', 'COBYLA', 'TNC', 'Basinhopping', 'Brute Force',
'Genetic Algorithm'
]
"""Because basinhopping, brute force, and GA are all in separate libraries, they are handled as separate cases."""
optmethod = optimization_method.get() - 1
if optmethod < 9:
model_params = optimization_suite.minimize_suite(
model.mcfunc,
methods=[
model_training_methods[optmethod],
],
guess=guess,
SS_stress=SS_stress)
elif optmethod == 9:
model_params = optimization_suite.minimize_suite(model.mcfunc,
methods=[
basinhopping,
],
guess=guess,
SS_stress=SS_stress)
elif optmethod == 10:
model_params = optimization_suite.minimize_suite(model.mcfunc,
methods=[
brute,
],
guess=guess,
SS_stress=SS_stress)
else:
model_params = GA(model.mcfunc, guess, minimize=True).learn()
"""Plots the data versus the fitted irreversible model data"""
plot.plotmult2D(data,
model.irreversible_model(model_params, SS_stress),
title='Fitted Thermodynamics',
xtitle='Strain ($\epsilon$)',
ytitle='Stress ($\sigma$)')
def GA_Factory():
return GA()
for x in range(0,numberOfAgents):
popList.append(initPopulation)
for x in range(0,numberOfAgents):
weights = initPopulation[0]
NNetListParams.append(weights)
i = 0
while (i < 50):
for x in range(0,len(NNetListParams)):
for k in range(0,len(NNetListParams)):
NNetList[k]._setParameters(NNetListParams[k])
workingNN = x
ga = GA(evaluator,NNetListParams[x],maxEvaluations = 20,initRangeScaling = 2,elitism = False,populationSize = populationSize,initialPopulation = popList[x],mutationProb = 0.1)
ga.minimize = True
result = ga.learn()
NNetListParams[x] = result[0]
popList[x] = ga.currentpop
i += 1
print i
for x in range(0,len(NNetListParams)):
for k in range(0,len(NNetListParams)):
NNetList[k]._setParameters(NNetListParams[k])