def runTrial(ds, iters):
kw = {
'minimize': True,
'verbose': True,
'mutationProb': 0.1,
'populationSize': 150,
'storeAllEvaluations': True,
'storeAllPopulations': True,
'elitism': True,
'tournament': False
}
e = setUpExperiment(ds, **kw)
ga = GA(testOnTrainData, e.nn, **kw)
e.results['trn_per_err'] = []
for i in range(iters):
print ga.learn(0)
e.results['trn_per_err'].append(ga._bestFound()[1])
fits = array(ga.fitnesses)
print '\t\t\tmean', mean(fits), 'std', std(fits), 'max', max(
fits), 'min', min(fits)
for i, t in trn_data:
o = ga._bestFound()[0].activate(i)
if where(o != t, True, False).any():
print i, o, argmax(t), '----'
e.results['tst_per_err'].append(
testOnTestData(ga._bestFound()[0], e.tst_data))
return e
def runTrial(iters):
kw = {
'minimize': True,
'verbose': True,
'mutationProb': 0.1,
'populationSize': 150,
'storeAllEvaluations': True,
'storeAllPopulations': True,
'elitism': True,
'tournament': False
}
e = Experiment(**kw)
# e.buildXor()
e.buildParity()
# e.buildIris()
# trainer = bp(e.nn, dataset=trn_data, verbose=True)
# trainer.trainEpochs(100)
ga = GA(testOnTrainData, e.nn, **kw)
# ga = GA(trn_data.evaluateModuleMSE, e.nn, **kw)
# tmp_nn = e.nn.newSimilarInstance()
perc_errs = []
for i in range(iters):
ga.learn(0)
perc_errs.append(ga._bestFound()[1])
fits = array(ga.fitnesses)
print '\t\t\tmean', mean(fits), 'std', std(fits), 'max', max(
fits), 'min', min(fits)
for i, t in trn_data:
o = ga._bestFound()[0].activate(i)
if o != t:
print i, o, t, '++' if o == t else '----'
# ranked_pop = ga._allGenerations[i]
# ranked_pop = array(sorted(ga._allGenerations[i], key=lambda x: x[1]))
# for ind in range(len(ranked_pop[0])):
# tmp_nn.params[:] = array(ranked_pop[0][ind]).copy()
# print ranked_pop[1][ind], testOnTrainData(tmp_nn), int(testOnTrainData(tmp_nn)) * '.'
# astrocytes = associateAstrocyteLayers(tmp_nn)
# trainNGA(tmp_nn, trn_data, astrocytes[0], astrocytes[1])
# ga.currentpop = new_pop
return perc_errs
def runTrial(ds, iters):
kw = {'minimize':True,
'verbose':True, 'mutationProb':0.1, 'populationSize':150,
'storeAllEvaluations':True, 'storeAllPopulations':True,
'elitism':True, 'tournament':False}
e = setUpExperiment(ds, **kw)
ga = GA(testOnTrainData, e.nn, **kw)
e.results['trn_per_err'] = []
for i in range(iters):
print ga.learn(0)
e.results['trn_per_err'].append(ga._bestFound()[1])
fits = array(ga.fitnesses)
print '\t\t\tmean', mean(fits), 'std', std(fits), 'max', max(fits), 'min', min(fits)
for i, t in trn_data:
o = ga._bestFound()[0].activate(i)
if where(o != t, True, False).any():
print i, o, argmax(t), '----'
e.results['tst_per_err'].append(testOnTestData(ga._bestFound()[0], e.tst_data))
return e
def runTrial(iters):
kw = {'minimize':True,
'verbose':True, 'mutationProb':0.1, 'populationSize':150,
'storeAllEvaluations':True, 'storeAllPopulations':True,
'elitism':True, 'tournament':False}
e = Experiment(**kw)
# e.buildXor()
e.buildParity()
# e.buildIris()
# trainer = bp(e.nn, dataset=trn_data, verbose=True)
# trainer.trainEpochs(100)
ga = GA(testOnTrainData, e.nn, **kw)
# ga = GA(trn_data.evaluateModuleMSE, e.nn, **kw)
# tmp_nn = e.nn.newSimilarInstance()
perc_errs = []
for i in range(iters):
ga.learn(0)
perc_errs.append(ga._bestFound()[1])
fits = array(ga.fitnesses)
print '\t\t\tmean', mean(fits), 'std', std(fits), 'max', max(fits), 'min', min(fits)
for i, t in trn_data:
o = ga._bestFound()[0].activate(i)
if o != t:
print i, o, t, '++' if o == t else '----'
# ranked_pop = ga._allGenerations[i]
# ranked_pop = array(sorted(ga._allGenerations[i], key=lambda x: x[1]))
# for ind in range(len(ranked_pop[0])):
# tmp_nn.params[:] = array(ranked_pop[0][ind]).copy()
# print ranked_pop[1][ind], testOnTrainData(tmp_nn), int(testOnTrainData(tmp_nn)) * '.'
# astrocytes = associateAstrocyteLayers(tmp_nn)
# trainNGA(tmp_nn, trn_data, astrocytes[0], astrocytes[1])
# ga.currentpop = new_pop
return perc_errs
