def testZeroIncrAndDecr(self): old_ws = [[x[0].copy(), x[1].copy()] for x in self.newpopW] new_ws = trainPP(self.newpopW, self.pop, incr=0, decr=0) for w1, w2 in zip(new_ws, old_ws): for io in range(2): npt.assert_array_equal(w1[io], w2[io]) self.assertIsNot(w1[io], w2[io])
def runTrial():
"""
Fresh population of NN individuals are evalutated, paired with
their fitness scores and ordered by fitness in descending order (fittest first)
"""
pop = createPop()
pairedPop = pairPop(pop)
rankedPop = sorted(pairedPop, key=itemgetter(-1), reverse=True)
all_results = {
'train_mses_top' : [],
'train_mses_avg' : [],
'train_percents_avg' : [],
'train_percents_top' : [],
'test_mse' : 0,
'test_percent' : 0
}
for iters in range(max_iterations):
# raw_input('==============NEXT==============')
if not iters % 1:
print 'Iteration'.rjust(120), iters
newpopW = evolveNewPop(rankedPop)
print '\n \nEVOLVE NEW POP'
rankedPop, top_mse, avg_mse, top_per, avg_per = rankPop(newpopW, pop, verbose=NN.pop_size)
if iters > 0 and iters < 1000:
newpopW = trainPP(rankedPop, pop, incr=0.25, decr=0.5)
print '\n \nTRAIN PP'
rankedPop, top_mse, avg_mse, top_per, avg_per = rankPop(newpopW, pop, verbose=NN.pop_size)
all_results['train_mses_top'].append(top_mse)
all_results['train_mses_avg'].append(avg_mse)
all_results['train_percents_top'].append(top_per)
all_results['train_percents_avg'].append(avg_per)
test_mse, test_per = testFittestInd(rankedPop)
all_results['test_mse'] = test_mse
all_results['test_percent'] = test_per
return all_results
def runTrial(iters):
"""
Fresh population of NN individuals are evalutated, paired with
their fitness scores and ordered by fitness in descending order (fittest first)
"""
pop = createPop()
pairedPop = pairPop(pop, verbose=False)
rankedPop = sorted(pairedPop, key=itemgetter(-1), reverse=True)
all_results = {
'train_mses_top': [],
'train_mses_avg': [],
'train_percents_avg': [],
'train_percents_top': [],
'test_mse': 0,
'test_percent': 0
}
for iters in range(iters):
# raw_input('==============NEXT==============')
if not iters % 1:
print 'Iteration'.rjust(120), iters
newpopW = evolveNewPop(rankedPop)
print '\n \nEVOLVE NEW POP'
rankedPop, top_mse, avg_mse, top_per, avg_per = rankPop(newpopW,
pop,
verbose=False)
print top_per
if iters > 500:
newpopW = trainPP(rankedPop, pop, incr=0.25, decr=0.5)
print '\n \nTRAIN PP'
rankedPop, top_mse, avg_mse, top_per, avg_per = rankPop(
newpopW, pop, verbose=False)
all_results['train_mses_top'].append(top_mse)
all_results['train_mses_avg'].append(avg_mse)
all_results['train_percents_top'].append(top_per)
all_results['train_percents_avg'].append(avg_per)
test_mse, test_per = testFittestInd(rankedPop)
all_results['test_mse'] = test_mse
all_results['test_percent'] = test_per
return all_results