def rank(dataset, algorithm, spatial):
_, _, labels, _ = datatools.preparedata(dataset, order=spatial)
allfeatuers = [labels[x] for x in labels if x < 0]
data = [x for x in history if x['algorithm']==algorithm and x['spatial']==spatial
and datatools.compatible(datatools.metadata(dataset), x['training set'])
and x['best genome'] is not None and x['generations']>100] #gen>100 to exclude some test runs to debug problems
if len(data)==0:
print("No run found")
return {},0
ranked = {key : {f :0.0 for f in allfeatuers} for key in data[0]['training set']['genres']}
if spatial==1:
ranked['Output'] = {f :0.0 for f in allfeatuers} ## workaround
scoresum=0.0
for datum in data:
best = datum['best genome']
scoresum+= datum['control score'] ** 2
for i in range(0, datum['spatial']):
crunch_output_node(i, ranked[labels[i]], best, labels)
for k in ranked:
for f in ranked[k]:
ranked[k][f] *= datum['control score']**2
for k in ranked:
for f in ranked[k]:
ranked[k][f] /= scoresum
if spatial == 1:
meta = datatools.metadata(dataset)
mapping = datatools.generalMapping(meta['genres'], spatial)
for key in ranked:
if key != 'Output':
for f in ranked[key]:
ranked[key][f] = ranked['Output'][f] * mapping[key][0] if mapping[key][0]>0 else -ranked['Output'][f]
ranked.pop('Output',None)
maximum = max([x for genre in ranked for x in ranked[genre].values()])
minimum = min([x for genre in ranked for x in ranked[genre].values()])
for genre in ranked:
for f in ranked[genre]:
ranked[genre][f] = (ranked[genre][f])/(maximum - minimum)
return ranked, len(data)
def bestPerformer(dataset):
m = datatools.metadata(dataset)
data = [x for x in history if datatools.compatible(x['training set'], m)]
best = (1,None)
for d in data:
if 1.0*len(d['control errors'])/d['control set']['size'] < best[0]:
best =(1.0*len(d['control errors'])/d['control set']['size'] ,d)
return best
def averagePerformance(dataset, algorithm, spatial):
m = datatools.metadata(dataset)
data = [x for x in history if datatools.compatible(x['training set'], m) and x['algorithm']==algorithm and x['spatial']==spatial
and x['generations']>100]
total=0
for x in data:
total+= 1.0*len(x['control errors'])/x['control set']['size']
return 1.0 - total/len(data) if len(data)>0 else 0
def rank(dataset, algorithm, spatial):
data = [x for x in history if x['algorithm']==algorithm and x['spatial']==spatial
and datatools.compatible(datatools.metadata(dataset), x['training set'])
and x['best genome'] is not None]
if len(data)==0:
print("No run found")
return {},0
ranked = {key : {f :0.0 for f in allfeatuers} for key in data[0]['training set']['genres']}
ranked['Output'] = {f :0.0 for f in allfeatuers} ## workaround
scoresum=0.0
for datum in data:
best = datum['best genome']
scoresum+= datum['control score']
for i in range(0, spatial):
crunch_output_node(i, ranked[labels[i]], best)
for k in ranked:
for f in ranked[k]:
ranked[k][f] *= datum['control score']
for k in ranked:
for f in ranked[k]:
ranked[k][f] /= scoresum
if spatial == 1:
#todo: ammesso che i generi siano 2
#todo: prendi ranked['Output']
#todo: associa al genere[0] l'opposto dei valori, e al genere[1] i valori originari
#todo: con più di 2 generi???????
pass
return ranked, len(data)
'./Datasets/classic-jazz-rock.dat', './Datasets/classic-rock.dat',
'./Datasets/classic-jazz.dat', './Datasets/jazz-rock.dat'
]
#for d,a,s in itertools.product(datasets, algorithms, spatials):
# print(d,a,'(spatial:',s,') ',len([x for x in history if x['algorithm']==a and x['spatial']==s and x['training set']['path']==d]))
"""for d in [x for x in history if x['training set']['path']==datasets[1] and x['spatial']==2]:
print('\n\n################################', d['generations'],'algorithm:', d['algorithm'])
datatools.showErrors(d)
exit(0)"""
for d, a, s in itertools.product(datasets, algorithms, spatials):
values = [
x['control score'] for x in history
if x['algorithm'] == a and x['spatial'] == s
and datatools.compatible(datatools.metadata(d), x['training set'])
]
if len(values) > 0:
print(d, a, s, ' \t', len(values), ':',
averagePerformance(d, a, s),
sum(values) / len(values))
for x in itertools.product(datasets, algorithms, spatials):
plotRank(x[0], x[1], x[2])
#showrank(x[0],x[1],x[2])
#for d in datasets:
# x = bestPerformer(d)
# print(d,':',x[0],x[1]['algorithm'], '(spatial:',x[1]['spatial'],')', x[1]['generations'], x[1]['control score'], len(x[1]['control errors']))