def test_gridminimise_return():
from milksets.wine import load
features,labels = load()
learner = fast_classifier()
gridminimise(learner, features, labels, { 'ignore' : [0] })
_,error = gridminimise(learner, features, labels, { 'ignore' : [0] }, return_value=True, nfolds=5)
cmat,_ = milk.nfoldcrossvalidation(features, labels, learner=learner, nfolds=5)
assert error == cmat.sum()-cmat.trace()
def test_gridminimise_return():
from milksets.wine import load
features, labels = load()
learner = fast_classifier()
gridminimise(learner, features, labels, {'ignore': [0]})
_, error = gridminimise(learner,
features,
labels, {'ignore': [0]},
return_value=True,
nfolds=5)
cmat, _ = milk.nfoldcrossvalidation(features,
labels,
learner=learner,
nfolds=5)
assert error == cmat.sum() - cmat.trace()
def test_gridminimise():
features = np.arange(100)
labels = np.tile((0, 1), 50)
paramspace = {
'a': np.arange(4),
'b': np.arange(-3, 3),
'c': np.linspace(2., 10)
}
best, value = gridminimise(simple_learner(),
features,
labels,
paramspace,
measure=(lambda _, p: p[0]),
return_value=True)
best = dict(best)
val = f(best['a'], best['b'], best['c'])
assert value == val * 100
for a in np.arange(4):
for b in np.arange(-3, 3):
for c in np.linspace(2., 10):
assert val <= f(a, b, c)
gs = gridsearch(simple_learner(),
paramspace,
measure=(lambda _, p: p[0]),
annotate=True)
model = gs.train(features, labels)
assert model.value == value
assert model.arguments == val
def test_gridminimise():
from milksets.wine import load
features, labels = load()
x = gridminimise(milk.supervised.svm_simple(kernel=np.dot, C=2.),
features[::2], labels[::2] == 0, {'C': (0.5, )})
cval, = x
assert cval == ('C', .5)
def test_gridminimise():
features = np.arange(100)
labels = np.tile((0,1), 50)
best = gridminimise(simple_learner(), features, labels, {'a': np.arange(4), 'b' : np.arange(-3,3), 'c' : np.linspace(2., 10) }, measure=(lambda _, p: p[0]))
best = dict(best)
val = f(best['a'], best['b'], best['c'])
for a in np.arange(4):
for b in np.arange(-3,3):
for c in np.linspace(2., 10):
assert val <= f(a,b,c)
def test_gridminimise():
features = np.arange(100)
labels = np.tile((0,1), 50)
paramspace = { 'a': np.arange(4), 'b' : np.arange(-3,3), 'c' : np.linspace(2., 10) }
best,value = gridminimise(simple_learner(), features, labels, paramspace, measure=(lambda _, p: p[0]), return_value=True)
best = dict(best)
val = f(best['a'], best['b'], best['c'])
assert value == val*100
for a in np.arange(4):
for b in np.arange(-3,3):
for c in np.linspace(2., 10):
assert val <= f(a,b,c)
gs = gridsearch(simple_learner(), paramspace, measure=(lambda _, p: p[0]), annotate=True)
model = gs.train(features, labels)
assert model.value == value
assert model.arguments == val
def train(self, features, labels, **kwargs):
from milk.supervised.gridsearch import gridminimise
c_features = np.concatenate([f for f, _ in features if len(f)])
c_features = c_features[::self.sample]
nprocs = parallel.get_procs(use_current=True)
tow = multiprocessing.Queue()
fromw = multiprocessing.Queue()
for k in self.ks:
for ri in xrange(self.rmax):
tow.put((k, ri))
for i in xrange(nprocs):
tow.put(('shutdown', None))
workers = [
kmeans_cluster(c_features, tow, fromw) for i in xrange(nprocs)
]
for w in workers:
if nprocs > 1:
w.start()
else:
w.execute()
try:
codebooks = [fromw.get() for i in xrange(len(self.ks) * self.rmax)]
finally:
tow.close()
tow.join_thread()
if nprocs > 1:
for w in workers:
w.join()
parallel.release_procs(len(workers), count_current=True)
base = codebook_learner(self.base)
base.normalise = self.normalise
if len(codebooks) > 1:
(best, ) = gridminimise(base,
features,
labels, {'codebook': codebooks},
nfolds=self.nfolds)
_, codebook = best
else:
(codebook, ) = codebooks
base.codebook = codebook
return base.train(features, labels)
def train(self, features, labels, **kwargs):
from milk.supervised.gridsearch import gridminimise
c_features = np.concatenate([f for f, _ in features if len(f)])
c_features = c_features[:: self.sample]
nprocs = parallel.get_procs(use_current=True)
tow = multiprocessing.Queue()
fromw = multiprocessing.Queue()
for k in self.ks:
for ri in xrange(self.rmax):
tow.put((k, ri))
for i in xrange(nprocs):
tow.put(("shutdown", None))
workers = [kmeans_cluster(c_features, tow, fromw) for i in xrange(nprocs)]
for w in workers:
if nprocs > 1:
w.start()
else:
w.execute()
try:
codebooks = [fromw.get() for i in xrange(len(self.ks) * self.rmax)]
finally:
tow.close()
tow.join_thread()
if nprocs > 1:
for w in workers:
w.join()
parallel.release_procs(len(workers), count_current=True)
base = codebook_learner(self.base)
base.normalise = self.normalise
if len(codebooks) > 1:
(best,) = gridminimise(base, features, labels, {"codebook": codebooks}, nfolds=self.nfolds)
_, codebook = best
else:
(codebook,) = codebooks
base.codebook = codebook
return base.train(features, labels)
def test_gridminimise():
from milksets.wine import load
features, labels = load()
x = gridminimise(milk.supervised.svm_simple(kernel=np.dot, C=2.), features[::2], labels[::2] == 0, {'C' : (0.5,) })
cval, = x
assert cval == ('C', .5)
