def test_iris(self):
num_constraints = 200
n = self.iris_points.shape[0]
C = ITML.prepare_constraints(self.iris_labels, n, num_constraints)
itml = ITML(self.iris_points, C)
itml.fit(verbose=False)
csep = class_separation(itml.transform(), self.iris_labels)
self.assertLess(csep, 0.4) # it's not great
def sandwich_demo():
x, y = sandwich_data()
knn = nearest_neighbors(x, k=2)
ax = pyplot.subplot(3, 1, 1) # take the whole top row
plot_sandwich_data(x, y, ax)
plot_neighborhood_graph(x, knn, y, ax)
ax.set_title('input space')
ax.set_aspect('equal')
ax.set_xticks([])
ax.set_yticks([])
num_constraints = 60
mls = [
LMNN(x, y),
ITML(x, ITML.prepare_constraints(y, len(x), num_constraints)),
SDML(x, SDML.prepare_constraints(y, len(x), num_constraints)),
LSML(x, LSML.prepare_constraints(y, num_constraints))
]
for ax_num, ml in zip(xrange(3,7), mls):
ml.fit()
tx = ml.transform()
ml_knn = nearest_neighbors(tx, k=2)
ax = pyplot.subplot(3,2,ax_num)
plot_sandwich_data(tx, y, ax)
plot_neighborhood_graph(tx, ml_knn, y, ax)
ax.set_title('%s space' % ml.__class__.__name__)
ax.set_xticks([])
ax.set_yticks([])
pyplot.show()
def applyAlgo(algo, p, Xtrain, Ytrain, Xtest, Ytest):
if a.startswith("o"):
nbMinority = len(Xtrain[Ytrain == minClass])
if nbMinority <= 5:
sm = SMOTE(random_state=42, k_neighbors=nbMinority-1)
else:
sm = SMOTE(random_state=42)
Xtrain2, Ytrain2 = sm.fit_sample(Xtrain, Ytrain)
elif a.startswith("u"):
rus = RandomUnderSampler(random_state=42)
Xtrain2, Ytrain2 = rus.fit_sample(Xtrain, Ytrain)
else:
Xtrain2, Ytrain2 = Xtrain, Ytrain
scaler = StandardScaler()
scaler.fit(Xtrain2)
Xtrain2 = scaler.transform(Xtrain2)
Xtest = scaler.transform(Xtest)
if algo.endswith("IML"):
ml = IML(pClass=minClass, k=K, m=p["m"], Lambda=p["Lambda"], a=p["a"])
elif algo.endswith("LMNN"):
ml = LMNN(k=K, mu=p["mu"], randomState=np.random.RandomState(1))
elif algo.endswith("GMML"):
ml = GMML(t=p["t"], randomState=np.random.RandomState(1))
elif algo.endswith("ITML"):
ml = ITML(gamma=p["gamma"], randomState=np.random.RandomState(1))
if not algo.endswith("Euclidean"):
ml.fit(Xtrain2, Ytrain2)
Xtrain2 = ml.transform(Xtrain2)
Xtest = ml.transform(Xtest)
# Apply kNN to predict classes of test examples
Ytrain_pred = knnSame(K, Xtrain2, Ytrain2)
Ytest_pred = knn(K, Xtrain2, Ytrain2, Xtest)
perf = {}
for true, pred, name in [(Ytrain2, Ytrain_pred, "train"),
(Ytest, Ytest_pred, "test")]:
# Compute performance measures by comparing prediction with true labels
tn, fp, fn, tp = confusion_matrix(true, pred,
labels=[majClass, minClass]).ravel()
perf[name] = ((int(tn), int(fp), int(fn), int(tp)))
return perf
def applyAlgo(algo, p, Xtrain, Ytrain, Xtest, Ytest):
Xtrain2, Ytrain2 = Xtrain, Ytrain
scaler = StandardScaler()
scaler.fit(Xtrain2)
Xtrain2 = scaler.transform(Xtrain2)
Xtest = scaler.transform(Xtest)
if algo == "IML":
ml = IML(pClass=minClass, k=K, m=p["m"], Lambda=p["Lambda"], a=p["a"])
elif algo == "ML2":
ml = ML2(pClass=minClass, k=K, m=p["m"], Lambda=p["Lambda"], a=p["a"])
elif algo == "ML1":
ml = ML1(num_const=2*K*len(Xtrain),
m=p["m"], Lambda=p["Lambda"], a=p["a"])
elif algo == "LMNN":
ml = LMNN(k=K, mu=p["mu"], randomState=np.random.RandomState(1))
elif algo == "GMML":
ml = GMML(t=p["t"], randomState=np.random.RandomState(1))
elif algo == "ITML":
ml = ITML(gamma=p["gamma"], randomState=np.random.RandomState(1))
if not algo.endswith("Euclidean"):
ml.fit(Xtrain2, Ytrain2)
Xtrain2 = ml.transform(Xtrain2)
Xtest = ml.transform(Xtest)
# Apply kNN to predict classes of test examples
Ytrain_pred = knnSame(K, Xtrain2, Ytrain2)
Ytest_pred = knn(K, Xtrain2, Ytrain2, Xtest)
perf = {}
for true, pred, name in [(Ytrain2, Ytrain_pred, "train"),
(Ytest, Ytest_pred, "test")]:
# Compute performance measures by comparing prediction with true labels
tn, fp, fn, tp = confusion_matrix(true, pred,
labels=[majClass, minClass]).ravel()
perf[name] = ((int(tn), int(fp), int(fn), int(tp)))
return perf
