def test_multiclass_encoding():
#X, y = make_classification(n_samples=100, n_features=20,
n_samples = 10
n_classes = 3
X, y = make_classification(n_samples=n_samples, n_features=10,
n_informative=6, n_redundant=2, n_repeated=0,
n_classes=n_classes, n_clusters_per_class=2,
weights=None, flip_y=0.01, class_sep=1.0,
hypercube=True, shift=0.0, scale=1.0,
shuffle=True, random_state=None)
rows = np.arange(len(y))
cols = y
values = np.ones_like(y, dtype=np.float64)
y_one_hot = sp.coo_matrix((values, (rows, cols)),
shape=(n_samples, n_classes), dtype=np.float64)
y_one_hot = y_one_hot.tocsr()
X = sp.csr_matrix(X)
X_ext, compars = multiclass_to_ranking(X, y_one_hot)
fm = FMRecommender(n_iter=2000,
init_stdev=0.01, l2_reg_w=.5, l2_reg_V=.5, rank=2,
step_size=.002, random_state=11)
X_train = X_ext.tocsc()
#assert False
fm.fit(X_train, compars)
def test_multiclass_to_ranking():
X = sp.csr_matrix(np.arange(6).reshape((3,2)))
y = sp.csr_matrix((3, 5))
y[0, 0] = 1
y[1, [2, 3]] = 1
y[2, [0, 4]] = 1
n_classes = y.shape[1]
n_samples = X.shape[0]
n_features = X.shape[1]
X_ext, compars = multiclass_to_ranking(X, y)
assert X_ext.shape[0] == n_classes * n_samples
assert X_ext.shape[1] == n_classes + n_features
# test that features are replicated
assert_array_equal(X_ext.tocsc()[:, n_classes:].sum(axis=0),
X.sum(axis=0) * n_classes)
# test class labels encoding structure
assert X_ext.tocsr()[:n_classes, :].sum(axis=0)[0, 0] == n_samples
assert X_ext.tocsr()[:, :n_classes].sum() == n_samples * n_classes
#assert_array_equal(X_ext.tocsc()[:, n_classes:].sum(axis=0),
print X_ext.todense()
print y.todense()
print compars
def test_multiclass_to_ranking():
X = sp.csr_matrix(np.arange(6).reshape((3, 2)))
y = sp.csr_matrix((3, 5))
y[0, 0] = 1
y[1, [2, 3]] = 1
y[2, [0, 4]] = 1
n_classes = y.shape[1]
n_samples = X.shape[0]
n_features = X.shape[1]
X_ext, compars = multiclass_to_ranking(X, y)
assert X_ext.shape[0] == n_classes * n_samples
assert X_ext.shape[1] == n_classes + n_features
# test that features are replicated
assert_array_equal(X_ext.tocsc()[:, n_classes:].sum(axis=0),
X.sum(axis=0) * n_classes)
# test class labels encoding structure
assert X_ext.tocsr()[:n_classes, :].sum(axis=0)[0, 0] == n_samples
assert X_ext.tocsr()[:, :n_classes].sum() == n_samples * n_classes
#assert_array_equal(X_ext.tocsc()[:, n_classes:].sum(axis=0),
print X_ext.todense()
print y.todense()
print compars
def test_multiclass_encoding():
#X, y = make_classification(n_samples=100, n_features=20,
n_samples = 10
n_classes = 3
X, y = make_classification(n_samples=n_samples,
n_features=10,
n_informative=6,
n_redundant=2,
n_repeated=0,
n_classes=n_classes,
n_clusters_per_class=2,
weights=None,
flip_y=0.01,
class_sep=1.0,
hypercube=True,
shift=0.0,
scale=1.0,
shuffle=True,
random_state=None)
rows = np.arange(len(y))
cols = y
values = np.ones_like(y, dtype=np.float64)
y_one_hot = sp.coo_matrix((values, (rows, cols)),
shape=(n_samples, n_classes),
dtype=np.float64)
y_one_hot = y_one_hot.tocsr()
X = sp.csr_matrix(X)
X_ext, compars = multiclass_to_ranking(X, y_one_hot)
fm = FMRecommender(n_iter=2000,
init_stdev=0.01,
l2_reg_w=.5,
l2_reg_V=.5,
rank=2,
step_size=.002,
random_state=11)
X_train = X_ext.tocsc()
#assert False
fm.fit(X_train, compars)
