def test_zeros():
tensor = zeros((2, 3, 4))
assert len(list(tensor)) == 2
assert tensor.op.gpu is False
tensor2 = zeros((2, 3, 4), chunk_size=1)
# tensor's op key must be equal to tensor2
assert tensor.op.key == tensor2.op.key
assert tensor.key != tensor2.key
tensor3 = zeros((2, 3, 3))
assert tensor.op.key != tensor3.op.key
assert tensor.key != tensor3.key
# test create chunk op of zeros manually
chunk_op1 = TensorZeros(dtype=tensor.dtype)
chunk1 = chunk_op1.new_chunk(None, shape=(3, 3), index=(0, 0))
chunk_op2 = TensorZeros(dtype=tensor.dtype)
chunk2 = chunk_op2.new_chunk(None, shape=(3, 4), index=(0, 1))
assert chunk1.op.key != chunk2.op.key
assert chunk1.key != chunk2.key
tensor = zeros((100, 100), chunk_size=50)
tensor = tile(tensor)
assert len({c.op.key for c in tensor.chunks}) == 1
assert len({c.key for c in tensor.chunks}) == 1
def testZeros(self):
tensor = zeros((2, 3, 4))
self.assertEqual(len(list(tensor)), 2)
self.assertFalse(tensor.op.gpu)
tensor2 = zeros((2, 3, 4), chunk_size=1)
# tensor's op key must be equal to tensor2
self.assertEqual(tensor.op.key, tensor2.op.key)
self.assertNotEqual(tensor.key, tensor2.key)
tensor3 = zeros((2, 3, 3))
self.assertNotEqual(tensor.op.key, tensor3.op.key)
self.assertNotEqual(tensor.key, tensor3.key)
# test create chunk op of zeros manually
chunk_op1 = TensorZeros(dtype=tensor.dtype)
chunk1 = chunk_op1.new_chunk(None, shape=(3, 3), index=(0, 0))
chunk_op2 = TensorZeros(dtype=tensor.dtype)
chunk2 = chunk_op2.new_chunk(None, shape=(3, 4), index=(0, 1))
self.assertNotEqual(chunk1.op.key, chunk2.op.key)
self.assertNotEqual(chunk1.key, chunk2.key)
tensor = zeros((100, 100), chunk_size=50)
tensor = tensor.tiles()
self.assertEqual(len({c.op.key for c in tensor.chunks}), 1)
self.assertEqual(len({c.key for c in tensor.chunks}), 1)
def testRuntimeError(self):
with option_context({'eager_mode': True}):
a = mt.zeros((10, 10))
with self.assertRaises(ValueError):
op = TileFailOp()
b = op.new_tileable(None)
b.build_graph(tiled=True)
r = a + 1
self.assertIn(repr(np.zeros((10, 10)) + 1), repr(r))
np.testing.assert_array_equal(r.fetch(), np.zeros((10, 10)) + 1)
a = mt.zeros((10, 10))
# a's data is used by r,
# if not deleted, a's data would not be gc collected
del r
with self.assertRaises(ValueError):
a.fetch()
def testRuntimeError(self):
from mars.utils import kernel_mode
@kernel_mode
def raise_error(*_):
raise ValueError
with option_context({'eager_mode': True}):
a = mt.zeros((10, 10))
with self.assertRaises(ValueError):
raise_error(a)
r = a + 1
self.assertIn(repr(np.zeros((10, 10)) + 1), repr(r))
np.testing.assert_array_equal(r.fetch(), np.zeros((10, 10)) + 1)
a = mt.zeros((10, 10))
with self.assertRaises(ValueError):
a.fetch()
def testRuntimeError(self):
with option_context({'eager_mode': True}):
a = mt.zeros((10, 10))
with self.assertRaises(ValueError):
op = TileFailOp()
b = op.new_tileable(None)
b.build_graph(tiled=True)
r = a + 1
self.assertIn(repr(np.zeros((10, 10)) + 1), repr(r))
np.testing.assert_array_equal(r.fetch(), np.zeros((10, 10)) + 1)
def test_pca_deterministic_output(self):
rng = np.random.RandomState(0)
X = mt.tensor(rng.rand(10, 10))
for solver in self.solver_list:
transformed_X = mt.zeros((20, 2))
for i in range(20):
pca = PCA(n_components=2, svd_solver=solver, random_state=rng)
transformed_X[i, :] = pca.fit_transform(X)[0]
np.testing.assert_allclose(
transformed_X.execute(), mt.tile(transformed_X[0, :], 20).reshape(20, 2).execute())
def test_min_max_scaler1d(setup):
X_list_1row = X_1row.to_numpy().tolist()
X_list_1col = X_1col.to_numpy().tolist()
# Test scaling of dataset along single axis
for X in [X_1row, X_1col, X_list_1row, X_list_1col]:
scaler = MinMaxScaler(copy=True)
X_scaled = scaler.fit(X).transform(X)
if isinstance(X, list):
X = mt.array(X) # cast only after scaling done
if _check_dim_1axis(X) == 1:
assert_array_almost_equal(X_scaled.min(axis=0),
mt.zeros(n_features))
assert_array_almost_equal(X_scaled.max(axis=0),
mt.zeros(n_features))
else:
assert_array_almost_equal(X_scaled.min(axis=0), .0)
assert_array_almost_equal(X_scaled.max(axis=0), 1.)
assert scaler.n_samples_seen_ == X.shape[0]
# check inverse transform
X_scaled_back = scaler.inverse_transform(X_scaled)
assert_array_almost_equal(X_scaled_back, X)
# Constant feature
X = mt.ones((5, 1))
scaler = MinMaxScaler()
X_scaled = scaler.fit(X).transform(X)
assert X_scaled.min().to_numpy() >= 0.
assert X_scaled.max().to_numpy() <= 1.
assert scaler.n_samples_seen_ == X.shape[0]
# Function interface
X_1d = X_1row.ravel()
min_ = X_1d.min()
max_ = X_1d.max()
assert_array_almost_equal((X_1d - min_) / (max_ - min_),
minmax_scale(X_1d, copy=True))
def test_pca_score3(setup):
# Check that probabilistic PCA selects the right model
n, p = 200, 3
rng = np.random.RandomState(0)
Xl = mt.tensor(rng.randn(n, p) + rng.randn(n, 1) * np.array([3, 4, 5]) +
np.array([1, 0, 7]))
Xt = mt.tensor(rng.randn(n, p) + rng.randn(n, 1) * np.array([3, 4, 5]) +
np.array([1, 0, 7]))
ll = mt.zeros(p)
for k in range(p):
pca = PCA(n_components=k, svd_solver='full')
pca.fit(Xl)
ll[k] = pca.score(Xt)
assert ll.argmax().to_numpy() == 1
