def test_abs(self):
float32_tensor = ht.arange(-10, 10, dtype=ht.float32, split=0)
absolute_values = ht.abs(float32_tensor)
# basic absolute test
self.assertIsInstance(absolute_values, ht.tensor)
self.assertEqual(absolute_values.dtype, ht.float32)
self.assertEqual(absolute_values.sum(axis=0), 100)
# check whether output works
output_tensor = ht.zeros(20, split=0)
self.assertEqual(output_tensor.sum(axis=0), 0)
ht.absolute(float32_tensor, out=output_tensor)
self.assertEqual(output_tensor.sum(axis=0), 100)
# dtype parameter
int64_tensor = ht.arange(-10, 10, dtype=ht.int64)
absolute_values = ht.abs(int64_tensor, dtype=ht.float32)
self.assertIsInstance(absolute_values, ht.tensor)
self.assertEqual(absolute_values.sum(axis=0), 100)
self.assertEqual(absolute_values.dtype, ht.float32)
self.assertEqual(absolute_values._tensor__array.dtype, torch.float32)
# exceptions
with self.assertRaises(TypeError):
ht.absolute('hello')
with self.assertRaises(TypeError):
float32_tensor.abs(out=1)
with self.assertRaises(TypeError):
float32_tensor.absolute(out=float32_tensor, dtype=3.2)
def test_fit_iris_unsplit(self):
split = 0
# get some test data
iris = ht.load("heat/datasets/iris.csv", sep=";", split=split)
ht.random.seed(1)
# fit the clusters
k = 3
kmedoid = ht.cluster.KMedoids(n_clusters=k, random_state=1)
kmedoid.fit(iris)
# check whether the results are correct
self.assertIsInstance(kmedoid.cluster_centers_, ht.DNDarray)
self.assertEqual(kmedoid.cluster_centers_.shape, (k, iris.shape[1]))
# same test with init=kmedoids++
kmedoid = ht.cluster.KMedoids(n_clusters=k, init="kmedoids++")
kmedoid.fit(iris)
# check whether the results are correct
self.assertIsInstance(kmedoid.cluster_centers_, ht.DNDarray)
self.assertEqual(kmedoid.cluster_centers_.shape, (k, iris.shape[1]))
# check whether result is actually a datapoint
for i in range(kmedoid.cluster_centers_.shape[0]):
self.assertTrue(
ht.any(
ht.sum(ht.abs(kmedoid.cluster_centers_[i, :] - iris),
axis=1) == 0))
def test_spherical_clusters(self):
seed = 1
n = 20 * ht.MPI_WORLD.size
data = self.create_spherical_dataset(num_samples_cluster=n,
radius=1.0,
offset=4.0,
dtype=ht.float32,
random_state=seed)
kmedoid = ht.cluster.KMedoids(n_clusters=4, init="kmedoids++")
kmedoid.fit(data)
self.assertIsInstance(kmedoid.cluster_centers_, ht.DNDarray)
self.assertEqual(kmedoid.cluster_centers_.shape, (4, 3))
for i in range(kmedoid.cluster_centers_.shape[0]):
self.assertTrue(
ht.any(
ht.sum(ht.abs(kmedoid.cluster_centers_[i, :] - data),
axis=1) == 0))
# More Samples
n = 100 * ht.MPI_WORLD.size
data = self.create_spherical_dataset(num_samples_cluster=n,
radius=1.0,
offset=4.0,
dtype=ht.float32,
random_state=seed)
kmedoid = ht.cluster.KMedoids(n_clusters=4, init="kmedoids++")
kmedoid.fit(data)
self.assertIsInstance(kmedoid.cluster_centers_, ht.DNDarray)
self.assertEqual(kmedoid.cluster_centers_.shape, (4, 3))
# check whether result is actually a datapoint
for i in range(kmedoid.cluster_centers_.shape[0]):
self.assertTrue(
ht.any(
ht.sum(ht.abs(kmedoid.cluster_centers_[i, :] - data),
axis=1) == 0))
# different datatype
n = 20 * ht.MPI_WORLD.size
data = self.create_spherical_dataset(num_samples_cluster=n,
radius=1.0,
offset=4.0,
dtype=ht.float64,
random_state=seed)
kmedoid = ht.cluster.KMedoids(n_clusters=4, init="kmedoids++")
kmedoid.fit(data)
self.assertIsInstance(kmedoid.cluster_centers_, ht.DNDarray)
self.assertEqual(kmedoid.cluster_centers_.shape, (4, 3))
for i in range(kmedoid.cluster_centers_.shape[0]):
self.assertTrue(
ht.any(
ht.sum(ht.abs(kmedoid.cluster_centers_[i, :] -
data.astype(ht.float32)),
axis=1) == 0))
# on Ints (different radius, offset and datatype
data = self.create_spherical_dataset(num_samples_cluster=n,
radius=10.0,
offset=40.0,
dtype=ht.int32,
random_state=seed)
kmedoid = ht.cluster.KMedoids(n_clusters=4, init="kmedoids++")
kmedoid.fit(data)
self.assertIsInstance(kmedoid.cluster_centers_, ht.DNDarray)
self.assertEqual(kmedoid.cluster_centers_.shape, (4, 3))
for i in range(kmedoid.cluster_centers_.shape[0]):
self.assertTrue(
ht.any(
ht.sum(ht.abs(kmedoid.cluster_centers_[i, :] - data),
axis=1) == 0))
def lanczos(
A: DNDarray,
m: int,
v0: Optional[DNDarray] = None,
V_out: Optional[DNDarray] = None,
T_out: Optional[DNDarray] = None,
) -> Tuple[DNDarray, DNDarray]:
r"""
The Lanczos algorithm is an iterative approximation of the solution to the eigenvalue problem, as an adaptation of
power methods to find the m "most useful" (tending towards extreme highest/lowest) eigenvalues and eigenvectors of
an :math:`n \times n` Hermitian matrix, where often :math:`m<<n`.
It returns two matrices :math:`V` and :math:`T`, where:
- :math:`V` is a Matrix of size :math:`n\times m`, with orthonormal columns, that span the Krylow subspace \n
- :math:`T` is a Tridiagonal matrix of size :math:`m\times m`, with coefficients :math:`\alpha_1,..., \alpha_n`
on the diagonal and coefficients :math:`\beta_1,...,\beta_{n-1}` on the side-diagonals\n
Parameters
----------
A : DNDarray
2D symmetric, positive definite Matrix
m : int
Number of Lanczos iterations
v0 : DNDarray, optional
1D starting vector of Euclidian norm 1. If not provided, a random vector will be used to start the algorithm
V_out : DNDarray, optional
Output Matrix for the Krylow vectors, Shape = (n, m)
T_out : DNDarray, optional
Output Matrix for the Tridiagonal matrix, Shape = (m, m)
"""
if not isinstance(A, DNDarray):
raise TypeError("A needs to be of type ht.dndarra, but was {}".format(
type(A)))
if not (A.ndim == 2):
raise RuntimeError("A needs to be a 2D matrix")
if not isinstance(m, (int, float)):
raise TypeError("m must be eiter int or float, but was {}".format(
type(m)))
n, column = A.shape
if n != column:
raise TypeError("Input Matrix A needs to be symmetric.")
T = ht.zeros((m, m))
if A.split == 0:
# This is done for better memory access in the reorthogonalization Gram-Schmidt algorithm
V = ht.ones((n, m), split=0, dtype=A.dtype, device=A.device)
else:
V = ht.ones((n, m), split=None, dtype=A.dtype, device=A.device)
if v0 is None:
vr = ht.random.rand(n, split=V.split)
v0 = vr / ht.norm(vr)
else:
if v0.split != V.split:
v0.resplit_(axis=V.split)
# # 0th iteration
# # vector v0 has euklidian norm = 1
w = ht.matmul(A, v0)
alpha = ht.dot(w, v0)
w = w - alpha * v0
T[0, 0] = alpha
V[:, 0] = v0
for i in range(1, int(m)):
beta = ht.norm(w)
if ht.abs(beta) < 1e-10:
# print("Lanczos breakdown in iteration {}".format(i))
# Lanczos Breakdown, pick a random vector to continue
vr = ht.random.rand(n, dtype=A.dtype, split=V.split)
# orthogonalize v_r with respect to all vectors v[i]
for j in range(i):
vi_loc = V.larray[:, j]
a = torch.dot(vr.larray, vi_loc)
b = torch.dot(vi_loc, vi_loc)
A.comm.Allreduce(ht.communication.MPI.IN_PLACE, a,
ht.communication.MPI.SUM)
A.comm.Allreduce(ht.communication.MPI.IN_PLACE, b,
ht.communication.MPI.SUM)
vr.larray = vr.larray - a / b * vi_loc
# normalize v_r to Euclidian norm 1 and set as ith vector v
vi = vr / ht.norm(vr)
else:
vr = w
# Reorthogonalization
# ToDo: Rethink this; mask torch calls, See issue #494
# This is the fast solution, using item access on the ht.dndarray level is way slower
for j in range(i):
vi_loc = V.larray[:, j]
a = torch.dot(vr._DNDarray__array, vi_loc)
b = torch.dot(vi_loc, vi_loc)
A.comm.Allreduce(ht.communication.MPI.IN_PLACE, a,
ht.communication.MPI.SUM)
A.comm.Allreduce(ht.communication.MPI.IN_PLACE, b,
ht.communication.MPI.SUM)
vr._DNDarray__array = vr._DNDarray__array - a / b * vi_loc
vi = vr / ht.norm(vr)
w = ht.matmul(A, vi)
alpha = ht.dot(w, vi)
w = w - alpha * vi - beta * V[:, i - 1]
T[i - 1, i] = beta
T[i, i - 1] = beta
T[i, i] = alpha
V[:, i] = vi
if V.split is not None:
V.resplit_(axis=None)
if T_out is not None:
T_out = T.copy()
if V_out is not None:
V_out = V.copy()
return V_out, T_out
return V, T_out
elif V_out is not None:
V_out = V.copy()
return V_out, T
return V, T
def test_abs(self):
# for abs==absolute
float32_tensor = ht.arange(-10, 10, dtype=ht.float32, split=0)
absolute_values = ht.abs(float32_tensor)
# for fabs
int8_tensor_fabs = ht.arange(-10.5, 10.5, dtype=ht.int8, split=0)
int8_absolute_values_fabs = ht.fabs(int8_tensor_fabs)
int16_tensor_fabs = ht.arange(-10.5, 10.5, dtype=ht.int16, split=0)
int16_absolute_values_fabs = ht.fabs(int16_tensor_fabs)
int32_tensor_fabs = ht.arange(-10.5, 10.5, dtype=ht.int32, split=0)
int32_absolute_values_fabs = ht.fabs(int32_tensor_fabs)
int64_tensor_fabs = ht.arange(-10.5, 10.5, dtype=ht.int64, split=0)
int64_absolute_values_fabs = ht.fabs(int64_tensor_fabs)
float32_tensor_fabs = ht.arange(-10.5, 10.5, dtype=ht.float32, split=0)
float32_absolute_values_fabs = ht.fabs(float32_tensor_fabs)
float64_tensor_fabs = ht.arange(-10.5, 10.5, dtype=ht.float64, split=0)
float64_absolute_values_fabs = ht.fabs(float64_tensor_fabs)
# basic absolute test
self.assertIsInstance(absolute_values, ht.DNDarray)
self.assertEqual(absolute_values.dtype, ht.float32)
self.assertEqual(absolute_values.sum(axis=0), 100)
# for fabs
self.assertEqual(int8_absolute_values_fabs.sum(axis=0), 100.0)
self.assertEqual(int16_absolute_values_fabs.sum(axis=0), 100.0)
self.assertEqual(int32_absolute_values_fabs.sum(axis=0), 100.0)
self.assertEqual(int64_absolute_values_fabs.sum(axis=0), 100.0)
self.assertEqual(float32_absolute_values_fabs.sum(axis=0), 110.5)
self.assertEqual(float64_absolute_values_fabs.sum(axis=0), 110.5)
# check whether output works
# for abs==absolute
output_tensor = ht.zeros(20, split=0)
self.assertEqual(output_tensor.sum(axis=0, keepdim=True), 0)
ht.absolute(float32_tensor, out=output_tensor)
self.assertEqual(output_tensor.sum(axis=0), 100)
# for fabs
output_tensor_fabs = ht.zeros(21, split=0)
self.assertEqual(output_tensor_fabs.sum(axis=0), 0)
ht.fabs(float32_tensor_fabs, out=output_tensor_fabs)
self.assertEqual(output_tensor_fabs.sum(axis=0), 110.5)
# dtype parameter
# for abs==absolute
int64_tensor = ht.arange(-10, 10, dtype=ht.int64)
absolute_values = ht.abs(int64_tensor, dtype=ht.float32)
self.assertIsInstance(absolute_values, ht.DNDarray)
self.assertEqual(absolute_values.sum(axis=0), 100)
self.assertEqual(absolute_values.dtype, ht.float32)
self.assertEqual(absolute_values._DNDarray__array.dtype, torch.float32)
# for fabs
self.assertEqual(int8_absolute_values_fabs.dtype, ht.float32)
self.assertEqual(int16_absolute_values_fabs.dtype, ht.float32)
self.assertEqual(int32_absolute_values_fabs.dtype, ht.float32)
self.assertEqual(int64_absolute_values_fabs.dtype, ht.float64)
self.assertEqual(float32_absolute_values_fabs.dtype, ht.float32)
self.assertEqual(float64_absolute_values_fabs.dtype, ht.float64)
# exceptions
# for abs==absolute
with self.assertRaises(TypeError):
ht.absolute("hello")
with self.assertRaises(TypeError):
float32_tensor.abs(out=1)
with self.assertRaises(TypeError):
float32_tensor.absolute(out=float32_tensor, dtype=3.2)
# for fabs
with self.assertRaises(TypeError):
ht.fabs("hello")
with self.assertRaises(TypeError):
float32_tensor_fabs.fabs(out=1)
# test with unsplit tensor
# for fabs
float32_unsplit_tensor_fabs = ht.arange(-10.5, 10.5, dtype=ht.float32)
float32_unsplit_absolute_values_fabs = ht.fabs(
float32_unsplit_tensor_fabs)
self.assertEqual(float32_unsplit_absolute_values_fabs.sum(), 110.5)
self.assertEqual(float32_unsplit_absolute_values_fabs.dtype,
ht.float32)
