wehub-resource-sync
300 lines
10 KiB
Python
300 lines
10 KiB
Python
# Copyright (c) 2023 PaddlePaddle Authors. All Rights Reserved.
|
|
#
|
|
# Licensed under the Apache License, Version 2.0 (the "License");
|
|
# you may not use this file except in compliance with the License.
|
|
# You may obtain a copy of the License at
|
|
#
|
|
# http://www.apache.org/licenses/LICENSE-2.0
|
|
#
|
|
# Unless required by applicable law or agreed to in writing, software
|
|
# distributed under the License is distributed on an "AS IS" BASIS,
|
|
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
|
# See the License for the specific language governing permissions and
|
|
# limitations under the License.
|
|
|
|
import unittest
|
|
|
|
import numpy as np
|
|
from op_test import get_device_place, is_custom_device
|
|
|
|
import paddle
|
|
|
|
|
|
def geqrf(x):
|
|
def _geqrf(x):
|
|
m, n = x.shape
|
|
tau = np.zeros([n, 1], dtype=x.dtype)
|
|
for i in range(min(n, m)):
|
|
alpha = x[i, i]
|
|
normx = np.linalg.norm(x[min(i + 1, m) :, i])
|
|
beta = np.linalg.norm(x[i:, i])
|
|
if x.dtype in [np.complex64, np.complex128]:
|
|
s = 1 if alpha < 0 else -1
|
|
else:
|
|
alphar = x[i, i].real
|
|
s = 1 if alphar < 0 else -1
|
|
u1 = alpha - s * beta
|
|
w = x[i:, i] / u1
|
|
w[0] = 1
|
|
x[i + 1 :, i] = w[1 : m - i + 1]
|
|
if normx == 0:
|
|
tau[i] = 0
|
|
else:
|
|
tau[i] = -s * u1 / beta
|
|
x[i, i] = s * beta
|
|
w = w.reshape([-1, 1])
|
|
if x.dtype in [np.complex64, np.complex128]:
|
|
x[i:, i + 1 :] = x[i:, i + 1 :] - (tau[i] * w) @ (
|
|
np.conj(w).T @ x[i:, i + 1 :]
|
|
)
|
|
else:
|
|
x[i:, i + 1 :] = x[i:, i + 1 :] - (tau[i] * w) @ (
|
|
w.T @ x[i:, i + 1 :]
|
|
)
|
|
return x, tau[: min(m, n)].reshape(-1)
|
|
|
|
if len(x.shape) == 2:
|
|
return _geqrf(x)
|
|
m, n = x.shape[-2:]
|
|
org_x_shape = x.shape
|
|
x = x.reshape((-1, x.shape[-2], x.shape[-1]))
|
|
n_batch = x.shape[0]
|
|
out = np.zeros([n_batch, m, n], dtype=x.dtype)
|
|
taus = np.zeros([n_batch, min(m, n)], dtype=x.dtype)
|
|
org_taus_shape = [*org_x_shape[:-2], min(m, n)]
|
|
for i in range(n_batch):
|
|
out[i], t = _geqrf(x[i])
|
|
taus[i, :] = t.reshape(-1)
|
|
return out.reshape(org_x_shape), taus.reshape(org_taus_shape)
|
|
|
|
|
|
def ref_ormqr(input, tau, y, left=True, transpose=False):
|
|
m, n = input.shape[-2:]
|
|
|
|
def _ref_ormqr(input_matrix, tau_vector):
|
|
k = tau_vector.shape[-1]
|
|
Q = np.eye(m, dtype=input_matrix.dtype)
|
|
for i in range(min(k, n)):
|
|
w = input_matrix[i:, i]
|
|
w[0] = 1
|
|
w = w.reshape(-1, 1)
|
|
if np.iscomplexobj(input_matrix):
|
|
Q[:, i:] = Q[:, i:] - (
|
|
Q[:, i:] @ w @ np.conj(w).T * tau_vector[i]
|
|
)
|
|
else:
|
|
Q[:, i:] = Q[:, i:] - (Q[:, i:] @ w @ w.T * tau_vector[i])
|
|
return Q[:, :n]
|
|
|
|
if input.ndim == 2:
|
|
Q = _ref_ormqr(input, tau)
|
|
Q = Q.T if transpose else Q
|
|
else:
|
|
org_input_shape = input.shape
|
|
org_tau_shape = tau.shape
|
|
|
|
input_reshaped = input.reshape(
|
|
(-1, org_input_shape[-2], org_input_shape[-1])
|
|
)
|
|
tau_reshaped = tau.reshape((-1, org_tau_shape[-1]))
|
|
|
|
n_batch = input_reshaped.shape[0]
|
|
Q = np.zeros((n_batch, m, n), dtype=input.dtype)
|
|
|
|
for i in range(n_batch):
|
|
Q[i] = _ref_ormqr(input_reshaped[i], tau_reshaped[i])
|
|
|
|
Q = (
|
|
np.transpose(Q.reshape(org_input_shape), (0, 2, 1))
|
|
if transpose
|
|
else Q.reshape(org_input_shape)
|
|
)
|
|
result = np.matmul(Q, y) if left else np.matmul(y, Q)
|
|
|
|
return result
|
|
|
|
|
|
class TestOrmqrAPI(unittest.TestCase):
|
|
def setUp(self):
|
|
paddle.seed(2024)
|
|
self.place = get_device_place()
|
|
self.init_input()
|
|
|
|
def init_input(self):
|
|
self.x = np.array(
|
|
[
|
|
[1, 2, 4],
|
|
[0, 0, 5],
|
|
[0, 3, 6],
|
|
],
|
|
dtype=np.float32,
|
|
)
|
|
self.y = np.array(
|
|
[
|
|
[1, 2, 4],
|
|
[0, 0, 5],
|
|
[0, 3, 6],
|
|
],
|
|
dtype=np.float32,
|
|
)
|
|
|
|
def test_static_api(self):
|
|
m, n = self.x.shape[-2:]
|
|
self.geqrf_x, self.tau = geqrf(self.x)
|
|
self._x = self.geqrf_x.copy()
|
|
self._tau = self.tau.copy()
|
|
self._y = self.y.copy()
|
|
paddle.enable_static()
|
|
with paddle.static.program_guard(paddle.static.Program()):
|
|
x = paddle.static.data(
|
|
'x', self.geqrf_x.shape, dtype=self.geqrf_x.dtype
|
|
)
|
|
tau = paddle.static.data(
|
|
'tau', self.tau.shape, dtype=self.tau.dtype
|
|
)
|
|
y = paddle.static.data('y', self.y.shape, dtype=self.y.dtype)
|
|
out = paddle.linalg.ormqr(x, tau, y)
|
|
exe = paddle.static.Executor(self.place)
|
|
res = exe.run(
|
|
feed={'x': self.geqrf_x, 'tau': self.tau, 'y': self.y},
|
|
fetch_list=[out],
|
|
)
|
|
out_ref = ref_ormqr(self._x, self._tau, self._y)
|
|
np.testing.assert_allclose(out_ref, res[0], rtol=1e-3, atol=1e-3)
|
|
|
|
def test_dygraph_api(self):
|
|
m, n = self.x.shape[-2:]
|
|
self.geqrf_x, self.tau = geqrf(self.x)
|
|
self._x = self.geqrf_x.copy()
|
|
self._tau = self.tau.copy()
|
|
self._y = self.y.copy()
|
|
paddle.disable_static(self.place)
|
|
x = paddle.to_tensor(self.geqrf_x)
|
|
tau = paddle.to_tensor(self.tau)
|
|
y = paddle.to_tensor(self.y)
|
|
out = paddle.linalg.ormqr(x, tau, y)
|
|
out_ref = ref_ormqr(self._x, self._tau, self._y)
|
|
np.testing.assert_allclose(out_ref, out.numpy(), rtol=1e-3, atol=1e-3)
|
|
paddle.enable_static()
|
|
|
|
def test_error(self):
|
|
pass
|
|
|
|
|
|
class TestOrmqrAPICase1(TestOrmqrAPI):
|
|
def init_input(self):
|
|
self.x = np.random.randn(4, 3).astype('float32')
|
|
self.y = np.random.randn(3, 4).astype('float32')
|
|
|
|
|
|
class TestOrmqrAPICase2(TestOrmqrAPI):
|
|
def init_input(self):
|
|
self.x = np.random.randn(4, 3).astype('float64')
|
|
self.y = np.random.randn(3, 4).astype('float64')
|
|
|
|
|
|
class TestOrmqrAPICase3(TestOrmqrAPI):
|
|
def init_input(self):
|
|
self.x = np.random.randn(5, 4, 3).astype('float32')
|
|
self.y = np.random.randn(5, 3, 4).astype('float32')
|
|
|
|
|
|
# complex dtype
|
|
class TestOrmqrAPICase4(TestOrmqrAPI):
|
|
def init_input(self):
|
|
self.x = np.random.randn(4, 3).astype('complex64')
|
|
self.y = np.random.randn(3, 4).astype('complex64')
|
|
|
|
|
|
class TestOrmqrAPICase5(TestOrmqrAPI):
|
|
def init_input(self):
|
|
self.x = np.random.randn(4, 3).astype('complex128')
|
|
self.y = np.random.randn(3, 4).astype('complex128')
|
|
|
|
|
|
class TestOrmqrAPICase6(TestOrmqrAPI):
|
|
def init_input(self):
|
|
if paddle.is_compiled_with_cuda() or is_custom_device():
|
|
self.x = np.random.randn(4, 3).astype('float16')
|
|
self.y = np.random.randn(3, 4).astype('float16')
|
|
else:
|
|
self.x = np.random.randn(4, 3).astype('float32')
|
|
self.y = np.random.randn(3, 4).astype('float32')
|
|
|
|
|
|
class TestOrmqrAPI_type_error(TestOrmqrAPI):
|
|
def test_error(self):
|
|
with self.assertRaises(AssertionError):
|
|
x = paddle.randn([4, 3], dtype='float64')
|
|
tau = paddle.randn([3], dtype='float32')
|
|
y = paddle.randn([3, 4], dtype='float64')
|
|
out = paddle.linalg.ormqr(x, tau, y)
|
|
|
|
|
|
class TestOrmqrAPI_shape_error(TestOrmqrAPI):
|
|
def test_error(self):
|
|
with self.assertRaises(AssertionError):
|
|
x = paddle.randn([3], dtype='float32')
|
|
tau = paddle.randn([], dtype='float32')
|
|
y = paddle.randn([3], dtype='float32')
|
|
out = paddle.linalg.ormqr(x, tau, y)
|
|
|
|
|
|
class TestOrmqrAPI_dim_error(TestOrmqrAPI):
|
|
def test_error(self):
|
|
with self.assertRaises(AssertionError):
|
|
x = paddle.randn([3, 4], dtype='float32')
|
|
tau = paddle.randn([3, 4], dtype='float32')
|
|
y = paddle.randn([4, 3], dtype='float32')
|
|
out = paddle.linalg.ormqr(x, tau, y)
|
|
|
|
|
|
class TestOrmqrAPI_householder_error(TestOrmqrAPI):
|
|
def test_error(self):
|
|
with self.assertRaises(AssertionError):
|
|
x = paddle.randn([4, 3], dtype='float32')
|
|
tau = paddle.randn([4], dtype='float32')
|
|
y = paddle.randn([3, 4], dtype='float32')
|
|
out = paddle.linalg.ormqr(x, tau, y)
|
|
|
|
|
|
class TestOrmqrAPI_y_error(TestOrmqrAPI):
|
|
def test_error(self):
|
|
with self.assertRaises(AssertionError):
|
|
x = paddle.randn([4, 3], dtype='float32')
|
|
tau = paddle.randn([3], dtype='float32')
|
|
y = paddle.randn([3, 4], dtype='float32')
|
|
out = paddle.linalg.ormqr(x, tau, y, left=True, transpose=True)
|
|
|
|
with self.assertRaises(AssertionError):
|
|
x = paddle.randn([4, 3], dtype='float32')
|
|
tau = paddle.randn([3], dtype='float32')
|
|
y = paddle.randn([4, 3], dtype='float32')
|
|
out = paddle.linalg.ormqr(x, tau, y, left=True, transpose=False)
|
|
|
|
with self.assertRaises(AssertionError):
|
|
x = paddle.randn([4, 3], dtype='float32')
|
|
tau = paddle.randn([3], dtype='float32')
|
|
y = paddle.randn([3, 4], dtype='float32')
|
|
out = paddle.linalg.ormqr(x, tau, y, left=False, transpose=False)
|
|
|
|
with self.assertRaises(AssertionError):
|
|
x = paddle.randn([4, 3], dtype='float32')
|
|
tau = paddle.randn([3], dtype='float32')
|
|
y = paddle.randn([4, 3], dtype='float32')
|
|
out = paddle.linalg.ormqr(x, tau, y, left=False, transpose=True)
|
|
|
|
|
|
class TestOrmqrAPI_batch_error(TestOrmqrAPI):
|
|
def test_error(self):
|
|
with self.assertRaises(AssertionError):
|
|
x = paddle.randn([5, 3, 4], dtype='float32')
|
|
tau = paddle.randn([5, 4], dtype='float32')
|
|
y = paddle.randn([4, 4, 3], dtype='float32')
|
|
out = paddle.linalg.ormqr(x, tau, y)
|
|
|
|
|
|
if __name__ == '__main__':
|
|
paddle.enable_static()
|
|
unittest.main()
|