mindspore/tests/st/ops/cpu/test_lerp_op.py

191 lines
7.4 KiB
Python

# Copyright 2022 Huawei Technologies Co., Ltd
#
# 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 numpy as np
import pytest
import mindspore.nn as nn
import mindspore.ops.operations as P
from mindspore import Tensor, context
from mindspore.ops.functional import vmap
from mindspore.ops.operations.math_ops import Lerp
class LerpNet(nn.Cell):
def __init__(self):
super(LerpNet, self).__init__()
self.lerp = Lerp()
def construct(self, start, end, weight):
output = self.lerp(start, end, weight)
return output
class DynamicShapeNet(nn.Cell):
def __init__(self, axis=0):
super(DynamicShapeNet, self).__init__()
self.unique = P.Unique()
self.gather = P.Gather()
self.lerp = LerpNet()
self.axis = axis
def construct(self, start, end, weight, indices):
unique_indices, _ = self.unique(indices)
real_start = self.gather(start, unique_indices, self.axis)
real_end = self.gather(end, unique_indices, self.axis)
return real_start, real_end, self.lerp(real_start, real_end, weight)
class LerpVMapNet(nn.Cell):
def __init__(self, forward_net, in_axes, out_axes):
super(LerpVMapNet, self).__init__()
self.net = forward_net
self.in_axes = in_axes
self.out_axes = out_axes
def construct(self, start, end, weight):
return vmap(self.net, self.in_axes, self.out_axes)(start, end, weight)
def lerp_vmap_case():
"""
Feature: test lerp vamp feature.
Description: test special case.
Expectation: match to mindspore.ops.Lerp.
"""
# Case 1: in_axes start batch remains 0, other remains None.
start = Tensor(np.array([[1., 2., 3., 4.], [1., 3., 5., 7.]]).astype(np.float32))
end = Tensor(np.array([10., 10., 10., 10.]).astype(np.float32))
weight = 0.5
benchmark_output = np.array([[5.5, 6., 6.5, 7.], [5.5, 6.5, 7.5, 8.5]]).astype(np.float32)
in_axes = (0, None, None)
out_axes = 0
output = LerpVMapNet(LerpNet(), in_axes, out_axes)(start, end, weight)
assert np.allclose(output.asnumpy(), benchmark_output)
# Case 2: start remains 0 batch, end remains 1 batch, weight remains None.
start = Tensor(np.array([[1., 2., 3., 4.], [1., 3., 5., 7.]]).astype(np.float32))
end = Tensor(np.array([[10., 4.], [10., 4.], [10., 4.], [10., 4.]]).astype(np.float32))
weight = 0.5
benchmark_output = np.array([[5.5, 6., 6.5, 7.], [2.5, 3.5, 4.5, 5.5]]).astype(np.float32)
in_axes = (0, 1, None)
out_axes = 0
output = LerpVMapNet(LerpNet(), in_axes, out_axes)(start, end, weight)
assert np.allclose(output.asnumpy(), benchmark_output)
# Case 3: start remains 1 batch,end remains 1 batch, weight remains 0 batch.
start = Tensor(np.array([[1., 1], [2., 3.], [3., 5.], [4., 7.]]).astype(np.float32))
end = Tensor(np.array([[10., 4.], [10., 4.], [10., 4.], [10., 4.]]).astype(np.float32))
weight = Tensor(np.array([0.5, 0.4]).astype(np.float32))
benchmark_output = np.array([[5.5, 6., 6.5, 7.], [2.2, 3.4, 4.6, 5.8]]).astype(np.float32)
in_axes = (1, 1, 0)
out_axes = 0
output = LerpVMapNet(LerpNet(), in_axes, out_axes)(start, end, weight)
assert np.allclose(output.asnumpy(), benchmark_output)
# Case 4: start remain None, end remains 1 batch, weight remains 0 batch.
start = Tensor(np.array([1., 2., 3., 4.]).astype(np.float32))
end = Tensor(np.array([[10., 4.], [10., 4.], [10., 4.], [10., 4.]]).astype(np.float32))
weight = Tensor(np.array([0.5, 0.4]).astype(np.float32))
benchmark_output = np.array([[5.5, 6., 6.5, 7.], [2.2, 2.8, 3.4, 4.]]).astype(np.float32)
in_axes = (None, 1, 0)
out_axes = 0
output = LerpVMapNet(LerpNet(), in_axes, out_axes)(start, end, weight)
assert np.allclose(output.asnumpy(), benchmark_output)
def lerp_np_bencmark(start, end, weight):
"""
Feature: generate a lerp numpy benchmark.
Description: The input shape may need to broadcast.
Expectation: match to np mindspore lerp.
"""
end = np.broadcast_to(end, start.shape)
weight = np.broadcast_to(weight, start.shape)
result = start + weight * (end - start)
return result
@pytest.mark.level0
@pytest.mark.env_onecard
@pytest.mark.platform_x86_cpu
@pytest.mark.parametrize("data_shape", [(4,), (3, 4), (4, 5, 7)])
@pytest.mark.parametrize("data_type", [np.float32, np.float16])
def test_lerp(data_shape, data_type):
"""
Feature: Test Lerp.
Description: The input shape may need to broadcast.
Expectation: match to np benchmark.
"""
context.set_context(mode=context.GRAPH_MODE)
start = np.random.random(data_shape).astype(data_type)
end = np.ones(data_shape).astype(data_type)
error = 1e-6
if data_type == np.float16:
error = 1e-3
weight = 0.5
benchmark_output = lerp_np_bencmark(start, end, weight)
lerp = LerpNet()
output = lerp(Tensor(start), Tensor(end), Tensor(np.array(weight, dtype=data_type)))
np.testing.assert_allclose(output.asnumpy(), benchmark_output, rtol=error, atol=error)
context.set_context(mode=context.PYNATIVE_MODE)
output = lerp(Tensor(start), Tensor(end), Tensor(np.array(weight, dtype=data_type)))
np.testing.assert_allclose(output.asnumpy(), benchmark_output, rtol=error, atol=error)
@pytest.mark.level0
@pytest.mark.env_onecard
@pytest.mark.platform_x86_cpu
@pytest.mark.parametrize("data_type", [np.float32, np.float16])
def test_lerp_dy_shape(data_type):
"""
Feature: Test Lerp DyNamicShape.
Description: The input shape may need to broadcast.
Expectation: match to np benchmark.
"""
context.set_context(mode=context.GRAPH_MODE)
np.random.seed(1)
data_shape = (4, 5, 7)
start = np.random.random(data_shape).astype(data_type)
end = np.ones(data_shape).astype(data_type)
indices = np.random.randint(0, 4, size=4)
weight = 0.5
loss = 1e-6
if data_type == np.float16:
loss = 1e-3
net = DynamicShapeNet()
real_start, real_end, ms_result = net(Tensor(start), Tensor(end), weight, Tensor(indices))
np_result = lerp_np_bencmark(real_start.asnumpy(), real_end.asnumpy(), weight)
np.testing.assert_allclose(np_result, ms_result.asnumpy(), rtol=loss, atol=loss)
context.set_context(mode=context.PYNATIVE_MODE)
net = DynamicShapeNet()
real_start, real_end, ms_result = net(Tensor(start), Tensor(end), weight, Tensor(indices))
np_result = lerp_np_bencmark(real_start.asnumpy(), real_end.asnumpy(), weight)
np.testing.assert_allclose(np_result, ms_result.asnumpy(), rtol=loss, atol=loss)
@pytest.mark.level0
@pytest.mark.platform_x86_cpu
@pytest.mark.env_onecard
def test_lerp_vmap_cpu():
"""
Feature: test Lerp vmap on CPU.
Description: inputs(start, end, weight) with batch.
Expectation: the result match with expect
"""
context.set_context(mode=context.GRAPH_MODE, device_target="CPU")
lerp_vmap_case()
context.set_context(mode=context.PYNATIVE_MODE, device_target="CPU")
lerp_vmap_case()