You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
mindspore/tests/ut/python/pynative_mode/vm/test_vm.py

274 lines
8.8 KiB

# Copyright 2020 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.
# ============================================================================
""" test_vm """
import numpy as np
from .....vm_impl import vm
def test_avg_pooling():
""" test_avg_pooling """
input_data = np.array([[[[-4., -3., 1., 9.],
[-9., -1., 3., 4.],
[1., -1., -3., -6.],
[-2., -1., -2., -15.]]]]).astype(np.float32)
out = vm.avg_pooling(input_data, pool_h=2, pool_w=2, stride=1)
expect_out = [[[[-4.25, 0.0, 4.25],
[-2.5, -0.5, -0.5],
[-0.75, -1.75, -6.5]]]]
assert (expect_out == out).all()
def test_avg_pool_grad():
""" test_avg_pool_grad """
# To do
input_data = np.array([[[[1., 2, 3, 4],
[5, 6, 7, 8],
[9, 10, 11, 12],
[13, 14, 15, 16]]]]).astype(np.float32)
dout = vm.avg_pooling(input_data, pool_h=2, pool_w=2, stride=1)
print("vm.avg_pooling dout: ", dout)
out = vm.avg_pool_grad(dout, input_data.shape, 2, 2, 1)
print("vm.avg_pool_grad: ", out)
assert True
def test_batch_norm():
""" test_batch_norm """
input_data = np.random.randint(0, 255, [1, 3, 224, 224])
print("input_data.shape: ", input_data.shape)
print("input_data: ", input_data)
output = vm.batch_norm(input_data)
print("vm.batch_norm: ", output)
def test_conv2d():
""" test_conv2d """
x = np.array([[[
[3, 0, 1, 2, 7, 4],
[1, 5, 8, 9, 3, 1],
[2, 7, 2, 5, 1, 3],
[0, 1, 3, 1, 7, 8],
[4, 2, 1, 6, 2, 8],
[2, 4, 5, 2, 3, 9]]]]).astype(np.float32)
weight = np.array([[[[1, 0, -1], [1, 0, -1], [1, 0, -1]]]]).astype(np.float32)
out = vm.conv2d(x, weight)
expect_out = np.array([[[
[-5., -4., 0., 8.],
[-10., -2., 2., 3.],
[0., -2., -4., -7.],
[-3., -2., -3., -16.]]]]).astype(np.float32)
assert (expect_out == out).all()
def test_conv2d_with_bias():
""" test_conv2d_with_bias """
x = np.array([[[
[3, 0, 1, 2, 7, 4],
[1, 5, 8, 9, 3, 1],
[2, 7, 2, 5, 1, 3],
[0, 1, 3, 1, 7, 8],
[4, 2, 1, 6, 2, 8],
[2, 4, 5, 2, 3, 9]]]]).astype(np.float32)
weight = np.array([[[[1, 0, -1], [1, 0, -1], [1, 0, -1]]]]).astype(np.float32)
bias = np.array([1]).astype(np.float32)
out = vm.conv2d(x, weight, bias)
expect_out = np.array([[[
[-4., -3., 1., 9.],
[-9., -1., 3., 4.],
[1., -1., -3., -6.],
[-2., -1., -2., -15.]]]]).astype(np.float32)
assert (expect_out == out).all()
def test_conv2d_backprop_filter():
""" test_conv2d_backprop_filter """
x = np.array([[[
[3, 0, 1, 2, 7, 4],
[1, 5, 8, 9, 3, 1],
[2, 7, 2, 5, 1, 3],
[0, 1, 3, 1, 7, 8],
[4, 2, 1, 6, 2, 8],
[2, 4, 5, 2, 3, 9]]]]).astype(np.float32)
weight = np.array([[[[1, 0, -1], [1, 0, -1], [1, 0, -1]]]]).astype(np.float32)
out = vm.conv2d(x, weight)
backprop_filter = vm.conv2d_backprop_filter(out, x, weight.shape)
print(backprop_filter)
assert True
def test_conv2d_backprop_input():
""" test_conv2d_backprop_input """
x = np.array([[[
[3, 0, 1, 2, 7, 4],
[1, 5, 8, 9, 3, 1],
[2, 7, 2, 5, 1, 3],
[0, 1, 3, 1, 7, 8],
[4, 2, 1, 6, 2, 8],
[2, 4, 5, 2, 3, 9]]]]).astype(np.float32)
weight = np.array([[[[1, 0, -1], [1, 0, -1], [1, 0, -1]]]]).astype(np.float32)
out = vm.conv2d(x, weight)
grad = vm.conv2d_backprop_input(out, x.shape, weight)
print(grad)
assert True
def test_flatten():
""" test_flatten """
x = np.array([[1, 2, 3], [4, 5, 6]]).astype(np.float32)
y = vm.flatten(x)
assert ([1, 2, 3, 4, 5, 6] == y.T).all()
assert np.float32 == y.dtype
def test_flatten2():
""" test_flatten2 """
x = np.array([[1, 2, 3], [4, 5, 6]]).astype(np.float32)
y = vm.flatten2(x)
assert ([1, 2, 3, 4, 5, 6] == y).all()
assert (1, 6) == y.shape
assert np.float32 == y.dtype
def test_flatten_batch():
""" test_flatten_batch """
x = np.array([[[9, 4, 14, 1],
[7, 10, 14, 13],
[1, 9, 16, 7],
[15, 16, 0, 4]],
[[16, 13, 13, 10],
[0, 12, 5, 9],
[15, 0, 11, 1],
[4, 16, 4, 1]],
[[2, 8, 1, 13],
[5, 15, 4, 11],
[8, 2, 17, 16],
[5, 13, 0, 2]],
[[14, 8, 6, 8],
[0, 8, 6, 15],
[9, 1, 8, 5],
[12, 6, 13, 8]],
[[13, 11, 6, 3],
[8, 6, 16, 5],
[7, 10, 0, 8],
[17, 17, 17, 3]]]).astype(np.float32)
y = vm.flatten_batch(x)
expect_out = np.array(
[[9, 4, 14, 1, 7, 10, 14, 13, 1, 9, 16, 7, 15, 16, 0, 4],
[16, 13, 13, 10, 0, 12, 5, 9, 15, 0, 11, 1, 4, 16, 4, 1],
[2, 8, 1, 13, 5, 15, 4, 11, 8, 2, 17, 16, 5, 13, 0, 2],
[14, 8, 6, 8, 0, 8, 6, 15, 9, 1, 8, 5, 12, 6, 13, 8],
[13, 11, 6, 3, 8, 6, 16, 5, 7, 10, 0, 8, 17, 17, 17, 3]]).astype(np.float32)
assert (expect_out == y).all()
assert expect_out.shape == y.shape
assert np.float32 == y.dtype
def test_im2col():
""" test_im2col """
img = np.ones([1, 1, 32, 32]).astype(np.float32) * 0.01
print("input img: ", img)
col = vm.im2col(img, 2, 3, 1, 1)
print("output col.shape : ", col.shape)
print("output col: ", col)
print("output col.dtype: ", col.dtype)
assert np.float32 == col.dtype
def test_matmul():
""" test_matmul """
x = np.array([1, 2, 3]).astype(np.float32)
w = np.array([0, 1, 0.5]).astype(np.float32)
y = vm.matmul(x, w)
assert y == 3.5
assert np.float32 == y.dtype
def test_max_pooling():
""" test_max_pooling """
input_data = np.array([[[
[-4., -3., 1., 9.],
[-9., -1., 3., 4.],
[1., -1., -3., -6.],
[-2., -1., -2., -15.]]]]).astype(np.float32)
out = vm.max_pooling(input_data, pool_h=2, pool_w=2, stride=1)
expect_out = [[[[-1., 3., 9.],
[1., 3., 4.],
[1., -1., -2.]]]]
assert (expect_out == out).all()
assert np.float32 == out.dtype
def test_np_convolve():
""" test_np_convolve """
out = np.convolve([1, 2, 3], [0, 1, 0.5]).astype(np.float32)
assert ([0.0, 1.0, 2.5, 4.0, 1.5] == out).all()
assert np.float32 == out.dtype
def test_np_convolve_same():
""" test_np_convolve_same """
out = np.convolve([1, 2, 3], [0, 1, 0.5], 'same').astype(np.float32)
assert ([1.0, 2.5, 4.0] == out).all()
assert np.float32 == out.dtype
def test_np_convolve_valid():
""" test_np_convolve_valid """
out = np.convolve([1, 2, 3], [0, 1, 0.5], 'valid').astype(np.float32)
assert ([2.5] == out).all()
assert np.float32 == out.dtype
def test_relu():
""" test_relu """
x = np.array([-0.32208174, 0.33999891]).astype(np.float32)
y = vm.relu(x)
assert np.allclose([-0., 0.33999891], y)
assert np.float32 == y.dtype
y = vm.relu_grad(y)
assert (y == [0., 1.]).all()
assert np.float32 == y.dtype
def test_softmax():
""" test_softmax """
logits = 2.84806275 * np.ones([1, 10]).astype(np.float32)
y = vm.softmax(logits)
assert np.allclose([0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1], y)
assert np.float32 == y.dtype
logits = np.array([[1, 2, 3], [4, 5, 6]]).astype(np.float32)
y = vm.softmax(logits, axis=1)
labels = [[0.09003057, 0.24472847, 0.66524096], [0.09003057, 0.24472847, 0.66524096]]
assert np.allclose(labels, y)
assert np.float32 == y.dtype
def test_softmax_cross_entropy_with_logit():
""" test_softmax_cross_entropy_with_logit """
logits = np.array([[1, 2, 3, 4, 2, 1, 0, 2, 1, 1], [1, 2, 4, 1, 0, 5, 0, 2, 1, 3]], dtype=np.float32)
labels = np.array([[0, 0, 0, 1, 0, 0, 0, 0, 0, 0], [0, 0, 1, 0, 0, 0, 0, 0, 0, 0]], dtype=np.float32)
loss, dx = vm.softmax_cross_entropy_with_logits(logits, labels)
print("logits.shape: ", logits.shape)
print("logits: ", logits)
print("softmax: ", vm.softmax(logits))
print("labels: ", labels)
print("loss: ", loss)
print("dx: ", dx)
assert np.float32 == loss.dtype
assert np.float32 == dx.dtype