p60975123
/
mindspore
forked from mindspore-Ecosystem/mindspore
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

!28975 [MD] Chained Sampler Python UT fixes and updates 

Merge pull request !28975 from cathwong/ckw_ut_sampler_chain_fixes2
This commit is contained in:
i-robot 2022-01-14 03:49:05 +00:00 committed by Gitee
parent 0b007e3973 9d9013ba79
commit cfe5669de9
No known key found for this signature in database
GPG Key ID: 173E9B9CA92EEF8F
3 changed files with 143 additions and 56 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

81
tests/ut/python/dataset/test_datasets_imagefolder.py
View File

@ -1,4 +1,4 @@
# Copyright 2019-2021 Huawei Technologies Co., Ltd
# Copyright 2019-2022 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.
@ -421,6 +421,11 @@ def test_weighted_random_sampler_exception():
def test_chained_sampler_01():
"""
Feature: Chained Sampler
Description: Chained Samplers: Random and Sequential, with repeat
Expectation: Get correct number of data
"""
logger.info("Test Case Chained Sampler - Random and Sequential, with repeat")
# Create chained sampler, random and sequential
@ -450,6 +455,11 @@ def test_chained_sampler_01():
def test_chained_sampler_02():
"""
Feature: Chained Sampler
Description: Chained Samplers: Random and Sequential, with batch then repeat
Expectation: Get correct number of data
"""
logger.info("Test Case Chained Sampler - Random and Sequential, with batch then repeat")
# Create chained sampler, random and sequential
@ -480,6 +490,11 @@ def test_chained_sampler_02():
def test_chained_sampler_03():
"""
Feature: Chained Sampler
Description: Chained Samplers: Random and Sequential, with repeat then batch
Expectation: Get correct number of data
"""
logger.info("Test Case Chained Sampler - Random and Sequential, with repeat then batch")
# Create chained sampler, random and sequential
@ -510,6 +525,11 @@ def test_chained_sampler_03():
def test_chained_sampler_04():
"""
Feature: Chained Sampler
Description: Chained Samplers: Distributed and Random, with batch then repeat
Expectation: Get correct number of data
"""
logger.info("Test Case Chained Sampler - Distributed and Random, with batch then repeat")
# Create chained sampler, distributed and random
@ -541,13 +561,18 @@ def test_chained_sampler_04():
assert num_iter == 6
def skip_test_chained_sampler_05():
logger.info("Test Case Chained Sampler - PKSampler and WeightedRandom")
def test_chained_sampler_05():
"""
Feature: Chained Sampler
Description: Chained Samplers: Distributed and WeightedRandom
Expectation: Get correct number of data
"""
logger.info("Test Case Chained Sampler - Distributed and WeightedRandom")
# Create chained sampler, PKSampler and WeightedRandom
sampler = ds.PKSampler(num_val=3) # Number of elements per class is 3 (and there are 4 classes)
# Create chained sampler, Distributed and WeightedRandom
sampler = ds.DistributedSampler(num_shards=2, shard_id=1)
weights = [1.0, 0.1, 0.02, 0.3, 0.4, 0.05, 1.2, 0.13, 0.14, 0.015, 0.16, 0.5]
child_sampler = ds.WeightedRandomSampler(weights, num_samples=12)
child_sampler = ds.WeightedRandomSampler(weights, num_samples=24)
sampler.add_child(child_sampler)
# Create ImageFolderDataset with sampler
data1 = ds.ImageFolderDataset(DATA_DIR, sampler=sampler)
@ -566,12 +591,17 @@ def skip_test_chained_sampler_05():
num_iter += 1
logger.info("Number of data in data1: {}".format(num_iter))
# Note: PKSampler produces 4x3=12 samples
# Note: Child WeightedRandomSampler produces 12 samples
# Note: Child WeightedRandomSampler produces 24 samples
# Note: DistributedSampler produces 24/2=12 samples
assert num_iter == 12
def test_chained_sampler_06():
"""
Feature: Chained Sampler
Description: Chained Samplers: WeightedRandom and PKSampler
Expectation: Get correct number of data
"""
logger.info("Test Case Chained Sampler - WeightedRandom and PKSampler")
# Create chained sampler, WeightedRandom and PKSampler
@ -602,6 +632,11 @@ def test_chained_sampler_06():
def test_chained_sampler_07():
"""
Feature: Chained Sampler
Description: Chained Samplers: SubsetRandom and Distributed, 2 shards
Expectation: Get correct number of data
"""
logger.info("Test Case Chained Sampler - SubsetRandom and Distributed, 2 shards")
# Create chained sampler, subset random and distributed
@ -628,17 +663,20 @@ def test_chained_sampler_07():
logger.info("Number of data in data1: {}".format(num_iter))
# Note: SubsetRandomSampler produces 12 samples
# Note: Each of 2 shards has 6 samples
# FIXME: Uncomment the following assert when code issue is resolved; at runtime, number of samples is 12 not 6
# assert num_iter == 6
assert num_iter == 12
def skip_test_chained_sampler_08():
def test_chained_sampler_08():
"""
Feature: Chained Sampler
Description: Chained Samplers: SubsetRandom and Distributed, 4 shards
Expectation: Get correct number of data
"""
logger.info("Test Case Chained Sampler - SubsetRandom and Distributed, 4 shards")
# Create chained sampler, subset random and distributed
indices = [0, 1, 2, 3, 4, 5, 12, 13, 14, 15, 16, 11]
sampler = ds.SubsetRandomSampler(indices, num_samples=12)
indices = [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10]
sampler = ds.SubsetRandomSampler(indices, num_samples=11)
child_sampler = ds.DistributedSampler(num_shards=4, shard_id=1)
sampler.add_child(child_sampler)
# Create ImageFolderDataset with sampler
@ -647,7 +685,7 @@ def skip_test_chained_sampler_08():
# Verify dataset size
data1_size = data1.get_dataset_size()
logger.info("dataset size is: {}".format(data1_size))
assert data1_size == 3
assert data1_size == 11
# Verify number of iterations
num_iter = 0
@ -658,9 +696,8 @@ def skip_test_chained_sampler_08():
num_iter += 1
logger.info("Number of data in data1: {}".format(num_iter))
# Note: SubsetRandomSampler returns 12 samples
# Note: Each of 4 shards has 3 samples
assert num_iter == 3
# Note: SubsetRandomSampler returns 11 samples
assert num_iter == 11
def test_imagefolder_rename():
@ -809,8 +846,8 @@ if __name__ == '__main__':
test_chained_sampler_04()
logger.info('test_chained_sampler_04 Ended.\n')
# test_chained_sampler_05()
# logger.info('test_chained_sampler_05 Ended.\n')
test_chained_sampler_05()
logger.info('test_chained_sampler_05 Ended.\n')
test_chained_sampler_06()
logger.info('test_chained_sampler_06 Ended.\n')
@ -818,8 +855,8 @@ if __name__ == '__main__':
test_chained_sampler_07()
logger.info('test_chained_sampler_07 Ended.\n')
# test_chained_sampler_08()
# logger.info('test_chained_sampler_07 Ended.\n')
test_chained_sampler_08()
logger.info('test_chained_sampler_08 Ended.\n')
test_imagefolder_numshards()
logger.info('test_imagefolder_numshards Ended.\n')

17
tests/ut/python/dataset/test_sampler.py
View File

@ -1,4 +1,4 @@
# Copyright 2020-2021 Huawei Technologies Co., Ltd
# Copyright 2020-2022 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.
@ -248,9 +248,9 @@ def test_sampler_chain():
manifest_file = "../data/dataset/testManifestData/test5trainimgs.json"
map_ = {(172876, 0): 0, (54214, 0): 1, (54214, 1): 2, (173673, 0): 3, (64631, 1): 4}
def test_config(num_shards, shard_id):
def test_config(num_shards, shard_id, start_index=0):
sampler = ds.DistributedSampler(num_shards, shard_id, shuffle=False, num_samples=5)
child_sampler = ds.SequentialSampler()
child_sampler = ds.SequentialSampler(start_index)
sampler.add_child(child_sampler)
data1 = ds.ManifestDataset(manifest_file, sampler=sampler)
@ -269,6 +269,13 @@ def test_sampler_chain():
assert test_config(5, 2) == [2]
assert test_config(5, 3) == [3]
assert test_config(5, 4) == [4]
assert test_config(2, 0, 1) == [1, 3]
assert test_config(2, 1, 1) == [2, 4]
assert test_config(5, 0, 1) == [1]
assert test_config(5, 1, 1) == [2]
assert test_config(5, 2, 1) == [3]
assert test_config(5, 3, 1) == [4]
assert test_config(5, 4, 1) == [1]
def test_add_sampler_invalid_input():
@ -286,13 +293,13 @@ def test_add_sampler_invalid_input():
sampler = ds.SequentialSampler()
with pytest.raises(RuntimeError) as info:
data2 = ds.ManifestDataset(manifest_file, sampler=sampler, num_samples=20)
_ = ds.ManifestDataset(manifest_file, sampler=sampler, num_samples=20)
assert "sampler and num_samples cannot be specified at the same time" in str(info.value)
def test_distributed_sampler_invalid_offset():
with pytest.raises(RuntimeError) as info:
sampler = ds.DistributedSampler(num_shards=4, shard_id=0, shuffle=False, num_samples=None, offset=5).parse()
_ = ds.DistributedSampler(num_shards=4, shard_id=0, shuffle=False, num_samples=None, offset=5).parse()
assert "DistributedSampler: offset must be no more than num_shards(4)" in str(info.value)

101
tests/ut/python/dataset/test_sampler_chain.py
View File

@ -1,4 +1,4 @@
# Copyright 2020 Huawei Technologies Co., Ltd
# Copyright 2020-2022 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.
@ -12,6 +12,7 @@
# See the License for the specific language governing permissions and
# limitations under the License.
# ==============================================================================
import numpy as np
import pytest
import mindspore.dataset as ds
import mindspore.dataset.transforms.c_transforms as c_transforms
@ -35,7 +36,9 @@ VOC_DATA_DIR = "../data/dataset/testVOC2012"
def test_numpyslices_sampler_no_chain():
"""
Test NumpySlicesDataset with sampler, no chain
Feature: Chained Sampler
Description: NumpySlicesDataset with sampler, no chain
Expectation: Data verified to be correct
"""
logger.info("test_numpyslices_sampler_no_chain")
@ -59,10 +62,14 @@ def test_numpyslices_sampler_no_chain():
res.append(item)
logger.info("dataset: {}".format(res))
np.testing.assert_array_equal(res, [[2], [3]])
def test_numpyslices_sampler_chain():
"""
Test NumpySlicesDataset sampler chain
Feature: Chained Sampler
Description: NumpySlicesDataset with sampler chain; add child sampler with 1 statement
Expectation: Data verified to be correct
"""
logger.info("test_numpyslices_sampler_chain")
@ -88,10 +95,14 @@ def test_numpyslices_sampler_chain():
res.append(item)
logger.info("dataset: {}".format(res))
np.testing.assert_array_equal(res, [[3]])
def test_numpyslices_sampler_chain2():
"""
Test NumpySlicesDataset sampler chain
Feature: Chained Sampler
Description: NumpySlicesDataset with sampler chain; add child sampler with 2 statements
Expectation: Data verified to be correct
"""
logger.info("test_numpyslices_sampler_chain2")
@ -118,6 +129,50 @@ def test_numpyslices_sampler_chain2():
res.append(item)
logger.info("dataset: {}".format(res))
np.testing.assert_array_equal(res, [[3]])
def test_numpyslices_sampler_chain_multi_add_child():
"""
Feature: Chained Sampler
Description: NumpySlicesDataset with sampler chain with multiple add_child() invocations
Expectation: Data verified to be correct. Only last add_child() invocation is effective.
"""
logger.info("test_numpyslices_sampler_chain_multi_add_child")
# Create NumpySlicesDataset with sampler chain
# Call add_child() multiple times in succession
# Note: A subsequent add_child() invocation replaces the prior child sampler (if any).
np_data = [1, 2, 3, 4, 5, 6, 7, 8]
sampler = ds.SequentialSampler(start_index=1, num_samples=None)
# 1st add_child invocation
sampler.add_child(ds.SequentialSampler(start_index=4, num_samples=1))
# 2nd add_child invocation
sampler.add_child(ds.SequentialSampler(start_index=4, num_samples=2))
# 3rd add_child invocation
sampler.add_child(ds.SequentialSampler(start_index=4, num_samples=3))
# 4th and last add_child invocation which is the effective child sampler
sampler.add_child(ds.SequentialSampler(start_index=1, num_samples=6))
data1 = ds.NumpySlicesDataset(np_data, sampler=sampler)
# Verify dataset size
data1_size = data1.get_dataset_size()
logger.info("dataset size is: {}".format(data1_size))
assert data1_size == 5
# Verify number of rows
assert sum([1 for _ in data1]) == 5
# Verify dataset contents
res = []
for item in data1.create_tuple_iterator(num_epochs=1, output_numpy=True):
logger.info("item: {}".format(item))
res.append(item)
logger.info("dataset: {}".format(res))
np.testing.assert_array_equal(res, [[3], [4], [5], [6], [7]])
def test_imagefolder_sampler_chain():
"""
@ -287,10 +342,10 @@ def test_numpyslices_sampler_chain_batch():
# Create NumpySlicesDataset with sampler chain
np_data = [1, 2, 3, 4, 5, 6, 7, 8, 9, 10]
sampler = ds.SequentialSampler(start_index=1, num_samples=3)
sampler = sampler.add_child(ds.SequentialSampler(start_index=1, num_samples=2))
sampler = ds.SequentialSampler(start_index=1, num_samples=8)
sampler.add_child(ds.SequentialSampler(start_index=1, num_samples=9))
data1 = ds.NumpySlicesDataset(np_data, sampler=sampler)
data1 = data1.batch(batch_size=3, drop_remainder=False)
data1 = data1.batch(batch_size=2, drop_remainder=False)
# Verify dataset size
data1_size = data1.get_dataset_size()
@ -307,6 +362,8 @@ def test_numpyslices_sampler_chain_batch():
res.append(item)
logger.info("dataset: {}".format(res))
np.testing.assert_array_equal(res, [[[3, 4]], [[5, 6]], [[7, 8]], [[9, 10]]])
def test_sampler_chain_errors():
"""
@ -321,19 +378,6 @@ def test_sampler_chain_errors():
with pytest.raises(AttributeError, match=error_msg_1):
sampler.add_child(ds.SequentialSampler(start_index=1, num_samples=2))
# error_msg_2 = "'NoneType' object has no attribute 'add_child'"
# Test add second and nested child sampler
sampler = ds.SequentialSampler(start_index=1, num_samples=2)
child_sampler = ds.SequentialSampler(start_index=1, num_samples=2)
sampler.add_child(child_sampler)
child_sampler2 = ds.SequentialSampler(start_index=1, num_samples=2)
sampler.add_child(child_sampler2)
# FIXME - no error is raised; uncomment after code issue is resolved
# with pytest.raises(AttributeError, match=error_msg_2):
# sampler.add_child(child_sampler2)
# np_data = [1, 2, 3, 4, 5, 6, 7, 8, 9, 10]
# data1 = ds.NumpySlicesDataset(np_data, sampler=sampler)
error_msg_3 = "Conflicting arguments during sampler assignments."
# Test conflicting arguments (sampler and shuffle=False) for sampler (no chain)
np_data = [1, 2, 3, 4, 5, 6, 7, 8, 9, 10]
@ -341,14 +385,13 @@ def test_sampler_chain_errors():
with pytest.raises(ValueError, match=error_msg_3):
ds.NumpySlicesDataset(np_data, shuffle=False, sampler=sampler)
# error_msg_4 = "Conflicting arguments during sampler assignments."
error_msg_4 = "Conflicting arguments during sampler assignments."
# Test conflicting arguments (sampler and shuffle=False) for sampler chaining
np_data = [1, 2, 3, 4, 5, 6, 7, 8, 9, 10]
sampler = ds.SequentialSampler(start_index=1, num_samples=3)
sampler = sampler.add_child(ds.SequentialSampler(start_index=1, num_samples=2))
# FIXME - no error is raised; uncomment after code issue is resolved
# with pytest.raises(ValueError, match=error_msg_4):
# ds.NumpySlicesDataset(np_data, shuffle=False, sampler=sampler)
sampler.add_child(ds.SequentialSampler(start_index=1, num_samples=2))
with pytest.raises(ValueError, match=error_msg_4):
ds.NumpySlicesDataset(np_data, shuffle=False, sampler=sampler)
def test_manifest_sampler_chain_repeat():
@ -396,23 +439,23 @@ def test_manifest_sampler_chain_batch_repeat():
data1 = ds.ManifestDataset(manifest_file, decode=True, sampler=sampler)
one_hot_encode = c_transforms.OneHot(3)
data1 = data1.map(operations=one_hot_encode, input_columns=["label"])
data1 = data1.batch(batch_size=5, drop_remainder=False)
data1 = data1.batch(batch_size=1, drop_remainder=False)
data1 = data1.repeat(count=2)
# Verify dataset size
data1_size = data1.get_dataset_size()
logger.info("dataset size is: {}".format(data1_size))
assert data1_size == 2
assert data1_size == 10
# Verify number of rows
# FIXME: Uncomment the following assert when code issue is resolved
# assert sum([1 for _ in data1]) == 2
assert sum([1 for _ in data1]) == 10
if __name__ == '__main__':
test_numpyslices_sampler_no_chain()
test_numpyslices_sampler_chain()
test_numpyslices_sampler_chain2()
test_numpyslices_sampler_chain_multi_add_child()
test_imagefolder_sampler_chain()
test_mnist_sampler_chain()
test_manifest_sampler_chain()