mindspore-Ecosystem
/
mindspore
21
0
Fork 27
Code Issues Pull Requests 6 Projects Releases Wiki Activity

add log warning and op error check

This commit is contained in:
jonyguo 2022-03-24 10:24:10 +08:00
parent 7ab93dcb37
commit 317c48a933
4 changed files with 169 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

18
mindspore/ccsrc/plugin/device/ascend/hal/device/ascend_kernel_runtime.cc
View File

@ -14,6 +14,7 @@
* limitations under the License.
*/
#include "plugin/device/ascend/hal/device/ascend_kernel_runtime.h"
#include <locale>
#include <string>
#include <vector>
#include <memory>
@ -719,6 +720,23 @@ void AscendKernelRuntime::DumpTaskExceptionInfo(const session::KernelGraph &) {
auto full_scope_name = node->fullname_with_scope();
MS_LOG(ERROR) << "Dump node (" << full_scope_name << ") task error input/output data to: " << path
<< trace::DumpSourceLines(node);
// full_scope_name: Default/GetNext-op1
std::string lower_full_scope_name(full_scope_name.length(), ' ');
(void)std::transform(full_scope_name.begin(), full_scope_name.end(), lower_full_scope_name.begin(), ::tolower);
if (lower_full_scope_name.find("getnext") != std::string::npos) {
MS_LOG(WARNING) << "GetNext error may be caused by slow data processing (bigger than 20s / batch) or "
<< "transfer data to device error.";
MS_LOG(WARNING) << "Suggestion: ";
MS_LOG(WARNING) << " 1) Set the parameter dataset_sink_mode=False of model.train(...) or "
<< "model.eval(...) and try again.";
MS_LOG(WARNING) << " 2) Reduce the batch_size in data processing and try again.";
MS_LOG(WARNING) << " 3) You can create iterator by interface create_dict_iterator() of dataset class to "
<< "independently verify the performance of data processing without training. "
<< "Refer to the link for data processing optimization suggestions: "
<< "https://www.mindspore.cn/docs/programming_guide/zh-CN/r1.6/optimize_data_processing.html";
}
E2eDump::DumpInputData(node, false, path, &full_scope_name);
E2eDump::DumpOutputData(node, false, path, &full_scope_name);
}

14
mindspore/python/mindspore/dataset/engine/datasets.py
View File

@ -1,4 +1,4 @@
# Copyright 2019-2022 Huawei Technologies Co., Ltd
# 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.
@ -55,6 +55,7 @@ from mindspore import log as logger
from mindspore.parallel._ps_context import _is_role_pserver, _is_role_sched
from mindspore.dataset.engine.offload import GetOffloadModel
import mindspore.dataset.transforms.c_transforms as c_transforms
import mindspore.dataset.transforms.py_transforms as py_transforms
from mindspore.dataset.text.utils import SentencePieceModel, DE_C_INTER_SENTENCEPIECE_MODE
from mindspore.parallel._utils import _get_device_num
@ -3180,6 +3181,17 @@ class MapDataset(UnionBaseDataset):
offload=None):
super().__init__(children=input_dataset, num_parallel_workers=num_parallel_workers, cache=cache)
self.operations = to_list(operations)
for op in self.operations:
# user define c_vision.HWC2CHW without parentheses is error
if type(op) == type: # pylint: disable=unidiomatic-typecheck
raise ValueError("Parameter operations's element of method map should be a dataset processing "
"operation instance, but got: {}. It may be missing parentheses for "
"instantiation.".format(op))
if not isinstance(op, (c_transforms.TensorOperation, py_transforms.PyTensorOperation)) \
and not callable(op):
raise ValueError("Parameter operations's element of method map should be a python function or "
"class method which should be callable, but got: {}. It doesn't need parentheses "
"for python function or class method.".format(op))
self.operations = py_transforms.Compose.reduce(self.operations)
self.input_columns = to_list(input_columns)
self.output_columns = to_list(output_columns)

13
mindspore/python/mindspore/dataset/transforms/validators.py
View File

@ -219,6 +219,7 @@ def check_random_transform_ops(method):
raise ValueError("op_list can not be empty.")
for ind, op in enumerate(arg_list[0]):
check_tensor_op(op, "op_list[{0}]".format(ind))
check_transform_op_type(ind, op)
if len(arg_list) == 2: # random apply takes an additional arg
type_check(arg_list[1], (float, int), "prob")
check_value(arg_list[1], (0, 1), "prob")
@ -227,6 +228,15 @@ def check_random_transform_ops(method):
return new_method
def check_transform_op_type(ind, op):
"""Check the operator."""
# c_vision.HWC2CHW error
# py_vision.HWC2CHW error
if type(op) == type: # pylint: disable=unidiomatic-typecheck
raise ValueError("op_list[{}] should be a dataset processing operation instance, "
"but got: {}. It may be missing parentheses for instantiation.".format(ind, op))
def check_compose_list(method):
"""Wrapper method to check the transform list of Python Compose."""
@ -240,6 +250,7 @@ def check_compose_list(method):
for i, transform in enumerate(transforms):
if not callable(transform):
raise ValueError("transforms[{}] is not callable.".format(i))
check_transform_op_type(i, transform)
return method(self, *args, **kwargs)
return new_method
@ -274,6 +285,7 @@ def check_random_apply(method):
raise ValueError(
"transforms[{}] is not a py transforms. Should not use a c transform in py transform" \
.format(i))
check_transform_op_type(i, transform)
if prob is not None:
type_check(prob, (float, int,), "prob")
@ -297,6 +309,7 @@ def check_transforms_list(method):
raise ValueError(
"transforms[{}] is not a py transforms. Should not use a c transform in py transform" \
.format(i))
check_transform_op_type(i, transform)
return method(self, *args, **kwargs)
return new_method

125
tests/ut/python/dataset/test_map.py Normal file
View File

@ -0,0 +1,125 @@
# 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 pytest
import mindspore.dataset as ds
from mindspore.dataset.transforms import c_transforms
from mindspore.dataset.transforms import py_transforms
import mindspore.dataset.vision.c_transforms as c_vision
import mindspore.dataset.vision.py_transforms as py_vision
DATA_DIR = "../data/dataset/testPK/data"
def test_map_c_transform_exception():
"""
Feature: test c error op def
Description: op defined like c_vision.HWC2CHW
Expectation: success
"""
data_set = ds.ImageFolderDataset(DATA_DIR, num_parallel_workers=1, shuffle=True)
train_image_size = 224
mean = [0.485 * 255, 0.456 * 255, 0.406 * 255]
std = [0.229 * 255, 0.224 * 255, 0.225 * 255]
# define map operations
random_crop_decode_resize_op = c_vision.RandomCropDecodeResize(train_image_size,
scale=(0.08, 1.0),
ratio=(0.75, 1.333))
random_horizontal_flip_op = c_vision.RandomHorizontalFlip(prob=0.5)
normalize_op = c_vision.Normalize(mean=mean, std=std)
hwc2chw_op = c_vision.HWC2CHW # exception
data_set = data_set.map(operations=random_crop_decode_resize_op, input_columns="image", num_parallel_workers=1)
data_set = data_set.map(operations=random_horizontal_flip_op, input_columns="image", num_parallel_workers=1)
data_set = data_set.map(operations=normalize_op, input_columns="image", num_parallel_workers=1)
with pytest.raises(ValueError) as info:
data_set = data_set.map(operations=hwc2chw_op, input_columns="image", num_parallel_workers=1)
assert "Parameter operations's element of method map should be a " in str(info.value)
# compose exception
with pytest.raises(ValueError) as info:
c_transforms.Compose([
c_vision.RandomCropDecodeResize(train_image_size, scale=(0.08, 1.0), ratio=(0.75, 1.333)),
c_vision.RandomHorizontalFlip,
c_vision.Normalize(mean=mean, std=std),
c_vision.HWC2CHW()])
assert " should be a " in str(info.value)
# randomapply exception
with pytest.raises(ValueError) as info:
c_transforms.RandomApply([
c_vision.RandomCropDecodeResize,
c_vision.RandomHorizontalFlip(prob=0.5),
c_vision.Normalize(mean=mean, std=std),
c_vision.HWC2CHW()])
assert " should be a " in str(info.value)
# randomchoice exception
with pytest.raises(ValueError) as info:
c_transforms.RandomChoice([
c_vision.RandomCropDecodeResize(train_image_size, scale=(0.08, 1.0), ratio=(0.75, 1.333)),
c_vision.RandomHorizontalFlip(prob=0.5),
c_vision.Normalize,
c_vision.HWC2CHW()])
assert " should be a " in str(info.value)
def test_map_py_transform_exception():
"""
Feature: test python error op def
Description: op defined like py_vision.RandomHorizontalFlip
Expectation: success
"""
data_set = ds.ImageFolderDataset(DATA_DIR, num_parallel_workers=1, shuffle=True)
# define map operations
decode_op = py_vision.Decode()
random_horizontal_flip_op = py_vision.RandomHorizontalFlip # exception
to_tensor_op = py_vision.ToTensor()
trans = [decode_op, random_horizontal_flip_op, to_tensor_op]
with pytest.raises(ValueError) as info:
data_set = data_set.map(operations=trans, input_columns="image", num_parallel_workers=1)
assert "Parameter operations's element of method map should be a " in str(info.value)
# compose exception
with pytest.raises(ValueError) as info:
py_transforms.Compose([
py_vision.Decode,
py_vision.RandomHorizontalFlip(),
py_vision.ToTensor()])
assert " should be a " in str(info.value)
# randomapply exception
with pytest.raises(ValueError) as info:
py_transforms.RandomApply([
py_vision.Decode(),
py_vision.RandomHorizontalFlip,
py_vision.ToTensor()])
assert " should be a " in str(info.value)
# randomchoice exception
with pytest.raises(ValueError) as info:
py_transforms.RandomChoice([
py_vision.Decode(),
py_vision.RandomHorizontalFlip(),
py_vision.ToTensor])
assert " should be a " in str(info.value)
if __name__ == '__main__':
test_map_c_transform_exception()
test_map_py_transform_exception()