m53297601
/
mindspore
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

del: zhwiki & enwiki & CLUERNER2020 preprocess script

Browse Source
pull/2796/head
jonyguo 5 years ago
parent 622b97f3b6
commit d0cd4cb83c
36 changed files with 0 additions and 45082 deletions
Show Stats Download Patch File Download Diff File

82
model_zoo/utils/nlp_to_mindrecord/CLUERNER2020/README.md
Unescape View File

@ -1,82 +0,0 @@
# Guideline to Convert Training Data CLUERNER2020 to MindRecord For Bert Fine Tuning
<!-- TOC -->
- [What does the example do](#what-does-the-example-do)
- [How to use the example to process CLUERNER2020](#how-to-use-the-example-to-process-cluerner2020)
- [Download CLUERNER2020 and unzip](#download-cluerner2020-and-unzip)
- [Generate MindRecord](#generate-mindrecord)
- [Create MindDataset By MindRecord](#create-minddataset-by-mindrecord)
<!-- /TOC -->
## What does the example do
This example is based on [CLUERNER2020](https://www.cluebenchmarks.com/introduce.html) training data, generating MindRecord file, and finally used for Bert Fine Tuning progress.
1. run.sh: generate MindRecord entry script
2. run_read.py: create MindDataset by MindRecord entry script.
- create_dataset.py: use MindDataset to read MindRecord to generate dataset.
## How to use the example to process CLUERNER2020
Download CLUERNER2020, convert it to MindRecord, use MindDataset to read MindRecord.
### Download CLUERNER2020 and unzip
1. Download the training data zip.
> [CLUERNER2020 dataset download address](https://www.cluebenchmarks.com/introduce.html) **-> 任务介绍 -> CLUENER 细粒度命名实体识别 -> cluener下载链接**
2. Unzip the training data to dir example/nlp_to_mindrecord/CLUERNER2020/cluener_public.
```
unzip -d {your-mindspore}/example/nlp_to_mindrecord/CLUERNER2020/data/cluener_public cluener_public.zip
```
### Generate MindRecord
1. Run the run.sh script.
```bash
bash run.sh
```
2. Output like this:
```
...
[INFO] ME(17603:139620983514944,MainProcess):2020-04-28-16:56:12.498.235 [mindspore/mindrecord/filewriter.py:313] The list of mindrecord files created are: ['data/train.mindrecord'], and the list of index files are: ['data/train.mindrecord.db']
...
[INFO] ME(17603,python):2020-04-28-16:56:13.400.175 [mindspore/ccsrc/mindrecord/io/shard_writer.cc:667] WriteRawData] Write 1 records successfully.
[INFO] ME(17603,python):2020-04-28-16:56:13.400.863 [mindspore/ccsrc/mindrecord/io/shard_writer.cc:667] WriteRawData] Write 1 records successfully.
[INFO] ME(17603,python):2020-04-28-16:56:13.401.534 [mindspore/ccsrc/mindrecord/io/shard_writer.cc:667] WriteRawData] Write 1 records successfully.
[INFO] ME(17603,python):2020-04-28-16:56:13.402.179 [mindspore/ccsrc/mindrecord/io/shard_writer.cc:667] WriteRawData] Write 1 records successfully.
[INFO] ME(17603,python):2020-04-28-16:56:13.402.702 [mindspore/ccsrc/mindrecord/io/shard_writer.cc:667] WriteRawData] Write 1 records successfully.
...
[INFO] ME(17603:139620983514944,MainProcess):2020-04-28-16:56:13.431.208 [mindspore/mindrecord/filewriter.py:313] The list of mindrecord files created are: ['data/dev.mindrecord'], and the list of index files are: ['data/dev.mindrecord.db']
```
3. Generate files like this:
```bash
$ ls output/
dev.mindrecord dev.mindrecord.db README.md train.mindrecord train.mindrecord.db
```
### Create MindDataset By MindRecord
1. Run the run_read.sh script.
```bash
bash run_read.sh
```
2. Output like this:
```
...
example 1340: input_ids: [ 101 3173 1290 4852 7676 3949 122 3299 123 126 3189 4510 8020 6381 5442 7357 2590 3636 8021 7676 3949 4294 1166 6121 3124 1277 6121 3124 7270 2135 3295 5789 3326 123 126 3189 1355 6134 1093 1325 3173 2399 6590 6791 8024 102 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0]
example 1340: input_mask: [1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0]
example 1340: segment_ids: [0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0]
example 1340: label_ids: [ 0 18 19 20 2 4 0 0 0 0 0 0 0 34 36 26 27 28 0 34 35 35 35 35 35 35 35 35 35 36 26 27 28 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0]
example 1341: input_ids: [ 101 1728 711 4293 3868 1168 2190 2150 3791 934 3633 3428 4638 6237 7025 8024 3297 1400 5310 3362 6206 5023 5401 1744 3297 7770 3791 7368 976 1139 1104 2137 511 102 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0]
example 1341: input_mask: [1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0]
example 1341: segment_ids: [0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0]
example 1341: label_ids: [ 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 18 19 19 19 19 20 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0]
...
```

36
model_zoo/utils/nlp_to_mindrecord/CLUERNER2020/create_dataset.py
Unescape View File

@ -1,36 +0,0 @@
# 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.
# ============================================================================
"""create MindDataset by MindRecord"""
import mindspore.dataset as ds
def create_dataset(data_file):
"""create MindDataset"""
num_readers = 4
data_set = ds.MindDataset(dataset_file=data_file, num_parallel_workers=num_readers, shuffle=True)
index = 0
for item in data_set.create_dict_iterator():
# print("example {}: {}".format(index, item))
print("example {}: input_ids: {}".format(index, item['input_ids']))
print("example {}: input_mask: {}".format(index, item['input_mask']))
print("example {}: segment_ids: {}".format(index, item['segment_ids']))
print("example {}: label_ids: {}".format(index, item['label_ids']))
index += 1
if index % 1000 == 0:
print("read rows: {}".format(index))
print("total rows: {}".format(index))
if __name__ == '__main__':
create_dataset('output/train.mindrecord')
create_dataset('output/dev.mindrecord')

1
model_zoo/utils/nlp_to_mindrecord/CLUERNER2020/data/.gitignore vendored
Unescape View File

@ -1 +0,0 @@
cluener_public

1
model_zoo/utils/nlp_to_mindrecord/CLUERNER2020/data/README.md
Unescape View File

@ -1 +0,0 @@
## The input dataset

1
model_zoo/utils/nlp_to_mindrecord/CLUERNER2020/output/README.md
Unescape View File

@ -1 +0,0 @@
## output dir

40
model_zoo/utils/nlp_to_mindrecord/CLUERNER2020/run.sh
Unescape View File

@ -1,40 +0,0 @@
#!/bin/bash
# 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.
# ============================================================================
rm -f output/train.mindrecord*
rm -f output/dev.mindrecord*
if [ ! -d "../../../../third_party/to_mindrecord/CLUERNER2020" ]; then
echo "The patch base dir ../../../../third_party/to_mindrecord/CLUERNER2020 is not exist."
exit 1
fi
if [ ! -f "../../../../third_party/patch/to_mindrecord/CLUERNER2020/data_processor_seq.patch" ]; then
echo "The patch file ../../../../third_party/patch/to_mindrecord/CLUERNER2020/data_processor_seq.patch is not exist."
exit 1
fi
# patch for data_processor_seq.py
patch -p0 -d ../../../../third_party/to_mindrecord/CLUERNER2020/ -o data_processor_seq_patched.py < ../../../../third_party/patch/to_mindrecord/CLUERNER2020/data_processor_seq.patch
if [ $? -ne 0 ]; then
echo "Patch ../../../../third_party/to_mindrecord/CLUERNER2020/data_processor_seq.py failed"
exit 1
fi
# use patched script
python ../../../../third_party/to_mindrecord/CLUERNER2020/data_processor_seq_patched.py \
--vocab_file=../../../../third_party/to_mindrecord/CLUERNER2020/vocab.txt \
--label2id_file=../../../../third_party/to_mindrecord/CLUERNER2020/label2id.json

17
model_zoo/utils/nlp_to_mindrecord/CLUERNER2020/run_read.sh
Unescape View File

@ -1,17 +0,0 @@
#!/bin/bash
# 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.
# ============================================================================
python create_dataset.py

173
model_zoo/utils/nlp_to_mindrecord/enwiki/README.md
Unescape View File

@ -1,173 +0,0 @@
# Guideline to Convert Training Data enwiki to MindRecord For Bert Pre Training
<!-- TOC -->
- [What does the example do](#what-does-the-example-do)
- [How to use the example to process enwiki](#how-to-use-the-example-to-process-enwiki)
- [Download enwiki training data](#download-enwiki-training-data)
- [Process the enwiki](#process-the-enwiki)
- [Generate MindRecord](#generate-mindrecord)
- [Create MindDataset By MindRecord](#create-minddataset-by-mindrecord)
<!-- /TOC -->
## What does the example do
This example is based on [enwiki](https://dumps.wikimedia.org/enwiki) training data, generating MindRecord file, and finally used for Bert network training.
1. run.sh: generate MindRecord entry script.
2. run_read.py: create MindDataset by MindRecord entry script.
- create_dataset.py: use MindDataset to read MindRecord to generate dataset.
## How to use the example to process enwiki
Download enwiki data, process it, convert it to MindRecord, use MindDataset to read MindRecord.
### Download enwiki training data
> [enwiki dataset download address](https://dumps.wikimedia.org/enwiki) **-> 20200501 -> enwiki-20200501-pages-articles-multistream.xml.bz2**
### Process the enwiki
1. Please follow the steps in [process enwiki](https://github.com/mlperf/training/tree/master/language_model/tensorflow/bert)
- All permissions of this step belong to the link address website.
### Generate MindRecord
1. Run the run.sh script.
```
bash run.sh input_dir output_dir vocab_file
```
- input_dir: the directory which contains files like 'part-00251-of-00500'.
- output_dir: which will store the output mindrecord files.
- vocab_file: the vocab file which you can download from other opensource project.
2. The output like this:
```
...
Begin preprocess Wed Jun 10 09:21:23 CST 2020
Begin preprocess input file: /mnt/data/results/part-00000-of-00500
Begin output file: part-00000-of-00500.mindrecord
Total task: 510, processing: 1
Begin preprocess input file: /mnt/data/results/part-00001-of-00500
Begin output file: part-00001-of-00500.mindrecord
Total task: 510, processing: 2
Begin preprocess input file: /mnt/data/results/part-00002-of-00500
Begin output file: part-00002-of-00500.mindrecord
Total task: 510, processing: 3
Begin preprocess input file: /mnt/data/results/part-00003-of-00500
Begin output file: part-00003-of-00500.mindrecord
Total task: 510, processing: 4
Begin preprocess input file: /mnt/data/results/part-00004-of-00500
Begin output file: part-00004-of-00500.mindrecord
Total task: 510, processing: 4
...
```
3. Generate files like this:
```bash
$ ls {your_output_dir}/
part-00000-of-00500.mindrecord part-00000-of-00500.mindrecord.db part-00001-of-00500.mindrecord part-00001-of-00500.mindrecord.db part-00002-of-00500.mindrecord part-00002-of-00500.mindrecord.db ...
```
### Create MindDataset By MindRecord
1. Run the run_read.sh script.
```bash
bash run_read.sh input_dir
```
- input_dir: the directory which contains mindrecord files.
2. The output like this:
```
...
example 633: input_ids: [ 101 2043 19781 4305 2140 4520 2041 1010 103 2034 2455 2002
7879 2003 1996 2455 1997 103 26378 4160 1012 102 7291 2001
1996 103 1011 2343 1997 6327 1010 3423 1998 103 4262 2005
1996 2118 1997 2329 3996 103 102 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0]
example 633: input_mask: [1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
1 1 1 1 1 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0]
example 633: segment_ids: [0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
1 1 1 1 1 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0]
example 633: masked_lm_positions: [ 8 17 20 25 33 41 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0]
example 633: masked_lm_ids: [ 1996 16137 1012 3580 2451 1012 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0]
example 633: masked_lm_weights: [1. 1. 1. 1. 1. 1. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.
0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.
0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.
0. 0. 0. 0.]
example 633: next_sentence_labels: [1]
...
```

43
model_zoo/utils/nlp_to_mindrecord/enwiki/create_dataset.py
Unescape View File

@ -1,43 +0,0 @@
# 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.
# ============================================================================
"""create MindDataset by MindRecord"""
import argparse
import mindspore.dataset as ds
def create_dataset(data_file):
"""create MindDataset"""
num_readers = 4
data_set = ds.MindDataset(dataset_file=data_file, num_parallel_workers=num_readers, shuffle=True)
index = 0
for item in data_set.create_dict_iterator():
# print("example {}: {}".format(index, item))
print("example {}: input_ids: {}".format(index, item['input_ids']))
print("example {}: input_mask: {}".format(index, item['input_mask']))
print("example {}: segment_ids: {}".format(index, item['segment_ids']))
print("example {}: masked_lm_positions: {}".format(index, item['masked_lm_positions']))
print("example {}: masked_lm_ids: {}".format(index, item['masked_lm_ids']))
print("example {}: masked_lm_weights: {}".format(index, item['masked_lm_weights']))
print("example {}: next_sentence_labels: {}".format(index, item['next_sentence_labels']))
index += 1
if index % 1000 == 0:
print("read rows: {}".format(index))
print("total rows: {}".format(index))
if __name__ == '__main__':
parser = argparse.ArgumentParser()
parser.add_argument("-i", "--input_file", nargs='+', type=str, help='Input mindreord file')
args = parser.parse_args()
create_dataset(args.input_file)

133
model_zoo/utils/nlp_to_mindrecord/enwiki/run.sh
Unescape View File

@ -1,133 +0,0 @@
#!/bin/bash
# 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.
# ============================================================================
if [ $# -ne 3 ]; then
echo "Usage: $0 input_dir output_dir vocab_file"
exit 1
fi
if [ ! -d $1 ]; then
echo "The input dir: $1 is not exist."
exit 1
fi
if [ ! -d $2 ]; then
echo "The output dir: $2 is not exist."
exit 1
fi
rm -fr $2/*.mindrecord*
if [ ! -f $3 ]; then
echo "The vocab file: $3 is not exist."
exit 1
fi
data_dir=$1
output_dir=$2
vocab_file=$3
file_list=()
output_filename=()
file_index=0
function getdir() {
elements=`ls $1`
for element in ${elements[*]};
do
dir_or_file=$1"/"$element
if [ -d $dir_or_file ];
then
getdir $dir_or_file
else
file_list[$file_index]=$dir_or_file
echo "${dir_or_file}" | tr '/' '\n' > dir_file_list.txt # dir dir file to mapfile
mapfile parent_dir < dir_file_list.txt
rm dir_file_list.txt >/dev/null 2>&1
tmp_output_filename=${parent_dir[${#parent_dir[@]}-1]}".mindrecord"
output_filename[$file_index]=`echo ${tmp_output_filename} | sed 's/ //g'`
file_index=`expr $file_index + 1`
fi
done
}
getdir "${data_dir}"
# echo "The input files: "${file_list[@]}
# echo "The output files: "${output_filename[@]}
if [ ! -d "../../../../third_party/to_mindrecord/zhwiki" ]; then
echo "The patch base dir ../../../../third_party/to_mindrecord/zhwiki is not exist."
exit 1
fi
if [ ! -f "../../../../third_party/patch/to_mindrecord/zhwiki/create_pretraining_data.patch" ]; then
echo "The patch file ../../../../third_party/patch/to_mindrecord/zhwiki/create_pretraining_data.patch is not exist."
exit 1
fi
# patch for create_pretraining_data.py
patch -p0 -d ../../../../third_party/to_mindrecord/zhwiki/ -o create_pretraining_data_patched.py < ../../../../third_party/patch/to_mindrecord/zhwiki/create_pretraining_data.patch
if [ $? -ne 0 ]; then
echo "Patch ../../../../third_party/to_mindrecord/zhwiki/create_pretraining_data.py failed"
exit 1
fi
# get the cpu core count
num_cpu_core=`cat /proc/cpuinfo | grep "processor" | wc -l`
avaiable_core_size=`expr $num_cpu_core / 3 \* 2`
echo "Begin preprocess `date`"
# using patched script to generate mindrecord
file_list_len=`expr ${#file_list[*]} - 1`
for index in $(seq 0 $file_list_len); do
echo "Begin preprocess input file: ${file_list[$index]}"
echo "Begin output file: ${output_filename[$index]}"
python ../../../../third_party/to_mindrecord/zhwiki/create_pretraining_data_patched.py \
--input_file=${file_list[$index]} \
--output_file=${output_dir}/${output_filename[$index]} \
--partition_number=1 \
--vocab_file=${vocab_file} \
--do_lower_case=True \
--max_seq_length=512 \
--max_predictions_per_seq=76 \
--masked_lm_prob=0.15 \
--random_seed=12345 \
--dupe_factor=10 >/tmp/${output_filename[$index]}.log 2>&1 &
process_count=`ps -ef | grep create_pretraining_data_patched | grep -v grep | wc -l`
echo "Total task: ${#file_list[*]}, processing: ${process_count}"
if [ $process_count -ge $avaiable_core_size ]; then
while [ 1 ]; do
process_num=`ps -ef | grep create_pretraining_data_patched | grep -v grep | wc -l`
if [ $process_count -gt $process_num ]; then
process_count=$process_num
break;
fi
sleep 2
done
fi
done
process_num=`ps -ef | grep create_pretraining_data_patched | grep -v grep | wc -l`
while [ 1 ]; do
if [ $process_num -eq 0 ]; then
break;
fi
echo "There are still ${process_num} preprocess running ..."
sleep 2
process_num=`ps -ef | grep create_pretraining_data_patched | grep -v grep | wc -l`
done
echo "Preprocess all the data success."
echo "End preprocess `date`"

44
model_zoo/utils/nlp_to_mindrecord/enwiki/run_read.sh
Unescape View File

@ -1,44 +0,0 @@
#!/bin/bash
# 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.
# ============================================================================
if [ $# -ne 1 ]; then
echo "Usage: $0 input_dir"
exit 1
fi
if [ ! -d $1 ]; then
echo "The input dir: $1 is not exist."
exit 1
fi
file_list=()
file_index=0
# get all the mindrecord file from output dir
function getdir() {
elements=`ls $1/part-*.mindrecord`
for element in ${elements[*]};
do
file_list[$file_index]=$element
file_index=`expr $file_index + 1`
done
}
getdir $1
echo "Get all the mindrecord files: "${file_list[*]}
# create dataset for train
python create_dataset.py --input_file ${file_list[*]}

113
model_zoo/utils/nlp_to_mindrecord/zhwiki/README.md
Unescape View File

@ -1,113 +0,0 @@
# Guideline to Convert Training Data zhwiki to MindRecord For Bert Pre Training
<!-- TOC -->
- [What does the example do](#what-does-the-example-do)
- [Run simple test](#run-simple-test)
- [How to use the example to process zhwiki](#how-to-use-the-example-to-process-zhwiki)
- [Download zhwiki training data](#download-zhwiki-training-data)
- [Extract the zhwiki](#extract-the-zhwiki)
- [Generate MindRecord](#generate-mindrecord)
- [Create MindDataset By MindRecord](#create-minddataset-by-mindrecord)
<!-- /TOC -->
## What does the example do
This example is based on [zhwiki](https://dumps.wikimedia.org/zhwiki) training data, generating MindRecord file, and finally used for Bert network training.
1. run.sh: generate MindRecord entry script.
2. run_read.py: create MindDataset by MindRecord entry script.
- create_dataset.py: use MindDataset to read MindRecord to generate dataset.
## Run simple test
Follow the step:
```bash
bash run_simple.sh # generate output/simple.mindrecord* by ../../../../third_party/to_mindrecord/zhwiki/sample_text.txt
bash run_read_simple.sh # use MindDataset to read output/simple.mindrecord*
```
## How to use the example to process zhwiki
Download zhwiki data, extract it, convert it to MindRecord, use MindDataset to read MindRecord.
### Download zhwiki training data
> [zhwiki dataset download address](https://dumps.wikimedia.org/zhwiki) **-> 20200401 -> zhwiki-20200401-pages-articles-multistream.xml.bz2**
- put the zhwiki-20200401-pages-articles-multistream.xml.bz2 in {your-mindspore}/example/nlp_to_mindrecord/zhwiki/data directory.
### Extract the zhwiki
1. Download [wikiextractor](https://github.com/attardi/wikiextractor) script to {your-mindspore}/example/nlp_to_mindrecord/zhwiki/data directory.
```
$ ls data/
README.md wikiextractor zhwiki-20200401-pages-articles-multistream.xml.bz2
```
2. Extract the zhwiki.
```python
python data/wikiextractor/WikiExtractor.py data/zhwiki-20200401-pages-articles-multistream.xml.bz2 --processes 4 --templates data/template --bytes 8M --min_text_length 0 --filter_disambig_pages --output data/extract
```
3. Generate like this:
```
$ ls data/extract
AA AB
```
### Generate MindRecord
1. Run the run.sh script.
```
bash run.sh
```
> Caution: This process maybe slow, please wait patiently. If you do not have a machine with enough memory and cpu, it is recommended that you modify the script to generate mindrecord in step by step.
2. The output like this:
```
patching file create_pretraining_data_patched.py (read from create_pretraining_data.py)
Begin preprocess input file: ./data/extract/AA/wiki_00
Begin output file: AAwiki_00.mindrecord
Total task: 5, processing: 1
Begin preprocess input file: ./data/extract/AA/wiki_01
Begin output file: AAwiki_01.mindrecord
Total task: 5, processing: 2
Begin preprocess input file: ./data/extract/AA/wiki_02
Begin output file: AAwiki_02.mindrecord
Total task: 5, processing: 3
Begin preprocess input file: ./data/extract/AB/wiki_02
Begin output file: ABwiki_02.mindrecord
Total task: 5, processing: 4
...
```
3. Generate files like this:
```bash
$ ls output/
AAwiki_00.mindrecord AAwiki_00.mindrecord.db AAwiki_01.mindrecord AAwiki_01.mindrecord.db AAwiki_02.mindrecord AAwiki_02.mindrecord.db ... ABwiki_00.mindrecord ABwiki_00.mindrecord.db ...
```
### Create MindDataset By MindRecord
1. Run the run_read.sh script.
```bash
bash run_read.sh
```
2. The output like this:
```
...
example 74: input_ids: [ 101 8168 118 12847 8783 9977 15908 117 8256 9245 11643 8168 8847 8588 11575 8154 8228 143 8384 8376 9197 10241 103 10564 11421 8199 12268 112 161 8228 11541 9586 8436 8174 8363 9864 9702 103 103 119 103 9947 10564 103 8436 8806 11479 103 8912 119 103 103 103 12209 8303 103 8757 8824 117 8256 103 8619 8168 11541 102 11684 8196 103 8228 8847 11523 117 9059 9064 12410 8358 8181 10764 117 11167 11706 9920 148 8332 11390 8936 8205 10951 11997 103 8154 117 103 8670 10467 112 161 10951 13139 12413 117 10288 143 10425 8205 152 10795 8472 8196 103 161 12126 9172 13129 12106 8217 8174 12244 8205 143 103 8461 8277 10628 160 8221 119 102]
example 74: input_mask: [1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1]
example 74: segment_ids: [0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1]
example 74: masked_lm_positions: [ 6 22 37 38 40 43 47 50 51 52 55 60 67 76 89 92 98 109 120 0]
example 74: masked_lm_ids: [ 8118 8165 8329 8890 8554 8458 119 8850 8565 10392 8174 11467 10291 8181 8549 12718 13139 112 158 0]
example 74: masked_lm_weights: [1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 0.]
example 74: next_sentence_labels: [0]
...
```

43
model_zoo/utils/nlp_to_mindrecord/zhwiki/create_dataset.py
Unescape View File

@ -1,43 +0,0 @@
# 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.
# ============================================================================
"""create MindDataset by MindRecord"""
import argparse
import mindspore.dataset as ds
def create_dataset(data_file):
"""create MindDataset"""
num_readers = 4
data_set = ds.MindDataset(dataset_file=data_file, num_parallel_workers=num_readers, shuffle=True)
index = 0
for item in data_set.create_dict_iterator():
# print("example {}: {}".format(index, item))
print("example {}: input_ids: {}".format(index, item['input_ids']))
print("example {}: input_mask: {}".format(index, item['input_mask']))
print("example {}: segment_ids: {}".format(index, item['segment_ids']))
print("example {}: masked_lm_positions: {}".format(index, item['masked_lm_positions']))
print("example {}: masked_lm_ids: {}".format(index, item['masked_lm_ids']))
print("example {}: masked_lm_weights: {}".format(index, item['masked_lm_weights']))
print("example {}: next_sentence_labels: {}".format(index, item['next_sentence_labels']))
index += 1
if index % 1000 == 0:
print("read rows: {}".format(index))
print("total rows: {}".format(index))
if __name__ == '__main__':
parser = argparse.ArgumentParser()
parser.add_argument("-i", "--input_file", nargs='+', type=str, help='Input mindreord file')
args = parser.parse_args()
create_dataset(args.input_file)

3
model_zoo/utils/nlp_to_mindrecord/zhwiki/data/.gitignore vendored
Unescape View File

@ -1,3 +0,0 @@
wikiextractor/
zhwiki-20200401-pages-articles-multistream.xml.bz2
extract/

1
model_zoo/utils/nlp_to_mindrecord/zhwiki/data/README.md
Unescape View File

@ -1 +0,0 @@
## The input dataset

1
model_zoo/utils/nlp_to_mindrecord/zhwiki/output/README.md
Unescape View File

@ -1 +0,0 @@
## Output the mindrecord

112
model_zoo/utils/nlp_to_mindrecord/zhwiki/run.sh
Unescape View File

@ -1,112 +0,0 @@
#!/bin/bash
# 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.
# ============================================================================
rm -f output/*.mindrecord*
data_dir="./data/extract"
file_list=()
output_filename=()
file_index=0
function getdir() {
elements=`ls $1`
for element in ${elements[*]};
do
dir_or_file=$1"/"$element
if [ -d $dir_or_file ];
then
getdir $dir_or_file
else
file_list[$file_index]=$dir_or_file
echo "${dir_or_file}" | tr '/' '\n' > dir_file_list.txt # dir dir file to mapfile
mapfile parent_dir < dir_file_list.txt
rm dir_file_list.txt >/dev/null 2>&1
tmp_output_filename=${parent_dir[${#parent_dir[@]}-2]}${parent_dir[${#parent_dir[@]}-1]}".mindrecord"
output_filename[$file_index]=`echo ${tmp_output_filename} | sed 's/ //g'`
file_index=`expr $file_index + 1`
fi
done
}
getdir "${data_dir}"
# echo "The input files: "${file_list[@]}
# echo "The output files: "${output_filename[@]}
if [ ! -d "../../../../third_party/to_mindrecord/zhwiki" ]; then
echo "The patch base dir ../../../../third_party/to_mindrecord/zhwiki is not exist."
exit 1
fi
if [ ! -f "../../../../third_party/patch/to_mindrecord/zhwiki/create_pretraining_data.patch" ]; then
echo "The patch file ../../../../third_party/patch/to_mindrecord/zhwiki/create_pretraining_data.patch is not exist."
exit 1
fi
# patch for create_pretraining_data.py
patch -p0 -d ../../../../third_party/to_mindrecord/zhwiki/ -o create_pretraining_data_patched.py < ../../../../third_party/patch/to_mindrecord/zhwiki/create_pretraining_data.patch
if [ $? -ne 0 ]; then
echo "Patch ../../../../third_party/to_mindrecord/zhwiki/create_pretraining_data.py failed"
exit 1
fi
# get the cpu core count
num_cpu_core=`cat /proc/cpuinfo | grep "processor" | wc -l`
avaiable_core_size=`expr $num_cpu_core / 3 \* 2`
echo "Begin preprocess `date`"
# using patched script to generate mindrecord
file_list_len=`expr ${#file_list[*]} - 1`
for index in $(seq 0 $file_list_len); do
echo "Begin preprocess input file: ${file_list[$index]}"
echo "Begin output file: ${output_filename[$index]}"
python ../../../../third_party/to_mindrecord/zhwiki/create_pretraining_data_patched.py \
--input_file=${file_list[$index]} \
--output_file=output/${output_filename[$index]} \
--partition_number=1 \
--vocab_file=../../../../third_party/to_mindrecord/zhwiki/vocab.txt \
--do_lower_case=True \
--max_seq_length=128 \
--max_predictions_per_seq=20 \
--masked_lm_prob=0.15 \
--random_seed=12345 \
--dupe_factor=10 >/tmp/${output_filename[$index]}.log 2>&1 & # user defined
process_count=`ps -ef | grep create_pretraining_data_patched | grep -v grep | wc -l`
echo "Total task: ${#file_list[*]}, processing: ${process_count}"
if [ $process_count -ge $avaiable_core_size ]; then
while [ 1 ]; do
process_num=`ps -ef | grep create_pretraining_data_patched | grep -v grep | wc -l`
if [ $process_count -gt $process_num ]; then
process_count=$process_num
break;
fi
sleep 2
done
fi
done
process_num=`ps -ef | grep create_pretraining_data_patched | grep -v grep | wc -l`
while [ 1 ]; do
if [ $process_num -eq 0 ]; then
break;
fi
echo "There are still ${process_num} preprocess running ..."
sleep 2
process_num=`ps -ef | grep create_pretraining_data_patched | grep -v grep | wc -l`
done
echo "Preprocess all the data success."
echo "End preprocess `date`"

34
model_zoo/utils/nlp_to_mindrecord/zhwiki/run_read.sh
Unescape View File

@ -1,34 +0,0 @@
#!/bin/bash
# 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.
# ============================================================================
file_list=()
file_index=0
# get all the mindrecord file from output dir
function getdir() {
elements=`ls $1/[A-Z]*.mindrecord`
for element in ${elements[*]};
do
file_list[$file_index]=$element
file_index=`expr $file_index + 1`
done
}
getdir "./output"
echo "Get all the mindrecord files: "${file_list[*]}
# create dataset for train
python create_dataset.py --input_file ${file_list[*]}

18
model_zoo/utils/nlp_to_mindrecord/zhwiki/run_read_simple.sh
Unescape View File

@ -1,18 +0,0 @@
#!/bin/bash
# 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.
# ============================================================================
# create dataset for train
python create_dataset.py --input_file=output/simple.mindrecord

47
model_zoo/utils/nlp_to_mindrecord/zhwiki/run_simple.sh
Unescape View File

@ -1,47 +0,0 @@
#!/bin/bash
# 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.
# ============================================================================
rm -f output/simple.mindrecord*
if [ ! -d "../../../../third_party/to_mindrecord/zhwiki" ]; then
echo "The patch base dir ../../../../third_party/to_mindrecord/zhwiki is not exist."
exit 1
fi
if [ ! -f "../../../../third_party/patch/to_mindrecord/zhwiki/create_pretraining_data.patch" ]; then
echo "The patch file ../../../../third_party/patch/to_mindrecord/zhwiki/create_pretraining_data.patch is not exist."
exit 1
fi
# patch for create_pretraining_data.py
patch -p0 -d ../../../../third_party/to_mindrecord/zhwiki/ -o create_pretraining_data_patched.py < ../../../../third_party/patch/to_mindrecord/zhwiki/create_pretraining_data.patch
if [ $? -ne 0 ]; then
echo "Patch ../../../../third_party/to_mindrecord/zhwiki/create_pretraining_data.py failed"
exit 1
fi
# using patched script to generate mindrecord
python ../../../../third_party/to_mindrecord/zhwiki/create_pretraining_data_patched.py \
--input_file=../../../../third_party/to_mindrecord/zhwiki/sample_text.txt \
--output_file=output/simple.mindrecord \
--partition_number=1 \
--vocab_file=../../../../third_party/to_mindrecord/zhwiki/vocab.txt \
--do_lower_case=True \
--max_seq_length=128 \
--max_predictions_per_seq=20 \
--masked_lm_prob=0.15 \
--random_seed=12345 \
--dupe_factor=10 # user defined

1
third_party/patch/to_mindrecord/CLUERNER2020/README.md vendored
Unescape View File

@ -1 +0,0 @@
## the file is a patch which is about just change data_processor_seq.py the part of generated tfrecord to MindRecord in [CLUEbenchmark/CLUENER2020](https://github.com/CLUEbenchmark/CLUENER2020/tree/master/tf_version)

105
third_party/patch/to_mindrecord/CLUERNER2020/data_processor_seq.patch vendored
Unescape View File

@ -1,105 +0,0 @@
--- data_processor_seq.py 2020-05-28 10:07:13.365947168 +0800
+++ data_processor_seq.py 2020-05-28 10:14:33.298177130 +0800
@@ -4,11 +4,18 @@
@author: Cong Yu
@time: 2019-12-07 17:03
"""
+import sys
+sys.path.append("../../../third_party/to_mindrecord/CLUERNER2020")
+
+import argparse
import json
import tokenization
import collections
-import tensorflow as tf
+import numpy as np
+from mindspore.mindrecord import FileWriter
+
+# pylint: skip-file
def _truncate_seq_pair(tokens_a, tokens_b, max_length):
"""Truncates a sequence pair in place to the maximum length."""
@@ -80,11 +87,18 @@ def process_one_example(tokenizer, label
return feature
-def prepare_tf_record_data(tokenizer, max_seq_len, label2id, path, out_path):
+def prepare_mindrecord_data(tokenizer, max_seq_len, label2id, path, out_path):
"""
- 生成训练数据， tf.record, 单标签分类模型, 随机打乱数据
+ 生成训练数据， *.mindrecord, 单标签分类模型, 随机打乱数据
"""
- writer = tf.python_io.TFRecordWriter(out_path)
+ writer = FileWriter(out_path)
+
+ data_schema = {"input_ids": {"type": "int64", "shape": [-1]},
+ "input_mask": {"type": "int64", "shape": [-1]},
+ "segment_ids": {"type": "int64", "shape": [-1]},
+ "label_ids": {"type": "int64", "shape": [-1]}}
+ writer.add_schema(data_schema, "CLUENER2020 schema")
+
example_count = 0
for line in open(path):
@@ -113,16 +127,12 @@ def prepare_tf_record_data(tokenizer, ma
feature = process_one_example(tokenizer, label2id, list(_["text"]), labels,
max_seq_len=max_seq_len)
- def create_int_feature(values):
- f = tf.train.Feature(int64_list=tf.train.Int64List(value=list(values)))
- return f
-
features = collections.OrderedDict()
# 序列标注任务
- features["input_ids"] = create_int_feature(feature[0])
- features["input_mask"] = create_int_feature(feature[1])
- features["segment_ids"] = create_int_feature(feature[2])
- features["label_ids"] = create_int_feature(feature[3])
+ features["input_ids"] = np.asarray(feature[0])
+ features["input_mask"] = np.asarray(feature[1])
+ features["segment_ids"] = np.asarray(feature[2])
+ features["label_ids"] = np.asarray(feature[3])
if example_count < 5:
print("*** Example ***")
print(_["text"])
@@ -132,8 +142,7 @@ def prepare_tf_record_data(tokenizer, ma
print("segment_ids: %s" % " ".join([str(x) for x in feature[2]]))
print("label: %s " % " ".join([str(x) for x in feature[3]]))
- tf_example = tf.train.Example(features=tf.train.Features(feature=features))
- writer.write(tf_example.SerializeToString())
+ writer.write_raw_data([features])
example_count += 1
# if example_count == 20:
@@ -141,17 +150,22 @@ def prepare_tf_record_data(tokenizer, ma
if example_count % 3000 == 0:
print(example_count)
print("total example:", example_count)
- writer.close()
+ writer.commit()
if __name__ == "__main__":
- vocab_file = "./vocab.txt"
+ parser = argparse.ArgumentParser()
+ parser.add_argument("--vocab_file", type=str, required=True, help='The vocabulary file.')
+ parser.add_argument("--label2id_file", type=str, required=True, help='The label2id.json file.')
+ args = parser.parse_args()
+
+ vocab_file = args.vocab_file
tokenizer = tokenization.FullTokenizer(vocab_file=vocab_file)
- label2id = json.loads(open("label2id.json").read())
+ label2id = json.loads(open(args.label2id_file).read())
max_seq_len = 64
- prepare_tf_record_data(tokenizer, max_seq_len, label2id, path="data/thuctc_train.json",
- out_path="data/train.tf_record")
- prepare_tf_record_data(tokenizer, max_seq_len, label2id, path="data/thuctc_valid.json",
- out_path="data/dev.tf_record")
+ prepare_mindrecord_data(tokenizer, max_seq_len, label2id, path="data/cluener_public/train.json",
+ out_path="output/train.mindrecord")
+ prepare_mindrecord_data(tokenizer, max_seq_len, label2id, path="data/cluener_public/dev.json",
+ out_path="output/dev.mindrecord")

1
third_party/patch/to_mindrecord/zhwiki/README.md vendored
Unescape View File

@ -1 +0,0 @@
## the file is a patch which is about just change create_pretraining_data.py the part of generated tfrecord to MindRecord in [google-research/bert](https://github.com/google-research/bert)

288
third_party/patch/to_mindrecord/zhwiki/create_pretraining_data.patch vendored
View File

File diff suppressed because it is too large Load Diff

1
third_party/to_mindrecord/CLUERNER2020/.gitignore vendored
Unescape View File

@ -1 +0,0 @@
data_processor_seq_patched.py

Some files were not shown because too many files have changed in this diff Show More

Write Preview
Loading…
Cancel
Save