Bump spaCy version, bugfixes, codestyle

This commit is contained in:
Patrick Jentsch 2022-01-27 16:50:22 +01:00
parent 29ccfac4f6
commit a2e8e72e54
9 changed files with 699 additions and 509 deletions

View File

@ -9,7 +9,14 @@ ENV LANG=C.UTF-8
RUN apt-get update \
&& apt-get install --no-install-recommends --yes \
wget
procps \
python3.7 \
python3-pip \
wget \
&& python3 -m pip install \
chardet \
setuptools \
wheel
# Install the NLP pipeline and it's dependencies #
## Install pyFlow ##
@ -21,12 +28,12 @@ RUN wget --no-check-certificate --quiet \
&& apt-get install --no-install-recommends --yes \
python2.7 \
&& python2.7 setup.py build install \
&& cd .. \
&& cd - > /dev/null \
&& rm -r "pyflow-${PYFLOW_VERSION}" "pyflow-${PYFLOW_VERSION}.tar.gz"
## Install spaCy ##
ENV SPACY_VERSION=3.0.5
ENV SPACY_VERSION=3.2.1
RUN apt-get install --no-install-recommends --yes \
python3.7 \
python3-pip \
@ -38,23 +45,14 @@ RUN apt-get install --no-install-recommends --yes \
&& pip3 install "spacy==${SPACY_VERSION}"
# Only models that include the following components are compatibel:
# lemmatizer, ner, parser, senter, tagger,
ENV SPACY_MODELS="de_core_news_md,en_core_web_md,it_core_news_md,nl_core_news_md,pl_core_news_md,zh_core_web_md"
ENV SPACY_MODELS_VERSION=3.0.0
ENV SPACY_MODELS="de_core_news_md,en_core_web_md,it_core_news_md,pl_core_news_md,zh_core_web_md"
ENV SPACY_MODELS_VERSION=3.2.0
RUN for spacy_model in $(echo ${SPACY_MODELS} | tr "," "\n"); do python3 -m spacy download "${spacy_model}-${SPACY_MODELS_VERSION}" --direct; done
## Further dependencies ##
RUN apt-get install --no-install-recommends --yes \
procps \
zip
COPY packages .
RUN cd stand-off-data-py \
&& python3 setup.py build \
&& python3 setup.py install \
&& python3 -m pip install . \
&& cd -

21
LICENSE Normal file
View File

@ -0,0 +1,21 @@
MIT License
Copyright (c) 2021 Bielefeld University - CRC 1288 - INF
Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in all
copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
SOFTWARE.

View File

@ -1,48 +1,41 @@
# NLP - Natural Language Processing
This software implements a heavily parallelized pipeline for Natural Language Processing of text files. It is used for nopaque's NLP service but you can also use it standalone, for that purpose a convenient wrapper script is provided.
This software implements a heavily parallelized pipeline for Natural Language Processing of text files. It is used for nopaque's NLP service but you can also use it standalone, for that purpose a convenient wrapper script is provided. The pipeline is designed to run on Linux operating systems, but with some tweaks it should also run on Windows with WSL installed.
## Software used in this pipeline implementation
- Official Debian Docker image (buster-slim) and programs from its free repositories: https://hub.docker.com/_/debian
- Official Debian Docker image (buster-slim): https://hub.docker.com/_/debian
- Software from Debian Buster's free repositories
- pyFlow (1.1.20): https://github.com/Illumina/pyflow/releases/tag/v1.1.20
- spaCy (3.0.5): https://github.com/tesseract-ocr/tesseract/releases/tag/4.1.1
- spaCy medium sized models (3.0.0):
- https://github.com/explosion/spacy-models/releases/tag/de_core_news_md-3.0.0
- https://github.com/explosion/spacy-models/releases/tag/en_core_web_md-3.0.0
- https://github.com/explosion/spacy-models/releases/tag/it_core_news_md-3.0.0
- https://github.com/explosion/spacy-models/releases/tag/nl_core_news_md-3.0.0
- https://github.com/explosion/spacy-models/releases/tag/pl_core_news_md-3.0.0
- https://github.com/explosion/spacy-models/releases/tag/zh_core_web_md-3.0.0
- spaCy (3.2.1): https://github.com/explosion/spaCy/releases/tag/v3.2.1
- spaCy medium sized models (3.2.0):
- https://github.com/explosion/spacy-models/releases/tag/de_core_news_md-3.2.0
- https://github.com/explosion/spacy-models/releases/tag/en_core_web_md-3.2.0
- https://github.com/explosion/spacy-models/releases/tag/it_core_news_md-3.2.0
- https://github.com/explosion/spacy-models/releases/tag/nl_core_news_md-3.2.0
- https://github.com/explosion/spacy-models/releases/tag/pl_core_news_md-3.2.0
- https://github.com/explosion/spacy-models/releases/tag/zh_core_web_md-3.2.0
## Use this image
## Installation
1. Create input and output directories for the pipeline.
``` bash
mkdir -p /<my_data_location>/input /<my_data_location>/output
```
1. Install Docker and Python 3.
2. Clone this repository: `git clone https://gitlab.ub.uni-bielefeld.de/sfb1288inf/nlp.git`
3. Build the Docker image: `docker build -t gitlab.ub.uni-bielefeld.de:4567/sfb1288inf/nlp:v0.1.0 nlp`
4. Add the wrapper script (`wrapper/nlp` relative to this README file) to your `${PATH}`.
5. Create working directories for the pipeline: `mkdir -p /<my_data_location>/{input,output}`.
2. Place your text files inside `/<my_data_location>/input`. Files should all contain text of the same language.
## Use the Pipeline
1. Place your plain text files inside `/<my_data_location>/input`. Files should all contain text of the same language.
2. Clear your `/<my_data_location>/output` directory.
3. Start the pipeline process. Check the pipeline help (`nlp --help`) for more details.
```
# Option one: Use the wrapper script
## Install the wrapper script (only on first run). Get it from https://gitlab.ub.uni-bielefeld.de/sfb1288inf/nlp/-/raw/1.0.0/wrapper/nlp, make it executeable and add it to your ${PATH}
```bash
cd /<my_data_location>
nlp -i input -l <language_code> -o output <optional_pipeline_arguments>
# Option two: Classic Docker style
docker run \
--rm \
-it \
-u $(id -u $USER):$(id -g $USER) \
-v /<my_data_location>/input:/input \
-v /<my_data_location>/output:/output \
gitlab.ub.uni-bielefeld.de:4567/sfb1288inf/nlp:1.0.0 \
-i /input \
-l <language_code>
-o /output \
<optional_pipeline_arguments>
nlp \
--input-dir input \
--output-dir output \
-m <model_code> <optional_pipeline_arguments>
```
4. Check your results in the `/<my_data_location>/output` directory.

339
nlp
View File

@ -1,73 +1,141 @@
#!/usr/bin/env python2.7
# coding=utf-8
"""A NLP pipeline for text file processing."""
__author__ = 'Patrick Jentsch <p.jentsch@uni-bielefeld.de>,' \
'Stephan Porada <porada@posteo.de>'
__version__ = '1.0.0'
''' A NLP pipeline for text file processing. '''
__version__ = '0.1.0'
from argparse import ArgumentParser
from pyflow import WorkflowRunner
import multiprocessing
import json
import os
import sys
SPACY_MODELS = {'de': 'de_core_news_md',
SPACY_MODELS = {
'de': 'de_core_news_md',
'en': 'en_core_web_md',
'it': 'it_core_news_md',
'nl': 'nl_core_news_md',
'pl': 'pl_core_news_md',
'zh': 'zh_core_web_md'}
'zh': 'zh_core_web_md'
}
class NLPPipelineJob:
"""An NLP pipeline job class
class PipelineJob:
'''
NLP pipeline job class.
Each input file of the pipeline is represented as an NLP pipeline job,
which holds all necessary information for the pipeline to process it.
Each plain text input file of the pipeline is represented as an NLP
pipeline job, which holds all necessary information for the pipeline to
process it.
Arguments:
file -- Path to the file
output_dir -- Path to a directory, where job results a stored
"""
output_dir -- Path to a directory, where job results are stored
'''
def __init__(self, file, output_dir):
self.file = file
self.name = os.path.basename(file).rsplit('.', 1)[0]
self.name = os.path.basename(file)[:-4]
self.output_dir = output_dir
catma_stand_off_data_file = file.rsplit('.', 1)[0] + '.catma-stand-off.json' # noqa
if os.path.exists(catma_stand_off_data_file):
self.catma_stand_off_data_file = catma_stand_off_data_file
else:
self.catma_stand_off_data_file = None
class NLPPipeline(WorkflowRunner):
def __init__(self, input_dir, output_dir, check_encoding, lang, zip):
class NLPWorkflow(WorkflowRunner):
def __init__(self, job, model, check_encoding=False, id_prefix=''):
self.job = job
self.model = model
self.check_encoding = check_encoding
self.id_prefix = id_prefix
def workflow(self):
'''
' ##################################################
' # spacy #
' ##################################################
'''
n_cores = 1
mem_mb = min(1024, self.getMemMb())
cmd = 'spacy-nlp'
cmd += ' --input-file "{}"'.format(self.job.file)
cmd += ' --output-file "{}"'.format(
os.path.join(self.job.output_dir, '{}.json'.format(self.job.name))
)
cmd += ' -m "{}"'.format(self.model)
if self.check_encoding:
cmd += ' --check-encoding'
cmd += ' --id-prefix "{}"'.format(self.id_prefix)
self.addTask(
'spacy',
command=cmd,
memMb=mem_mb,
nCores=n_cores
)
class CreateVrtWorkflow(WorkflowRunner):
def __init__(self, job):
self.job = job
def workflow(self):
'''
' ##################################################
' # vrt-creator #
' ##################################################
'''
n_cores = 1
mem_mb = min(256, self.getMemMb())
cmd = 'vrt-creator'
cmd += ' --stand-off-data-file "{}"'.format(
os.path.join(self.job.output_dir, '{}.json'.format(self.job.name))
)
cmd += ' --text-file "{}"'.format(self.job.file)
cmd += ' --output-file "{}"'.format(
os.path.join(self.job.output_dir, '{}.vrt'.format(self.job.name))
)
self.addTask(
'vrt_creator',
command=cmd,
memMb=mem_mb,
nCores=n_cores
)
class MainWorkflow(WorkflowRunner):
def __init__(
self,
input_dir,
model,
output_dir,
check_encoding=False,
id_prefix=''
):
self.input_dir = input_dir
self.model = model
self.output_dir = output_dir
self.check_encoding = check_encoding
self.lang = lang
self.zip = zip
self.jobs = collect_jobs(self.input_dir, self.output_dir)
self.id_prefix = id_prefix
self.jobs = []
def collect_jobs(self):
self.jobs = []
for file in os.listdir(self.input_dir):
if os.path.isdir(os.path.join(self.input_dir, file)):
continue
if not file.lower().endswith('.txt'):
continue
job = PipelineJob(
os.path.join(self.input_dir, file),
os.path.join(self.output_dir, file)
)
self.jobs.append(job)
def workflow(self):
if not self.jobs:
return
'''
' ##################################################
' # setup output directory #
' ##################################################
'''
setup_output_directory_tasks = []
for i, job in enumerate(self.jobs):
cmd = 'mkdir -p "{}"'.format(job.output_dir)
lbl = 'setup_output_directory_-_{}'.format(i)
task = self.addTask(command=cmd, label=lbl)
setup_output_directory_tasks.append(task)
# Create output and temporary directories
for job in self.jobs:
os.mkdir(job.output_dir)
'''
' ##################################################
@ -75,106 +143,116 @@ class NLPPipeline(WorkflowRunner):
' ##################################################
'''
nlp_tasks = []
n_cores = max(1, int(self.getNCores() / len(self.jobs)))
mem_mb = min(n_cores * 2048, int(self.getMemMb() / len(self.jobs)))
for i, job in enumerate(self.jobs):
output_file = os.path.join(job.output_dir, '{}.nopaque-stand-off.json'.format(job.name)) # noqa
cmd = 'spacy-nlp'
cmd += ' -l "{}"'.format(self.lang)
cmd += ' --check-encoding' if self.check_encoding else ''
cmd += ' "{}"'.format(job.file)
cmd += ' "{}"'.format(output_file)
deps = 'setup_output_directory_-_{}'.format(i)
lbl = 'nlp_-_{}'.format(i)
task = self.addTask(command=cmd, dependencies=deps, label=lbl,
memMb=mem_mb, nCores=n_cores)
task = self.addWorkflowTask(
'nlp_-_{}'.format(i),
NLPWorkflow(
job,
self.model,
check_encoding=self.check_encoding,
id_prefix=self.id_prefix
)
)
nlp_tasks.append(task)
'''
' ##################################################
' # vrt creation #
' # create vrt #
' ##################################################
'''
vrt_creation_tasks = []
create_vrt_tasks = []
for i, job in enumerate(self.jobs):
output_file = os.path.join(job.output_dir, '{}.vrt'.format(job.name)) # noqa
nopaque_stand_off_data_file = os.path.join(job.output_dir, '{}.nopaque-stand-off.json'.format(job.name)) # noqa
cmd = 'vrt-creator'
cmd += ' "{}"'.format(job.file)
cmd += ' "{}"'.format(nopaque_stand_off_data_file)
if job.catma_stand_off_data_file is not None:
cmd += ' --catma-stand-off-data "{}"'.format(job.catma_stand_off_data_file) # noqa
cmd += ' "{}"'.format(output_file)
deps = 'nlp_-_{}'.format(i)
lbl = 'vrt_creation_-_{}'.format(i)
task = self.addTask(command=cmd, dependencies=deps, label=lbl)
vrt_creation_tasks.append(task)
task = self.addWorkflowTask(
'create_vrt_-_{}'.format(i),
CreateVrtWorkflow(job),
dependencies='nlp_-_{}'.format(i)
)
create_vrt_tasks.append(task)
'''
' ##################################################
' # zip creation #
' ##################################################
'''
zip_creation_tasks = []
if self.zip is not None:
cmd = 'cd "{}"'.format(self.output_dir)
cmd += ' && '
cmd += 'zip'
cmd += ' -r'
cmd += ' "{}.zip" .'.format(self.zip)
cmd += ' -x "pyflow.data*"'
cmd += ' -i "*.vrt" "*.json"'
cmd += ' && '
cmd += 'cd -'
deps = vrt_creation_tasks
lbl = 'zip_creation'
task = self.addTask(command=cmd, dependencies=deps, label=lbl)
zip_creation_tasks.append(task)
def collect_jobs(input_dir, output_dir):
jobs = []
for file in os.listdir(input_dir):
if os.path.isdir(os.path.join(input_dir, file)):
continue
if file.lower().endswith('.txt'):
job = NLPPipelineJob(os.path.join(input_dir, file),
os.path.join(output_dir, file))
jobs.append(job)
return jobs
self.waitForTasks()
outputs = []
for job in self.jobs:
# Track output files
relative_output_dir = os.path.relpath(
job.output_dir,
start=self.output_dir
)
outputs.append(
{
'description': 'JSON stand off data',
'file': os.path.join(
relative_output_dir,
'{}.json'.format(job.name)
),
'mimetype': 'application/json'
}
)
outputs.append(
{
'description': 'CWB vrt file',
'file': os.path.join(
relative_output_dir,
'{}.vrt'.format(job.name)
),
'mimetype': 'application/vrt+xml'
}
)
with open(os.path.join(self.output_dir, 'outputs.json'), 'w') as f:
json.dump(outputs, f, indent=4)
def parse_args():
parser = ArgumentParser(description='NLP pipeline for TXT file processing',
prog='NLP pipeline')
parser.add_argument('-i', '--input-dir',
parser = ArgumentParser(
description='NLP pipeline for plain text file processing'
)
parser.add_argument(
'-i', '--input-dir',
help='Input directory',
required=True)
parser.add_argument('-o', '--output-dir',
required=True
)
parser.add_argument(
'-o', '--output-dir',
help='Output directory',
required=True)
parser.add_argument('-l', '--language',
required=True
)
parser.add_argument(
'-m', '--model',
choices=SPACY_MODELS.keys(),
help='Language of the input (2-character ISO 639-1 language codes)', # noqa
required=True)
parser.add_argument('--check-encoding',
help='The model to be used',
required=True
)
parser.add_argument(
'--check-encoding',
action='store_true',
help='Check encoding of the input file, UTF-8 is used instead') # noqa
parser.add_argument('--log-dir',
help='Logging directory')
parser.add_argument('--mem-mb',
help='Amount of system memory to be used (Default: min(--n-cores * 2048, available system memory))', # noqa
type=int)
parser.add_argument('--n-cores',
default=min(4, multiprocessing.cpu_count()),
help='Number of CPU threads to be used (Default: min(4, number of CPUs))', # noqa
type=int)
parser.add_argument('--zip',
help='Create one zip file per filetype')
parser.add_argument('-v', '--version',
help='Check encoding of the input file, UTF-8 is used instead'
)
parser.add_argument(
'--id-prefix',
default='',
help='A prefix for all the ids within the stand off annotations'
)
parser.add_argument(
'--log-dir',
help='Logging directory (Default: --output-dir)'
)
parser.add_argument(
'--mem-mb',
help='Amount of system memory to be used '
'(Default: min(--n-cores * 1024, available system memory))',
type=int
)
parser.add_argument(
'--n-cores',
default=1,
help='Number of CPU threads to be used',
type=int
)
parser.add_argument(
'-v', '--version',
action='version',
help='Returns the current version of the NLP pipeline',
version='%(prog)s {}'.format(__version__))
version='%(prog)s {}'.format(__version__)
)
args = parser.parse_args()
# Set some tricky default values and check for insufficient input
@ -184,20 +262,27 @@ def parse_args():
raise Exception('--n-cores must be greater or equal 1')
if args.mem_mb is None:
max_mem_mb = int(os.popen('free -t -m').readlines()[-1].split()[1:][0])
args.mem_mb = min(args.n_cores * 2048, max_mem_mb)
if args.mem_mb < 2048:
raise Exception('--mem-mb must be greater or equal 2048')
if args.zip is not None and args.zip.lower().endswith('.zip'):
# Remove .zip file extension if provided
args.zip = args.zip[:-4]
args.zip = args.zip if args.zip else 'output'
args.mem_mb = min(args.n_cores * 1024, max_mem_mb)
if args.mem_mb < 1024:
raise Exception('--mem-mb must be greater or equal 1024')
return args
def main():
args = parse_args()
nlp_pipeline = NLPPipeline(args.input_dir, args.output_dir, args.check_encoding, args.language, args.zip) # noqa
retval = nlp_pipeline.run(dataDirRoot=args.log_dir, memMb=args.mem_mb, nCores=args.n_cores) # noqa
main_workflow = MainWorkflow(
args.input_dir,
args.model,
args.output_dir,
check_encoding=args.check_encoding,
id_prefix=args.id_prefix
)
main_workflow.collect_jobs()
retval = main_workflow.run(
dataDirRoot=args.log_dir,
memMb=args.mem_mb,
nCores=args.n_cores
)
sys.exit(retval)

View File

@ -1,14 +1,14 @@
import setuptools
setuptools.setup(
name='stand-off-data',
name='Stand off data',
author='Patrick Jentsch',
author_email='p.jentsch@uni-bielefeld.de',
description='A python library to handle stand off data.',
py_modules=['stand_off_data'],
classifiers=[
'Programming Language :: Python :: 3',
'Operating System :: OS Independent',
],
packages=setuptools.find_packages(),
python_requires='>=3.5'
)

View File

@ -7,13 +7,15 @@ class StandOffData:
self.lookup = {}
for x in attrs.get('tags', []):
self.add_tag_definition(x)
self.annotations = [TagAnnotation(x, self.lookup)
for x in attrs.get('annotations', [])]
self.annotations = [
TagAnnotation(x, self.lookup)
for x in attrs.get('annotations', [])
]
def add_tag_definition(self, attrs):
tag_definition = TagDefinition(attrs)
if tag_definition.id in self.lookup:
raise Exception('Tag id already in use: {}'.format(self.to_dict()))
raise Exception(f'Tag id already in use: {self.to_dict()}')
self.lookup[tag_definition.id] = tag_definition
def to_dict(self):
@ -42,7 +44,9 @@ class StandOffData:
if ((p_attr.start >= next_p_attr.start) and (p_attr.start < next_p_attr.end) # noqa
or (p_attr.end > next_p_attr.start) and (p_attr.end <= next_p_attr.end)): # noqa
raise Exception(
'Positional attribute overlaps another: {}<->{}'.format(p_attr.to_dict(), next_p_attr.to_dict()))
'Positional attribute overlaps another: '
f'{p_attr.to_dict()}<->{next_p_attr.to_dict()}'
)
# Check for s_attr<->p_attr overlap
for i, s_attr in enumerate(s_attrs):
for p_attr in p_attrs:
@ -56,8 +60,11 @@ class StandOffData:
s_attrs[i].end = p_attr.end
# Check if s_attr starts/ends before/after p_attr
if p_attr.start >= s_attr.end or p_attr.end <= s_attr.start:
# No further Checking needed (just because p_attrs are sorted)
# No further Checking needed (because p_attrs are sorted)
break
p_attr_buffer = {}
for i, p_attr in enumerate(p_attrs):
p_attr_buffer[p_attr.start] = i
s_attr_start_buffer = {}
s_attr_end_buffer = {}
for i, s_attr in enumerate(s_attrs):
@ -66,34 +73,56 @@ class StandOffData:
else:
s_attr_start_buffer[s_attr.start] = [i]
if s_attr.end in s_attr_end_buffer:
s_attr_end_buffer[s_attr.end].append(i)
s_attr_end_buffer[s_attr.end].insert(0, i)
else:
s_attr_end_buffer[s_attr.end] = [i]
vrt = ''
vrt += '<text>\n'
for p_attr in p_attrs:
# s_attr_ends
for k in {k: v for k, v in s_attr_end_buffer.items() if k <= p_attr.start}: # noqa
s_attr_indexes = s_attr_end_buffer.pop(k)
current_position = 0
text_len = len(text)
# As long as we have something in our buffers we process it
while current_position <= text_len:
# s_attr endings
# for k in {k: v for k, v in s_attr_end_buffer.items() if k <= current_position}: # noqa
if current_position in s_attr_end_buffer:
# s_attr_indexes = s_attr_end_buffer.pop(k)
s_attr_indexes = s_attr_end_buffer.pop(current_position)
for s_attr_index in s_attr_indexes:
s_attr = s_attrs[s_attr_index]
vrt += '</{}>\n'.format(escape(s_attr.name))
# s_attr_starts
for k in {k: v for k, v in s_attr_start_buffer.items() if k <= p_attr.start}: # noqa
s_attr_indexes = s_attr_start_buffer.pop(k)
vrt += f'</{escape(s_attr.name)}>\n'
# s_attrs starts
# for k in {k: v for k, v in s_attr_start_buffer.items() if k <= current_position}: # noqa
if current_position in s_attr_start_buffer:
# s_attr_indexes = s_attr_start_buffer.pop(k)
s_attr_indexes = s_attr_start_buffer.pop(current_position)
for s_attr_index in s_attr_indexes:
s_attr = s_attrs[s_attr_index]
foo = ''
vrt += f'<{escape(s_attr.name)}'
for property in s_attr.properties:
foo += ' {}="{}"'.format(escape(property.name),
escape(property.value))
vrt += '<{}{}>\n'.format(escape(s_attr.name), foo)
foo = {'lemma': None, 'ner': None, 'pos': None, 'simple_pos': None, 'word': None} # noqa
vrt += f' {escape(property.name)}="{escape(str(property.value))}"' # noqa
vrt += '>\n'
# p_attrs
if current_position not in p_attr_buffer:
current_position += 1
continue
p_attr_index = p_attr_buffer.pop(current_position)
p_attr = p_attrs[p_attr_index]
if text[p_attr.start:p_attr.end].isspace():
current_position = p_attr.end
continue
_p_attr = {
'lemma': 'None',
'pos': 'None',
'simple_pos': 'None',
'word': 'None'
}
for property in p_attr.properties:
foo[property.name] = escape(property.value)
foo['word'] = escape(text[p_attr.start:p_attr.end])
vrt += '{word}\t{pos}\t{lemma}\t{simple_pos}\t{ner}\n'.format(
**foo)
if property.name not in _p_attr:
continue
_p_attr[property.name] = escape(str(property.value))
_p_attr['word'] = escape(text[p_attr.start:p_attr.end])
vrt += '{word}\t{pos}\t{lemma}\t{simple_pos}\n'.format(**_p_attr)
current_position = p_attr.end
vrt += '</text>\n'
return vrt
@ -110,15 +139,15 @@ class TagAnnotation:
]
''' Sanity checks '''
if self.tag_id not in self.lookup:
raise Exception('Unknown tag: {}'.format(self.to_dict()))
raise Exception(f'Unknown tag: {self.to_dict()}')
if self.end < self.start:
raise Exception('Annotation end less then start: '
'{}'.format(self.to_dict()))
property_ids = [x.property_id for x in self.properties]
for required_property_id, required_property in self.lookup[self.tag_id].required_properties.items(): # noqa
if required_property_id not in property_ids:
raise Exception('Missing required property: '
'{}'.format(required_property.to_dict()))
raise Exception(f'Annotation end less then start: {self.to_dict()}') # noqa
# property_ids = [x.property_id for x in self.properties]
# for required_property_id, required_property in self.lookup[self.tag_id].required_properties.items(): # noqa
# if required_property_id not in property_ids:
# raise Exception(
# f'Missing required property: {required_property.to_dict()}'
# )
@property
def name(self):
@ -134,33 +163,45 @@ class TagAnnotation:
def __lt__(self, other):
if self.start == other.start:
return self.name == 'token' and other.name != 'token'
if self.name == 'token' and other.name != 'token':
return False
elif self.name != 'token' and other.name == 'token':
return True
else:
return self.end > other.end
else:
return self.start < other.start
def __le__(self, other):
if self.start == other.start:
return self.name == 'token' or other.name != 'token'
if self.name == 'token' and other.name != 'token':
return False
elif self.name != 'token' and other.name == 'token':
return True
else:
return self.start < other.start
return self.end >= other.end
else:
return self.start <= other.start
def __eq__(self, other):
return self.start == other.start and self.name == other.name
if self.start == other.start:
if self.name == 'token' and other.name != 'token':
return False
elif self.name != 'token' and other.name == 'token':
return False
else:
return self.end == other.end
else:
return False
def __ne__(self, other):
return self.start != other.start and self.name != other.name
return not self == other
def __gt__(self, other):
if self.start == other.start:
return self.name != 'token' and other.name == 'token'
else:
return self.start > other.start
return not self <= other
def __ge__(self, other):
if self.start == other.start:
return self.name != 'token' or other.name == 'token'
else:
return self.start > other.start
return not self < other
class PropertyAnnotation:
@ -171,7 +212,7 @@ class PropertyAnnotation:
# TODO: Process attrs['possibleValues'] as self.labels (no id?)
''' Sanity checks '''
if self.property_id not in self.lookup:
raise Exception('Unknown property: {}'.format(self.to_dict()))
raise Exception(f'Unknown property: {self.to_dict()}')
@property
def name(self):
@ -197,14 +238,14 @@ class TagDefinition:
def add_property_definition(self, attrs):
property_definition = PropertyDefinition(attrs)
if property_definition.id in self.properties:
raise Exception('Property id already in use: '
'{}'.format(property_definition.to_dict()))
raise Exception(
f'Property id already in use: {property_definition.to_dict()}')
self.properties[property_definition.id] = property_definition
@property
def required_properties(self):
return {property.id: property for property in self.properties.values()
if property.is_required}
# @property
# def required_properties(self):
# return {property.id: property for property in self.properties.values()
# if property.is_required}
def to_dict(self):
return {
@ -223,9 +264,9 @@ class PropertyDefinition:
self.flags = attrs.get('flags', [])
self.labels = attrs.get('labels', [])
@property
def is_required(self):
return 'required' in self.flags
# @property
# def is_required(self):
# return 'required' in self.flags
@property
def has_multiple_values(self):

299
spacy-nlp
View File

@ -11,46 +11,64 @@ import textwrap
import uuid
def UUIDnopaque(name):
return 'nopaque_{}'.format(
uuid.uuid3(uuid.NAMESPACE_DNS,
'{}@nopaque.sfb1288.uni-bielefeld.de'.format(name))
)
spacy_models = {spacy.info(pipeline)['lang']: pipeline
for pipeline in spacy.info()['pipelines']}
spacy_models = {
spacy.info(pipeline)['lang']: pipeline
for pipeline in spacy.info()['pipelines']
}
# Parse the given arguments
parser = ArgumentParser(description='Create annotations for a given txt file')
parser.add_argument('input', help='Path to txt input file')
parser.add_argument('output', help='Path to JSON output file')
parser.add_argument('-l', '--language',
parser = ArgumentParser(
description='Create annotations for a given plain txt file'
)
parser.add_argument(
'-i', '--input-file',
help='Input file'
)
parser.add_argument(
'-o', '--output-file',
help='Output file',
required=True
)
parser.add_argument(
'-m', '--model',
choices=spacy_models.keys(),
help='Language of the input (2-character ISO 639-1 language codes)', # noqa
required=True)
parser.add_argument('-c', '--check-encoding',
help='The model to be used',
required=True
)
parser.add_argument(
'-c', '--check-encoding',
action='store_true',
help='Check encoding of the input file, UTF-8 is used instead') # noqa
help='Check encoding of the input file, UTF-8 is used instead'
)
parser.add_argument(
'--id-prefix',
default='',
help='A prefix for all the ids within the stand off annotations'
)
args = parser.parse_args()
with open(args.input, "rb") as text_file:
def generate_id(name):
return f'{args.id_prefix}{uuid.uuid3(uuid.NAMESPACE_DNS, name)}'
with open(args.input_file, "rb") as input_file:
if args.check_encoding:
encoding = chardet.detect(text_file.read())['encoding']
encoding = chardet.detect(input_file.read())['encoding']
else:
encoding = 'utf-8'
text_file.seek(0)
input_file.seek(0)
text_md5 = hashlib.md5()
for chunk in iter(lambda: text_file.read(128 * text_md5.block_size), b''):
for chunk in iter(lambda: input_file.read(128 * text_md5.block_size), b''):
text_md5.update(chunk)
# Load the text contents from the input file
with open(args.input, encoding=encoding) as text_file:
with open(args.input_file, encoding=encoding) as input_file:
# spaCy NLP is limited to strings with a maximum of 1 million characters at
# once. So we split it into suitable chunks.
text_chunks = textwrap.wrap(
text_file.read(),
input_file.read(),
1000000,
break_long_words=False,
break_on_hyphens=False,
@ -59,186 +77,197 @@ with open(args.input, encoding=encoding) as text_file:
replace_whitespace=False
)
model = spacy_models[args.language]
nlp = spacy.load(model)
model_name = spacy_models[args.model]
nlp = spacy.load(model_name)
meta = {
'generator': {
'name': 'nopaque NLP service',
'version': '1.0.0',
'name': 'nopaque spacy NLP',
'version': '0.1.0',
'arguments': {
'check_encoding': args.check_encoding,
'language': args.language
'model': args.model
}
},
'file': {
'encoding': encoding,
'md5': text_md5.hexdigest(),
'name': os.path.basename(args.input)
'name': os.path.basename(args.input_file)
}
}
tags = [
{
'id': UUIDnopaque('token'),
tags = []
token = {
'id': generate_id('token'),
'name': 'token',
'description': 'An individual token — i.e. a word, punctuation symbol, whitespace, etc.',
'properties': [
'description': 'An individual token — i.e. a word, punctuation symbol, whitespace, etc.', # noqa
'properties': []
}
# TODO: Check if all languages support token.sentiment
token['properties'].append(
{
'id': UUIDnopaque('token.lemma'),
'id': generate_id('token.sentiment'),
'name': 'sentiment',
'description': 'A scalar value indicating the positivity or negativity of the token.' # noqa
}
)
if nlp.has_pipe('lemmatizer'):
token['properties'].append(
{
'id': generate_id('token.lemma'),
'name': 'lemma',
'description': 'The base form of the word',
'flags': ['required'],
'labels': []
},
'description': 'The base form of the word'
}
)
if nlp.has_pipe('morphologizer') or nlp.has_pipe('tagger'):
token['properties'].append(
{
'id': UUIDnopaque('token.pos'),
'name': 'pos',
'description': 'The detailed part-of-speech tag',
'flags': ['required'],
'labels': [
{
'id': UUIDnopaque('token.pos={}'.format(label)),
'name': label,
'description': spacy.explain(label) or ''
} for label in spacy.info(model)['labels']['tagger']
]
},
{
'id': UUIDnopaque('token.simple_pos'),
'id': generate_id('token.simple_pos'),
'name': 'simple_pos',
'description': 'The simple UPOS part-of-speech tag',
'flags': ['required'],
'labels': [
{
'id': UUIDnopaque('token.simple_pos=ADJ'),
'id': generate_id('token.simple_pos=ADJ'),
'name': 'ADJ',
'description': 'adjective'
},
{
'id': UUIDnopaque('token.simple_pos=ADJ'),
'id': generate_id('token.simple_pos=ADJ'),
'name': 'ADP',
'description': 'adposition'
},
{
'id': UUIDnopaque('token.simple_pos=ADJ'),
'id': generate_id('token.simple_pos=ADJ'),
'name': 'ADV',
'description': 'adverb'
},
{
'id': UUIDnopaque('token.simple_pos=ADJ'),
'id': generate_id('token.simple_pos=ADJ'),
'name': 'AUX',
'description': 'auxiliary verb'
},
{
'id': UUIDnopaque('token.simple_pos=ADJ'),
'id': generate_id('token.simple_pos=ADJ'),
'name': 'CONJ',
'description': 'coordinating conjunction'
},
{
'id': UUIDnopaque('token.simple_pos=ADJ'),
'id': generate_id('token.simple_pos=ADJ'),
'name': 'DET',
'description': 'determiner'
},
{
'id': UUIDnopaque('token.simple_pos=ADJ'),
'id': generate_id('token.simple_pos=ADJ'),
'name': 'INTJ',
'description': 'interjection'
},
{
'id': UUIDnopaque('token.simple_pos=ADJ'),
'id': generate_id('token.simple_pos=ADJ'),
'name': 'NOUN',
'description': 'noun'
},
{
'id': UUIDnopaque('token.simple_pos=ADJ'),
'id': generate_id('token.simple_pos=ADJ'),
'name': 'NUM',
'description': 'numeral'
},
{
'id': UUIDnopaque('token.simple_pos=ADJ'),
'id': generate_id('token.simple_pos=ADJ'),
'name': 'PART',
'description': 'particle'
},
{
'id': UUIDnopaque('token.simple_pos=ADJ'),
'id': generate_id('token.simple_pos=ADJ'),
'name': 'PRON',
'description': 'pronoun'
},
{
'id': UUIDnopaque('token.simple_pos=ADJ'),
'id': generate_id('token.simple_pos=ADJ'),
'name': 'PROPN',
'description': 'proper noun'
},
{
'id': UUIDnopaque('token.simple_pos=ADJ'),
'id': generate_id('token.simple_pos=ADJ'),
'name': 'PUNCT',
'description': 'punctuation'
},
{
'id': UUIDnopaque('token.simple_pos=ADJ'),
'id': generate_id('token.simple_pos=ADJ'),
'name': 'SCONJ',
'description': 'subordinating conjunction'
},
{
'id': UUIDnopaque('token.simple_pos=ADJ'),
'id': generate_id('token.simple_pos=ADJ'),
'name': 'SYM',
'description': 'symbol'
},
{
'id': UUIDnopaque('token.simple_pos=ADJ'),
'id': generate_id('token.simple_pos=ADJ'),
'name': 'VERB',
'description': 'verb'
},
{
'id': UUIDnopaque('token.simple_pos=ADJ'),
'id': generate_id('token.simple_pos=ADJ'),
'name': 'X',
'description': 'other'
}
]
},
}
)
if nlp.has_pipe('tagger'):
token['properties'].append(
{
'id': UUIDnopaque('token.ner'),
'name': 'ner',
'description': 'Label indicating the type of the entity',
'flags': ['required'],
'id': generate_id('token.pos'),
'name': 'pos',
'description': 'The detailed part-of-speech tag',
'labels': [
{
'id': UUIDnopaque('token.ner={}'.format(label)),
'id': generate_id(f'token.pos={label}'),
'name': label,
'description': spacy.explain(label) or ''
} for label in spacy.info(model)['labels']['ner']
} for label in spacy.info(model_name)['labels']['tagger']
]
}
]
},
)
if nlp.has_pipe('ner') or nlp.has_pipe('entity_ruler'):
tags.append(
{
'id': UUIDnopaque('s'),
'name': 's',
'description': 'Encodes the start and end of a sentence',
'properties': []
},
{
'id': UUIDnopaque('ent'),
'id': generate_id('ent'),
'name': 'ent',
'description': 'Encodes the start and end of a named entity',
'properties': [
{
'id': UUIDnopaque('ent.type'),
'id': generate_id('ent.type'),
'name': 'type',
'description': 'Label indicating the type of the entity',
'flags': ['required'],
'labels': [
{
'id': UUIDnopaque('ent.type={}'.format(label)),
'id': generate_id('ent.type={}'.format(label)),
'name': label,
'description': spacy.explain(label) or ''
} for label in spacy.info(model)['labels']['ner']
} for label in spacy.info(model_name)['labels']['ner']
]
}
]
}
]
)
if nlp.has_pipe('parser') or nlp.has_pipe('senter') or nlp.has_pipe('sentencizer'): # noqa
# TODO: Check if all languages support sent.sentiment
tags.append(
{
'id': generate_id('s'),
'name': 's',
'description': 'Encodes the start and end of a sentence',
'properties': [
{
'id': generate_id('s.sentiment'),
'name': 'sentiment',
'description': 'A scalar value indicating the positivity or negativity of the sentence.' # noqa
}
]
}
)
tags.append(token)
annotations = []
@ -246,60 +275,78 @@ chunk_offset = 0
while text_chunks:
text_chunk = text_chunks.pop(0)
doc = nlp(text_chunk)
for token in doc:
if token.is_space:
continue
if token.is_sent_start:
annotation = {'start': token.sent.start_char + chunk_offset,
'end': token.sent.end_char + chunk_offset,
'tag_id': UUIDnopaque('s'),
'properties': []}
annotations.append(annotation)
# Check if the token is the start of an entity
if token.ent_iob == 3:
for ent_candidate in token.sent.ents:
if ent_candidate.start_char == token.idx:
ent = ent_candidate
if hasattr(doc, 'ents'):
for ent in doc.ents:
annotation = {
'start': ent.start_char + chunk_offset,
'end': ent.end_char + chunk_offset,
'tag_id': UUIDnopaque('ent'),
'tag_id': generate_id('ent'),
'properties': [
{
'property_id': UUIDnopaque('ent.type'),
'value': token.ent_type_
'property_id': generate_id('ent.type'),
'value': ent.label_
}
]
}
annotations.append(annotation)
break
if hasattr(doc, 'sents'):
for sent in doc.sents:
annotation = {
'start': sent.start_char + chunk_offset,
'end': sent.end_char + chunk_offset,
'tag_id': generate_id('s'),
'properties': []
}
if hasattr(sent, 'sentiment'):
annotation['properties'].append(
{
'property_id': generate_id('s.sentiment'),
'value': sent.sentiment
}
)
annotations.append(annotation)
for token in doc:
annotation = {
'start': token.idx + chunk_offset,
'end': token.idx + len(token.text) + chunk_offset,
'tag_id': UUIDnopaque('token'),
'properties': [
'tag_id': generate_id('token'),
'properties': []
}
if hasattr(token, 'lemma_'):
annotation['properties'].append(
{
'property_id': UUIDnopaque('token.pos'),
'value': token.tag_
},
{
'property_id': UUIDnopaque('token.lemma'),
'property_id': generate_id('token.lemma'),
'value': token.lemma_
},
}
)
if hasattr(token, 'pos_'):
annotation['properties'].append(
{
'property_id': UUIDnopaque('token.simple_pos'),
'property_id': generate_id('token.simple_pos'),
'value': token.pos_
},
}
)
if hasattr(token, 'sentiment'):
annotation['properties'].append(
{
'property_id': UUIDnopaque('token.ner'),
'value': token.ent_type_ if token.ent_type_ else 'None'
'property_id': generate_id('token.sentiment'),
'value': token.sentiment
}
]
)
if hasattr(token, 'tag_'):
annotation['properties'].append(
{
'property_id': generate_id('token.pos'),
'value': token.tag_
}
)
annotations.append(annotation)
chunk_offset += len(text_chunk)
text_chunk = None
with open(args.output, 'w') as output_file:
json.dump({'meta': meta, 'tags': tags, 'annotations': annotations},
output_file, indent=4)
with open(args.output_file, 'w') as output_file:
json.dump(
{'meta': meta, 'tags': tags, 'annotations': annotations},
output_file,
indent=4
)

View File

@ -6,31 +6,36 @@ from stand_off_data import StandOffData
import hashlib
import json
parser = ArgumentParser(
description='Convert plain text and JSON stand off to a CWB vrt file'
)
parser.add_argument(
'-s', '--stand-off-data-file',
help='JSON stand off data input file'
)
parser.add_argument(
'-t', '--text-file',
help='Plain text input file'
)
parser.add_argument(
'-o', '--output-file',
help='Output file',
required=True
)
args = parser.parse_args()
def main():
# Parse the given arguments
parser = ArgumentParser(description='Create a vrt from JSON and txt')
parser.add_argument('text', help='Path to txt file')
parser.add_argument('stand_off_data', help='Path to JSON file')
parser.add_argument('output', help='Path to vrt output file')
args = parser.parse_args()
with open(args.stand_off_data) as stand_of_data_file:
with open(args.stand_off_data_file) as stand_of_data_file:
stand_off_data = StandOffData(json.load(stand_of_data_file))
with open(args.text, "rb") as text_file:
with open(args.text_file, "rb") as text_file:
text_md5 = hashlib.md5()
for chunk in iter(lambda: text_file.read(128 * text_md5.block_size), b''): # noqa
for chunk in iter(lambda: text_file.read(128 * text_md5.block_size), b''):
text_md5.update(chunk)
if text_md5.hexdigest() != stand_off_data.meta['file']['md5']:
raise Exception('md5 not equal')
with open(args.text, encoding=stand_off_data.meta['file']['encoding']) as text_file:
with open(args.text_file, encoding=stand_off_data.meta['file']['encoding']) as text_file: # noqa
text = text_file.read()
with open(args.output, 'w') as vrt_file:
with open(args.output_file, 'w') as vrt_file:
vrt_file.write(stand_off_data.to_vrt(text))
if __name__ == '__main__':
main()

View File

@ -6,7 +6,7 @@ import os
import subprocess
import sys
CONTAINER_IMAGE = 'gitlab.ub.uni-bielefeld.de:4567/sfb1288inf/nlp:1.0.0b'
CONTAINER_IMAGE = 'gitlab.ub.uni-bielefeld.de:4567/sfb1288inf/nlp:v0.1.0'
CONTAINER_INPUT_DIR = '/input'
CONTAINER_OUTPUT_DIR = '/output'
CONTAINER_LOG_DIR = '/logs'
@ -19,17 +19,17 @@ parser.add_argument('-o', '--output-dir')
parser.add_argument('--log-dir')
args, remaining_args = parser.parse_known_args()
cmd = ['docker', 'run', '--rm', '-it', '-u', '{}:{}'.format(UID, GID)]
cmd = ['docker', 'run', '--rm', '-it', '-u', f'{UID}:{GID}']
if args.input_dir is not None:
mapping = os.path.abspath(args.input_dir) + ':' + CONTAINER_INPUT_DIR
mapping = f'{os.path.abspath(args.input_dir)}:{CONTAINER_INPUT_DIR}'
cmd += ['-v', mapping]
remaining_args += ['-i', CONTAINER_INPUT_DIR]
if args.output_dir is not None:
mapping = os.path.abspath(args.output_dir) + ':' + CONTAINER_OUTPUT_DIR
mapping = f'{os.path.abspath(args.output_dir)}:{CONTAINER_OUTPUT_DIR}'
cmd += ['-v', mapping]
remaining_args += ['-o', CONTAINER_OUTPUT_DIR]
if args.log_dir is not None:
mapping = os.path.abspath(args.log_dir) + ':' + CONTAINER_LOG_DIR
mapping = '{os.path.abspath(args.log_dir)}:{CONTAINER_LOG_DIR}'
cmd += ['-v', mapping]
remaining_args += ['--log-dir', CONTAINER_LOG_DIR]
cmd.append(CONTAINER_IMAGE)