Update NLP Pipeline

2025-10-31 06:42:45 +00:00 · 2020-04-03 17:35:05 +02:00
parent 41910afb79
commit e061a7426d
5 changed files with 198 additions and 231 deletions
--- a/32
+++ b/32
@@ -11,8 +11,10 @@ ENV LANG=C.UTF-8
 # Install prerequisites
 RUN apt-get update \
 && apt-get install -y --no-install-recommends \
+      build-essential \
      python2.7 \
      python3.5 \
+      python3-dev \
      python3-pip \
      zip \
 && rm -rf /var/lib/apt/lists/* \
@@ -31,30 +33,24 @@ RUN tar -xzf "pyflow-${PYFLOW_VERSION}.tar.gz" \
      "pyflow-${PYFLOW_VERSION}" \
      "pyflow-${PYFLOW_VERSION}.tar.gz"

-ENV SPACY_MODEL_DE=de_core_news_sm \
-    SPACY_MODEL_EL=el_core_news_sm \
-    SPACY_MODEL_EN=en_core_web_sm \
-    SPACY_MODEL_ES=es_core_news_sm \
-    SPACY_MODEL_FR=fr_core_news_sm \
-    SPACY_MODEL_IT=it_core_news_sm \
-    SPACY_MODEL_NL=nl_core_news_sm \
-    SPACY_MODEL_PT=pt_core_news_sm \
-    SPACY_VERSION=2.2.0
+ENV SPACY_VERSION=2.2.4
+ENV SPACY_MODELS_VERSION=2.2.5
+
 RUN pip3 install \
      "spacy==${SPACY_VERSION}" \
- && python3 -m spacy download "${SPACY_MODEL_DE}-${SPACY_VERSION}" --direct \
- && python3 -m spacy download "${SPACY_MODEL_EL}-${SPACY_VERSION}" --direct \
- && python3 -m spacy download "${SPACY_MODEL_EN}-${SPACY_VERSION}" --direct \
- && python3 -m spacy download "${SPACY_MODEL_ES}-${SPACY_VERSION}" --direct \
- && python3 -m spacy download "${SPACY_MODEL_FR}-${SPACY_VERSION}" --direct \
- && python3 -m spacy download "${SPACY_MODEL_IT}-${SPACY_VERSION}" --direct \
- && python3 -m spacy download "${SPACY_MODEL_NL}-${SPACY_VERSION}" --direct \
- && python3 -m spacy download "${SPACY_MODEL_PT}-${SPACY_VERSION}" --direct
+ && python3 -m spacy download "de_core_news_sm-${SPACY_MODELS_VERSION}" --direct \
+ && python3 -m spacy download "el_core_news_sm-${SPACY_MODELS_VERSION}" --direct \
+ && python3 -m spacy download "en_core_web_sm-${SPACY_MODELS_VERSION}" --direct \
+ && python3 -m spacy download "es_core_news_sm-${SPACY_MODELS_VERSION}" --direct \
+ && python3 -m spacy download "fr_core_news_sm-${SPACY_MODELS_VERSION}" --direct \
+ && python3 -m spacy download "it_core_news_sm-${SPACY_MODELS_VERSION}" --direct \
+ && python3 -m spacy download "nl_core_news_sm-${SPACY_MODELS_VERSION}" --direct \
+ && python3 -m spacy download "pt_core_news_sm-${SPACY_MODELS_VERSION}" --direct


 # Install NLP pipeline
 COPY nlp /usr/local/bin
-COPY spacy_nlp /usr/local/bin
+COPY spacy-nlp /usr/local/bin


 ENTRYPOINT ["nlp"]
--- a/198
+++ b/198
@@ -9,147 +9,131 @@ Author: Patrick Jentsch <p.jentsch@uni-bielefeld.de>
 """


-import argparse
+from argparse import ArgumentParser
+from pyflow import WorkflowRunner
 import multiprocessing
 import os
 import sys
-from pyflow import WorkflowRunner


-def parse_arguments():
-    parser = argparse.ArgumentParser(
-        description=('Performs NLP of documents utilizing spaCy. The results '
-                     'are served as verticalized text files.')
-    )
-    parser.add_argument('-i', dest='input_dir', required=True)
-    parser.add_argument(
-        '-l',
-        choices=['de', 'el', 'en', 'es', 'fr', 'it', 'nl', 'pt'],
-        dest='lang',
-        required=True
-    )
-    parser.add_argument('-o', dest='output_dir', required=True)
-    parser.add_argument('--nCores',
+SPACY_MODELS = {'de': 'de_core_news_sm',
+                'el': 'el_core_news_sm',
+                'en': 'en_core_web_sm',
+                'es': 'es_core_news_sm',
+                'fr': 'fr_core_news_sm',
+                'it': 'it_core_news_sm',
+                'nl': 'nl_core_news_sm',
+                'pt': 'pt_core_news_sm'}
+
+
+def parse_args():
+    parser = ArgumentParser(description='NLP Pipeline utilizing spaCy.')
+    parser.add_argument('i')
+    parser.add_argument('o')
+    parser.add_argument('-l', '--language', choices=SPACY_MODELS.keys(),
+                        required=True)
+    parser.add_argument('--n-cores',
                        default=min(4, multiprocessing.cpu_count()),
-                        dest='n_cores',
-                        help='total number of cores available',
-                        required=False,
-                        type=int)
-    parser.add_argument('--zip',
-                        default='vrt-results',
-                        dest='zip',
-                        type=str,
-                        help='''package result files in zip bundles takes a
-                                string as a filename as an optional paramteer''',
-                        required=False)
-    parser.add_argument('--check-encoding',
-                        action='store_true',
-                        default=False,
-                        dest="check_encoding",
-                        help='''if used the nlp process will know hat the
-                                encoding of the input files is unkown and
-                                thus != utf-8. The process will try to determine
-                                the encoding of the input files and use this.
-                                encoding.'''
-                        )
+                        help='total number of cores available', type=int)
+    parser.add_argument('--check-encoding', action='store_true')
+    parser.add_argument('--log-dir')
+    parser.add_argument('--zip')
    return parser.parse_args()


-class NLPWorkflow(WorkflowRunner):
-    def __init__(self, args):
-        self.jobs = analyze_jobs(args.input_dir, args.output_dir)
-        self.lang = args.lang
-        self.n_cores = args.n_cores
-        self.output_dir = args.output_dir
-        self.zip = args.zip
-        self.check_encoding = args.check_encoding
+class NLPPipelineJob:
+    def __init__(self, file, output_dir):
+        self.file = file
+        self.name = os.path.basename(file).rsplit('.', 1)[0]
+        self.output_dir = output_dir
+
+
+class NLPPipeline(WorkflowRunner):
+    def __init__(self, check_encoding, jobs, lang, n_cores, output_dir, zip):
+        self.check_encoding = check_encoding
+        self.jobs = jobs
+        self.lang = lang
+        self.n_cores = n_cores
+        self.output_dir = output_dir
+        self.zip = zip

    def workflow(self):
-        if len(self.jobs) == 0:
+        if not self.jobs:
            return

        '''
        ' ##################################################
-        ' # Create output directories                      #
+        ' # mkdir_jobs                                     #
        ' ##################################################
        '''
-        create_output_directories_jobs = []
-        for index, job in enumerate(self.jobs):
-            cmd = 'mkdir -p "{}"'.format(job['output_dir'])
-            create_output_directories_jobs.append(
-                self.addTask(
-                    command=cmd,
-                    label='create_output_directories_job_-_{}'.format(index)
-                )
-            )
+        mkdir_jobs = []
+        for i, job in enumerate(self.jobs):
+            cmd = 'mkdir'
+            cmd += ' -p'
+            cmd += ' "{}"'.format(job.output_dir)
+            lbl = 'mkdir_job_-_{}'.format(i)
+            mkdir_jobs.append(self.addTask(command=cmd, label=lbl))

        '''
        ' ##################################################
-        ' # Natural language processing                    #
+        ' # spacy_nlp_jobs                                 #
        ' ##################################################
        '''
-        nlp_jobs = []
-        nlp_job_n_cores = min(
-            self.n_cores,
-            max(1, int(self.n_cores / len(self.jobs)))
-        )
-        for index, job in enumerate(self.jobs):
-            cmd = 'spacy_nlp -l "{}" "{}" "{}" {}'.format(
-                self.lang,
-                job['path'],
-                os.path.join(job['output_dir'], job['name'] + '.vrt'),
-                "--check-encoding" if self.check_encoding else ""
-            )
-            nlp_jobs.append(
-                self.addTask(
-                    command=cmd,
-                    dependencies='create_output_directories_job_-_{}'.format(
-                        index
-                    ),
-                    label='nlp_job_-_{}'.format(index),
-                    nCores=nlp_job_n_cores
-                )
-            )
+        spacy_nlp_jobs = []
+        n_cores = min(self.n_cores, max(1, int(self.n_cores / len(self.jobs))))
+        for i, job in enumerate(self.jobs):
+            output_file = os.path.join(job.output_dir,
+                                       '{}.vrt'.format(job.name))
+            cmd = 'spacy-nlp "{}" "{}"'.format(job.file, output_file)
+            cmd += ' -l "{}"'.format(self.lang)
+            cmd += ' --check-encoding' if self.check_encoding else ''
+            deps = 'mkdir_job_-_{}'.format(i)
+            lbl = 'spacy_nlp_job_-_{}'.format(i)
+            spacy_nlp_jobs.append(self.addTask(command=cmd, dependencies=deps,
+                                               label=lbl, nCores=n_cores))

-        if zip:
-            vrt_zip_jobs = []
-            vrt_zip_job_dependencies = nlp_jobs
-            cmd = 'cd "%s" && zip -m "%s"-nlp.zip */*.vrt -x "pyflow.data*" && cd -' % (
-                self.output_dir,
-                self.zip
-            )
-            vrt_zip_jobs.append(
-                self.addTask(
-                    command=cmd,
-                    dependencies=vrt_zip_job_dependencies,
-                    label='vrt_zip_job'
-                )
-            )
+        '''
+        ' ##################################################
+        ' # zip_jobs                                       #
+        ' ##################################################
+        '''
+        zip_jobs = []
+        if self.zip is not None:
+            cmd = 'cd "{}"'.format(self.output_dir)
+            cmd += ' && '
+            cmd += 'zip'
+            cmd += ' -m'
+            cmd += ' -r'
+            cmd += ' "{}_-_vrt" .'.format(self.zip)
+            cmd += ' -x "pyflow.data*"'
+            cmd += ' -i "*.vrt"'
+            cmd += ' && '
+            cmd += 'cd -'
+            deps = spacy_nlp_jobs
+            lbl = 'zip_job'
+            zip_jobs.append(self.addTask(command=cmd, dependencies=deps,
+                                         label=lbl))


-def analyze_jobs(input_dir, output_dir):
+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)):
-            jobs += analyze_jobs(os.path.join(input_dir, file),
+            jobs += collect_jobs(os.path.join(input_dir, file),
                                 os.path.join(output_dir, file))
        elif file.endswith('.txt'):
-            jobs.append({'filename': file,
-                         'name': file.rsplit('.', 1)[0],
-                         'output_dir': os.path.join(output_dir, file),
-                         'path': os.path.join(input_dir, file)})
-
+            jobs.append(NLPPipelineJob(os.path.join(input_dir, file),
+                                       os.path.join(output_dir, file)))
    return jobs


 def main():
-    args = parse_arguments()
-
-    wflow = NLPWorkflow(args)
-
-    retval = wflow.run(dataDirRoot=args.output_dir, nCores=args.n_cores)
-
+    args = parse_args()
+    jobs = collect_jobs(args.i, args.o)
+    nlp_pipeline = NLPPipeline(args.check_encoding, jobs, args.language,
+                               args.n_cores, args.o, args.zip)
+    retval = nlp_pipeline.run(dataDirRoot=(args.log_dir or args.o),
+                              nCores=args.n_cores)
    sys.exit(retval)


--- a/76
+++ b/76
@@ -0,0 +1,76 @@
+#!/usr/bin/env python3.5
+# coding=utf-8
+
+from argparse import ArgumentParser
+from xml.sax.saxutils import escape
+import chardet
+import spacy
+import textwrap
+
+
+SPACY_MODELS = {'de': 'de_core_news_sm',
+                'el': 'el_core_news_sm',
+                'en': 'en_core_web_sm',
+                'es': 'es_core_news_sm',
+                'fr': 'fr_core_news_sm',
+                'it': 'it_core_news_sm',
+                'nl': 'nl_core_news_sm',
+                'pt': 'pt_core_news_sm'}
+
+
+# Parse the given arguments
+parser = ArgumentParser(description=('Tag a text file with spaCy and save it '
+                                     'as a verticalized text file.'))
+parser.add_argument('i', metavar='txt-sourcefile')
+parser.add_argument('o', metavar='vrt-destfile')
+parser.add_argument('-l', '--language', choices=SPACY_MODELS.keys(),
+                    required=True)
+parser.add_argument('--check-encoding', action='store_true')
+args = parser.parse_args()
+
+
+# If requested: Check the encoding of the text contents from the input file
+# Else: Use utf-8
+if args.check_encoding:
+    with open(args.i, "rb") as input_file:
+        bytes = input_file.read()
+        encoding = chardet.detect(bytes)['encoding']
+else:
+    encoding = 'utf-8'
+
+
+# Load the text contents from the input file
+with open(args.i, encoding=encoding) as input_file:
+    text = input_file.read()
+    # spaCys NLP is limited to strings with maximum 1 million characters at
+    # once. So we split it into suitable chunks.
+    text_chunks = textwrap.wrap(text, 1000000, break_long_words=False)
+    # the text variable potentially occupies a lot of system memory and is no
+    # longer needed...
+    del text
+
+
+# Setup the spaCy toolkit by loading the chosen language model
+nlp = spacy.load(SPACY_MODELS[args.language])
+
+
+# Create the output file in verticalized text format
+# See: http://cwb.sourceforge.net/files/CWB_Encoding_Tutorial/node3.html
+output_file = open(args.o, 'w+')
+output_file.write('<?xml version="1.0" encoding="UTF-8"?>\n<corpus>\n<text>\n')
+for text_chunk in text_chunks:
+    doc = nlp(text_chunk)
+    for sent in doc.sents:
+        output_file.write('<s>\n')
+        for token in sent:
+            # Skip whitespace tokens
+            if token.text.isspace():
+                continue
+            output_file.write('{}'.format(escape(token.text))
+                              + '\t{}'.format(escape(token.lemma_))
+                              + '\t{}'.format(token.pos_)
+                              + '\t{}'.format(token.tag_)
+                              + '\t{}\n'.format(token.ent_type_ or 'NULL'))
+        output_file.write('</s>\n')
+output_file.write('</text>\n</corpus>')
+output_file.close()
--- a/83
+++ b/83
@@ -1,83 +0,0 @@
-#!/usr/bin/env python3.5
-# coding=utf-8
-
-from xml.sax.saxutils import escape
-import argparse
-import chardet
-import spacy
-import textwrap
-
-parser = argparse.ArgumentParser(
-    description=('Tag a text file with spaCy and save it as a verticalized '
-                 'text file.')
-)
-parser.add_argument('i', metavar='txt-sourcefile')
-parser.add_argument('-l',
-                    choices=['de', 'el', 'en', 'es', 'fr', 'it', 'nl', 'pt'],
-                    dest='lang',
-                    required=True)
-parser.add_argument('o', metavar='vrt-destfile')
-parser.add_argument('--check-encoding',
-                    default=False,
-                    action='store_true',
-                    dest='check_encoding'
-                    )
-args = parser.parse_args()
-
-SPACY_MODELS = {'de': 'de_core_news_sm',
-                'el': 'el_core_news_sm',
-                'en': 'en_core_web_sm',
-                'es': 'es_core_news_sm',
-                'fr': 'fr_core_news_sm',
-                'it': 'it_core_news_sm',
-                'nl': 'nl_core_news_sm',
-                'pt': 'pt_core_news_sm'}
-
-# Set the language model for spacy
-nlp = spacy.load(SPACY_MODELS[args.lang])
-
-# Try to determine the encoding of the text in the input file
-if args.check_encoding:
-    with open(args.i, "rb") as input_file:
-        bytes = input_file.read()
-        encoding = chardet.detect(bytes)['encoding']
-else:
-    encoding = 'utf-8'
-# Read text from the input file and if neccessary split it into parts with a
-# length of less than 1 million characters.
-with open(args.i, encoding=encoding) as input_file:
-    text = input_file.read()
-    texts = textwrap.wrap(text, 1000000, break_long_words=False)
-    text = None
-
-# Create and open the output file
-output_file = open(args.o, 'w+')
-
-output_file.write('<?xml version="1.0" encoding="UTF-8"?>\n'
-                  '<corpus>\n'
-                  '<text>\n')
-for text in texts:
-    # Run spacy nlp over the text (partial string if above 1 million chars)
-    doc = nlp(text)
-    for sent in doc.sents:
-        output_file.write('<s>\n')
-        for token in sent:
-            # Skip whitespace tokens like "\n" or "\t"
-            if token.text.isspace():
-                continue
-            # Write all information in .vrt style to the output file
-            # text, lemma, simple_pos, pos, ner
-            output_file.write(
-                '{}\t{}\t{}\t{}\t{}\n'.format(
-                    escape(token.text),
-                    escape(token.lemma_),
-                    token.pos_,
-                    token.tag_,
-                    token.ent_type_ if token.ent_type_ != '' else 'NULL'
-                )
-            )
-        output_file.write('</s>\n')
-output_file.write('</text>\n'
-                  '</corpus>')
-
-output_file.close()
--- a/wrapper/nlp
+++ b/wrapper/nlp
@@ -1,35 +1,29 @@
 #!/usr/bin/env python3
 # coding=utf-8

-import argparse
+from argparse import ArgumentParser
 import os
 import subprocess

-container_image = 'gitlab.ub.uni-bielefeld.de:4567/sfb1288inf/nlp:latest'
-container_input_dir = '/input'
-container_output_dir = '/output'
-uid = str(os.getuid())
-gid = str(os.getgid())
+CONTAINER_IMAGE = 'gitlab.ub.uni-bielefeld.de:4567/sfb1288inf/nlp:latest'
+CONTAINER_INPUT_DIR = '/input'
+CONTAINER_OUTPUT_DIR = '/output'
+UID = str(os.getuid())
+GID = str(os.getgid())

-parser = argparse.ArgumentParser(add_help=False)
-parser.add_argument('-i',
-                    dest='input_dir',
-                    required=False)
-parser.add_argument('-o',
-                    dest='output_dir',
-                    required=False)
+parser = ArgumentParser(add_help=False)
+parser.add_argument('-i')
+parser.add_argument('-o')
 args, remaining_args = parser.parse_known_args()

-cmd = ['docker', 'run', '--rm', '-it', '-u', uid + ':' + gid]
-if args.input_dir is not None:
-    host_input_dir = os.path.abspath(args.input_dir)
-    cmd += ['-v', host_input_dir + ':' + container_input_dir]
-    remaining_args += ['-i', container_input_dir]
-if args.output_dir is not None:
-    host_output_dir = os.path.abspath(args.output_dir)
-    cmd += ['-v', host_output_dir + ':' + container_output_dir]
-    remaining_args += ['-o', container_output_dir]
-cmd.append(container_image)
+cmd = ['docker', 'run', '--rm', '-it', '-u', '{}:{}'.format(UID, GID)]
+if args.o is not None:
+    cmd += ['-v', '{}:{}'.format(os.path.abspath(args.o), CONTAINER_OUTPUT_DIR)]
+    remaining_args.insert(0, CONTAINER_OUTPUT_DIR)
+if args.i is not None:
+    cmd += ['-v', '{}:{}'.format(os.path.abspath(args.i), CONTAINER_INPUT_DIR)]
+    remaining_args.insert(0, CONTAINER_INPUT_DIR)
+cmd.append(CONTAINER_IMAGE)
 cmd += remaining_args

 subprocess.run(cmd)