biblenlp-corpus.py

# coding=utf-8
# Copyright 2020 HuggingFace Datasets Authors.
#
# 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.

# Lint as: python3
import os

import datasets
from tqdm import tqdm
import numpy as np
import ijson

_VERSION = "0.0.1"
_LANGUAGES = ['san', 'dji', 'eng', 'epo', 'cmn', 'cop', 'deu', 'hrv', 'kky', 'haw', 'bre', 'nld', 'bea', 'pon', 'pes', 'dif', 'heb', 'cha', 'kos', 'nay', 'hbo', 'arb', 'ita', 'rus', 'awk', 'arp', 'hat', 'swh', 'lat', 'bla', 'ces', 'spa', 'vie', 'grc', 'srp', 'fra', 'hlt', 'ukr', 'wro', 'ton', 'als', 'jpn']
_DESCRIPTION = "Bible Parallel Corpus"
_HOMEPAGE = 'https://github.com/BibleNLP/ebible-corpus'
_PAIR = 'all'
_CITATION = """\
@InProceedings{huggingface:dataset,
title = {A great new dataset},
author={huggingface, Inc.
},
year={2020}
}
"""
_LICENSE = 'Creative Commons and Public Domain, specified in the dataset'
_FILES = {'corpus':'corpus.json','vref':'vref.txt'}

class BiblenlpCorpusConfig(datasets.BuilderConfig):
    def __init__(self, languages=[], pair='all', **kwargs):
        '''
        languages: list of languages to include
        pair: 'all', 'range', or 'single' to specify whether verse ranges, single pairings, or all pairs are included
        **kwargs: additional arguments to pass to the superclass'''
        super(BiblenlpCorpusConfig, self).__init__(name=f'{"-".join([x for x in languages])}', **kwargs)
        self.languages = languages
        self.pair = pair

class BiblenlpCorpus(datasets.GeneratorBasedBuilder):
    BUILDER_CONFIGS = [
        BiblenlpCorpusConfig(
            languages=_LANGUAGES, 
            pair=_PAIR,
            description = f'Parallel Bible verses with {_PAIR} pairings of {"-".join([x for x in _LANGUAGES])} languages',
        )
    ]

    BUILDER_CONFIG_CLASS = BiblenlpCorpusConfig

    def _info(self):
        return datasets.DatasetInfo(
            description=_DESCRIPTION,
            features=datasets.Features(
                {
                    "translation": datasets.features.TranslationVariableLanguages(languages=_LANGUAGES),     
                    "files": datasets.features.Sequence( {'lang':datasets.Value("string"), 'file':datasets.Value("string")} ),                
                    "ref": datasets.features.Sequence( datasets.Value("string") ),
                    "licenses": datasets.features.Sequence( datasets.Value("string")  ),
                    "copyrights": datasets.features.Sequence( datasets.Value("string")  )
                },        
                ),
            supervised_keys=None,
            homepage=_HOMEPAGE,
            citation=_CITATION,
            license=_LICENSE,
            version=_VERSION
            )   #

    def _split_generators(self, dl_manager):
        downloaded_files = dl_manager.download(_FILES)
        
        def parse_json(json_filename, langs):
            with open(json_filename, 'rb') as input_file:
                # load json iteratively
                parser = ijson.parse(input_file)
                parsed_dict = {lang: [] for lang in langs}
                idx = {lang: -1 for lang in langs}
                for prefix, event, value in parser:
                    #print(f'prefix:{prefix}, event:{event}, value:{value}')
                    if any([prefix.startswith(f'{lang}.') for lang in langs]):
                        #print(f'prefix:{prefix}, event:{event}, value:{value}')
                        if event == 'start_map':
                            idx[prefix.split('.')[0]] += 1
                            tmpdict ={}
                            tmpdict['verses']=[]
                        if prefix.endswith('verses.item'):
                            tmpdict['verses'].append(value)
                        if prefix.endswith('copyright'):
                            tmpdict['copyright'] = value
                        if prefix.endswith('text'):
                            tmpdict['text'] = value              
                        if prefix.endswith('file'):
                            tmpdict['file'] = value            
                        if prefix.endswith('license'):
                            tmpdict['license'] = value
                        if event == 'end_map':
                            #write the dictionary
                            parsed_dict[prefix.split('.')[0]].append( tmpdict.copy() )
            return(parsed_dict)
        
        def define_splits(corpus_slice, langs):
            verse2split = {}
            langtexts = {}
            textverses = {}
            vindexes = {}
            vgroups = []
            versesets ={}

            np.random.seed(42)

            for lang in tqdm(langs):
                langtexts[lang] = set()
                textverses[lang]={}
                vindexes[lang]={}
                for idx,line in enumerate(corpus_slice[lang]):
                    langtexts[lang].add(line['file'])
                    for v in line['verses']:
                        if not textverses[lang].get(line['file']): 
                            textverses[lang][line['file']] = set()
                        textverses[lang][line['file']].add(v)
                        if not vindexes[lang].get(v):
                            vindexes[lang][v] = [idx] 
                        else: 
                            vindexes[lang][v].append(idx)

            #print(list(vindexes['eng'].keys())[0:10])
            #for f in langtexts['eng']:
            #    print(len(textverses['eng'][f]))

            for line in tqdm(corpus_slice[langs[0]]):
                versesets = {line['file']:line['verses']}
                if any([verse2split.get(z) for z in line['verses']]):
                    for verse in line['verses']:
                        if verse2split.get(verse):
                            prevsplit = verse2split.get(verse)
                            break
                    split = prevsplit
                else:  
                    split = np.random.choice(['train','test','dev'],p=[.9,.05,.05])
                if not all([verse2split.get(z) for z in line['verses']]):
                    all_verses = set()
                    for v in line['verses']:
                        all_verses.add(v)
                    while True:
                        verses_added = False  
                        idxes = {k:set() for k in langs}
                        #get indexes for verses
                        for v in all_verses:
                            for lang in langs:
                                if vindexes[lang].get(v):
                                    for idx in vindexes[lang][v]:
                                        idxes[lang].add(idx)

                        for lang in langs:
                            for compline in [corpus_slice[lang][x] for x in idxes[lang] if x != set()]:

                                if all(x in textverses[lang][compline['file']] for x in all_verses) and any([x in list(all_verses) for x in compline['verses']]):
                                    if not versesets.get(compline['file']):
                                        versesets[compline['file']] = compline['verses'].copy()               
                                    else: 
                                        versesets[compline['file']].extend(compline['verses'].copy()) 
                                    for v in compline['verses']:
                                        pre_size = len(all_verses)
                                        all_verses.add(v)
                                        if len(all_verses) > pre_size: 
                                            verses_added = True      

                        if verses_added == False or all([set(versesets[q]) == all_verses for q in versesets.keys()]):
                            vgroups.append(all_verses)
                            for v in all_verses:
                                verse2split[v] = split
                            break

            return(vgroups,vindexes, verse2split)    


        corpus_slice = parse_json(downloaded_files['corpus'], self.config.languages)
        vgroups, vindexes, verse2split = define_splits(corpus_slice, self.config.languages)


        return [
            datasets.SplitGenerator(
                name=datasets.Split.TRAIN,
                gen_kwargs={
                    'langs': self.config.languages,
                    'corpus': corpus_slice,
                    'vgroups': vgroups,
                    'vindexes': vindexes,
                    'v2split': verse2split,
                    'split': 'train',
                    'vref': downloaded_files['vref'],
                }
            ),
            datasets.SplitGenerator(
                name=datasets.Split.VALIDATION,
                gen_kwargs={
                    'langs': self.config.languages,
                    'corpus': corpus_slice,
                    'vgroups': vgroups,
                    'vindexes': vindexes,
                    'v2split': verse2split,
                    'split': 'dev',
                    'vref': downloaded_files['vref'],
                }
            ),
            datasets.SplitGenerator(
                name=datasets.Split.TEST,
                gen_kwargs={
                    'langs': self.config.languages,
                    'corpus': corpus_slice,
                    'vgroups': vgroups,
                    'vindexes': vindexes,
                    'v2split': verse2split,
                    'split': 'test',
                    'vref': downloaded_files['vref'],
                }
            ),                                  
        ]



        

    def _generate_examples(self, vref, corpus, vgroups, vindexes, v2split, split, langs):
        #print(f'Generating {split} examples...')
        with open(vref, 'r') as txtfile:
            #print('file opened')
            lines = txtfile.readlines()
            #print('lines read')
            verse2index = {k.strip():v for v,k in enumerate(lines) if k.strip() != ''}
            #print('v2i created')

        def order_verses(verse_list):
            lines_list = [int(verse2index[x]) for x in verse_list]
            verse_order = np.argsort(lines_list)
            return(verse_order)

        trans_license = {}
        trans_copyright = {}
        id = -1
        #print(f'beginning groups, starting with {list(vgroups)[0]}')
        for group in vgroups:
            #print(group)
            if v2split.get(list(group)[0]) == split:
                #print('correct split')
                if len(group) > 1:
                    v_order = order_verses(group)
                    o_group = list(np.array(list(group))[v_order])

                else:
                    o_group = list(group)
                trans_dict={k:{} for k in langs}
                trans_texts = {k:{} for k in langs}
                used_idxes = {k:[] for k in langs}
                for v in o_group:
                    #print(f'trying verse {v}')
                    single_texts = {k:{} for k in langs}
                    single_dict = {k:{} for k in langs}
                    for lang in langs:
                        if vindexes[lang].get(v):
                            try:
                                for i in range(0,len(list(vindexes[lang][v]))):
                                    if list(vindexes[lang][v])[i] not in used_idxes[lang]:
                                        used_idxes[lang].append(vindexes[lang][v])
                                        line = corpus[lang][vindexes[lang][v][i]]
                                        if not trans_texts.get(line['file']):
                                            trans_texts[lang][line['file']]=line['text']
                                        else:
                                            trans_texts[lang][line['file']] += f' {line["text"]}'
                                        if line.get('license'):
                                            trans_license[line['file']]=line['license']
                                        if line.get('copyright'):
                                            trans_copyright[line['file']]=line['copyright']
                                        if len(line['verses']) == 1:
                                            single_texts[lang][line['file']] = line['text']
                            except:
                                print(lang,v)
                                raise
                    
                        single_dict[lang]=list(single_texts[lang].values())
                        trans_dict[lang]=list(trans_texts[lang].values())
                        
                    single_file = {x:list(single_texts[x].keys()) for x in langs}
                    single_lic_list =[]
                    single_copy_list = []
                    for lang in langs:
                        single_lic_list.extend([trans_license.get(x,'') for x in single_file[lang]])
                        single_copy_list.extend([trans_copyright.get(x,'') for x in single_file[lang]])
                    if all([single_dict.get(x) and single_dict.get(x) != [{}] and list(single_dict.get(x)) for x in langs]) and len(list(single_dict.keys())) == len(langs) and self.config.pair != 'range':
                        id = id + 1 
                        sfile_list = []
                        for key in langs:
                            for value in single_file.get(key):
                                sfile_list.append({'lang':key,'file':value})

                        #print('outputting single example')
                       # print(id, single_dict, sfile_list, v, single_lic_list, single_copy_list)
                        try:
                            yield(id, {'translation': single_dict, 'files': sfile_list, 'ref':[v], 'licenses':single_lic_list, 'copyrights':single_copy_list}) 
                        except:
                            print(id, single_dict, sfile_list, v, single_lic_list, single_copy_list)
                            raise

                file_list = {x:list(trans_texts[x].keys()) for x in langs}
                #license_list = [trans_license.get(x,'') for x in [y for y in file_list.values()]]
                #copyright_list = [trans_copyright.get(x,'') for x in [y for y in file_list.values()]]
                license_list = []
                copyright_list = []
                for lang in langs:
                    license_list.extend([trans_license.get(x,'') for x in file_list[lang]])
                    copyright_list .extend([trans_copyright.get(x,'') for x in file_list[lang]])
                if len(o_group)>1 and all([trans_dict.get(x)  and trans_dict.get(x) != [{}] and list(trans_dict.get(x)) for x in langs]) and len(list(trans_dict.keys())) == len(langs) and self.config.pair != 'single':
                    id = id + 1
                    ofile_list = []
                    for key in langs:
                        for value in file_list.get(key):
                            ofile_list.append({'lang':key,'file':value})
                    #print('outputting range example')
                    try:
                        yield(id, {'translation': trans_dict, 'files': ofile_list, 'ref':o_group, 'licenses':license_list, 'copyrights':copyright_list})
                    except:
                        print(id, trans_dict, ofile_list, o_group, license_list, copyright_list)
                        raise