"""
This file contains all the indexers.
"""
from enum import Enum
import pandas as pd
import os
import json
import tempfile
import contextlib
import threading
import select
import math
from warnings import warn
from deprecated import deprecated
from collections import deque
from typing import List, Dict, Union, Any, Callable, Type
import more_itertools
import pyterrier as pt
from pyterrier.terrier.stemmer import TerrierStemmer
from pyterrier.terrier.tokeniser import TerrierTokeniser
from pyterrier.terrier.stopwords import TerrierStopwords
# These classes are only defined after pt.java.init() in pyterrier.terrier.java._post_init
DocListIterator: Any = None
PythonListIterator: Any = None
FlatJSONDocumentIterator: Any = None
TQDMCollection: Any = None
TQDMSizeCollection: Any = None
# lastdoc ensures that a Document instance from a Collection is not GCd before Java has used it.
lastdoc = None
type_to_class = {
'trec' : 'org.terrier.indexing.TRECCollection',
'trecweb' : 'org.terrier.indexing.TRECWebCollection',
'warc' : 'org.terrier.indexing.WARC10Collection'
}
@pt.java.required
def createCollection(files_path : List[str], coll_type : str = 'trec', props = {}):
if coll_type in type_to_class:
collectionClzName = type_to_class[coll_type]
else:
collectionClzName = coll_type
_props = pt.java.J.HashMap()
for k, v in props.items():
_props[k] = v
pt.terrier.J.ApplicationSetup.getProperties().putAll(_props)
cls_string = pt.java.J.String._class
cls_list = pt.java.J.List._class
if len(files_path) == 0:
raise ValueError("list files_path cannot be empty")
asList = createAsList(files_path)
colObj = pt.terrier.J.CollectionFactory.loadCollections(
collectionClzName.split(","),
[cls_list, cls_string, cls_string, cls_string],
[asList, pt.terrier.J.TagSet.TREC_DOC_TAGS, "", ""])
return colObj
[docs]
def treccollection2textgen(
files : List[str],
meta : List[str] = ["docno"],
meta_tags : Dict[str,str] = {"text":"ELSE"},
verbose = False,
num_docs = None,
tag_text_length : int = 4096):
"""
Creates a generator of dictionaries on parsing TREC formatted files. This is useful
for parsing TREC-formatted corpora in indexers like IterDictIndexer, or similar
indexers in other plugins (e.g. ColBERTIndexer).
"Arguments:
-" files(List[str]): list of files to parse in TREC format.
- meta(List[str]): list of attributes to expose in the dictionaries as metadata.
- meta_tags(Dict[str,str]): mapping of TREC tags as metadata.
- tag_text_length(int): maximium length of metadata. Defaults to 4096.
- verbose(bool): set to true to show a TQDM progress bar. Defaults to True.
- num_docs(int): a hint for TQDM to size the progress bar based on document counts rather than file count.
Example::
files = pt.io.find_files("/path/to/Disk45")
gen = pt.index.treccollection2textgen(files)
index = pt.IterDictIndexer("./index45").index(gen)
"""
props = {
"TrecDocTags.doctag": "DOC",
"TrecDocTags.idtag": "DOCNO",
"TrecDocTags.skip": "DOCHDR",
"TrecDocTags.casesensitive": "false",
# Should the tags from which we create abstracts be case-sensitive?
'TaggedDocument.abstracts.tags.casesensitive':'false'
}
props['TaggedDocument.abstracts'] = ','.join(meta_tags.keys())
# The tags from which to save the text. ELSE is special tag name, which means anything not consumed by other tags.
props['TaggedDocument.abstracts.tags'] = ','.join(meta_tags.values())
# The max lengths of the abstracts. Abstracts will be truncated to this length. Defaults to empty.
props['TaggedDocument.abstracts.lengths'] = ','.join([str(tag_text_length)] * len(meta_tags) )
collection = createCollection(files, props=props)
if verbose:
if num_docs is not None:
collection = TQDMSizeCollection(collection, num_docs)
else:
collection = TQDMCollection(collection)
while collection.nextDocument():
d = collection.getDocument()
while not d.endOfDocument():
d.getNextTerm()
rtr = {}
for k in meta:
rtr[k] = d.getProperty(k)
for k in meta_tags:
rtr[k] = d.getProperty(k)
yield rtr
@pt.java.required
def _TaggedDocumentSetup(
meta : Dict[str,int], #mapping from meta-key to length
meta_tags : Dict[str,str] #mapping from meta-key to tag
):
"""
Property setup for TaggedDocument etc to generate abstracts
"""
abstract_tags=meta_tags.values()
abstract_names=meta_tags.keys()
abstract_lengths=[str(meta[name]) for name in abstract_names]
pt.terrier.J.ApplicationSetup.setProperty("TaggedDocument.abstracts", ",".join(abstract_names))
# The tags from which to save the text. ELSE is special tag name, which means anything not consumed by other tags.
pt.terrier.J.ApplicationSetup.setProperty("TaggedDocument.abstracts.tags", ",".join(abstract_tags))
# The max lengths of the abstracts. Abstracts will be truncated to this length. Defaults to empty.
pt.terrier.J.ApplicationSetup.setProperty("TaggedDocument.abstracts.lengths", ",".join(abstract_lengths))
# Should the tags from which we create abstracts be case-sensitive
pt.terrier.J.ApplicationSetup.setProperty("TaggedDocument.abstracts.tags.casesensitive", "false")
@pt.java.required
def _FileDocumentSetup(
meta : Dict[str,int], #mapping from meta-key to length
meta_tags : Dict[str,str] #mapping from meta-key to tag
):
"""
Property setup for FileDocument etc to generate abstracts
"""
meta_name_for_abstract = None
for k, v in meta_tags.items():
if v == 'ELSE':
meta_name_for_abstract = k
if meta_name_for_abstract is None:
return
if meta_name_for_abstract not in meta:
raise ValueError("You need to specify a meta length for " + meta_name_for_abstract)
abstract_length = meta[meta_name_for_abstract]
pt.terrier.J.ApplicationSetup.setProperty("FileDocument.abstract", meta_name_for_abstract)
pt.terrier.J.ApplicationSetup.setProperty("FileDocument.abstract.length", str(abstract_length))
@pt.java.required
def createAsList(files_path : Union[str, List[str]]):
"""
Helper method to be used by child indexers to add files to Java List
Returns:
Created Java List
"""
if isinstance(files_path, str):
asList = pt.java.J.Arrays.asList(files_path)
elif isinstance(files_path, list):
asList = pt.java.J.Arrays.asList(*files_path)
else:
raise ValueError(f"{files_path}: {type(files_path)} must be a List[str] or str")
return asList
# Using enum class create enumerations
[docs]
class IndexingType(Enum):
"""
This enum is used to determine the type of index built by Terrier. The default is CLASSIC. For more information,
see the relevant Terrier `indexer <https://terrier-core.readthedocs.io/en/latest/indexer_details.html>`_
and `realtime <https://terrier-core.readthedocs.io/en/latest/realtime_indices.html>`_ documentation.
"""
CLASSIC = 1 #: A classical indexing regime, which also creates a direct index structure, useful for query expansion
SINGLEPASS = 2 #: A single-pass indexing regime, which builds an inverted index directly. No direct index structure is created. Typically is faster than classical indexing.
MEMORY = 3 #: An in-memory index. No persistent index is created.
@pt.java.required
class TerrierIndexer:
"""
This is the super class for all of the Terrier-based indexers exposed by PyTerrier. It hosts common configuration
for all index types.
"""
schematic = {'label': 'TerrierIndexer'}
default_properties = {
"TrecDocTags.doctag": "DOC",
"TrecDocTags.idtag": "DOCNO",
"TrecDocTags.skip": "DOCHDR",
"TrecDocTags.casesensitive": "false",
"trec.collection.class": "TRECCollection",
}
def __init__(self,
index_path : str,
*,
blocks : bool = False,
overwrite: bool = False,
verbose : bool = False,
meta_reverse : List[str] = ["docno"],
stemmer : Union[None, str, TerrierStemmer] = TerrierStemmer.porter,
stopwords : Union[None, TerrierStopwords, List[str]] = TerrierStopwords.terrier,
tokeniser : Union[str,TerrierTokeniser] = TerrierTokeniser.english,
type=IndexingType.CLASSIC,
properties : Dict[str,str] = {}
):
"""
Constructor called by all indexer subclasses. All arguments listed below are available in
IterDictIndexer, DFIndexer, TRECCollectionIndexer and FilesIndsexer.
:param index_path: Directory to store index. Ignored for IndexingType.MEMORY.
:param blocks: Create indexer with blocks if true, else without blocks. Default is False.
:param overwrite: If index already present at `index_path`, True would overwrite it, False throws an Exception. Default is False.
:param verbose: Provide progess bars if possible. Default is False.
:param stemmer: the stemmer to apply. Default is ``TerrierStemmer.porter``.
:param stopwords: the stopwords list to apply. Default is ``TerrierStemmer.terrier``.
:param tokeniser: the stemmer to apply. Default is ``TerrierTokeniser.english``.
:param type: the specific indexing procedure to use. Default is ``IndexingType.CLASSIC``.
:param properties: Terrier properties that you wish to overrride.
"""
if type is IndexingType.MEMORY:
self.path = None
else:
self.path = os.path.join(index_path, "data.properties")
if not os.path.isdir(index_path):
os.makedirs(index_path)
self.index_called = False
self.index_dir = index_path
self.blocks = blocks
self.type = type
self.stemmer = TerrierStemmer._to_obj(stemmer)
self.stopwords, self.stopword_list = TerrierStopwords._to_obj(stopwords)
self.tokeniser = TerrierTokeniser._to_obj(tokeniser)
self.properties = pt.java.J.Properties()
self.setProperties(**self.default_properties)
for k,v in properties.items():
self.properties[k] = v
self.overwrite = overwrite
self.verbose = verbose
self.meta_reverse = meta_reverse
self.cleanup_hooks: List[Callable[[TerrierIndexer, Any], None]] = []
def setProperty(self, k, v):
"""
Set the named property to the specified value.
Args:
k(str): name of the Terrier property
v(str): value of the Terrier property
Usage::
indexer.setProperty("termpipelines", "")
"""
self.properties.put(str(k), str(v))
def setProperties(self, **kwargs):
"""
Set the properties to the given ones.
Args:
**kwargs: Properties to set to.
Usage:
>>> setProperties(**{property1:value1, property2:value2})
"""
for key, value in kwargs.items():
self.properties.put(str(key), str(value))
def checkIndexExists(self):
"""
Check if index exists at the `path` given when object was created
"""
if self.path is None:
return
if os.path.isfile(self.path):
if not self.overwrite:
raise ValueError("Index already exists at " + self.path)
if self.index_called:
raise Exception("Index method can be called only once")
def createIndexer(self):
"""
Checks `self.type` and
- if false, check `blocks` and create a BlockIndexer if true, else create BasicIndexer
- if true, check `blocks` and create a BlockSinglePassIndexer if true, else create BasicSinglePassIndexer
Returns:
Created Java object extending org.terrier.structures.indexing.Indexer
"""
Indexer, _ = self.indexerAndMergerClasses()
if Indexer is pt.terrier.J.BasicMemoryIndexer:
index = Indexer()
else:
index = Indexer(self.index_dir, "data")
assert index is not None
return index
def indexerAndMergerClasses(self):
"""
Check `single_pass` and
- if false, check `blocks` and create a BlockIndexer if true, else create BasicIndexer
- if true, check `blocks` and create a BlockSinglePassIndexer if true, else create BasicSinglePassIndexer
Returns:
type objects for indexer and merger for the given configuration
"""
# configure the meta index
self.properties['indexer.meta.forward.keys'] = ','.join(self.meta.keys())
self.properties['indexer.meta.forward.keylens'] = ','.join([str(le) for le in self.meta.values()])
self.properties['indexer.meta.reverse.keys'] = ','.join(self.meta_reverse)
# configure the term pipeline
if 'termpipelines' in self.properties:
# use existing configuration if present
warn(
"Setting of termpipelines property directly is deprecated", stacklevel=4, category=DeprecationWarning)
else:
termpipeline = []
TerrierStopwords._indexing_config(self.stopwords, self.stopword_list, termpipeline, self.properties, self.cleanup_hooks)
stemmer_clz = TerrierStemmer._to_class(self.stemmer)
if stemmer_clz is not None:
termpipeline.append(stemmer_clz)
self.properties['termpipelines'] = ','.join(termpipeline)
if "tokeniser" in self.properties:
warn(
"Setting of tokeniser property directly is deprecated", stacklevel=4, category=DeprecationWarning)
else:
self.properties['tokeniser'] = TerrierTokeniser._to_class(self.tokeniser)
# inform terrier of all properties
pt.terrier.J.ApplicationSetup.getProperties().putAll(self.properties)
# now create the indexers
if self.type is IndexingType.SINGLEPASS:
if self.blocks:
Indexer = pt.terrier.J.BlockSinglePassIndexer
Merger = pt.terrier.J.BlockStructureMerger
else:
Indexer = pt.terrier.J.BasicSinglePassIndexer
Merger = pt.terrier.J.StructureMerger
elif self.type is IndexingType.CLASSIC:
if self.blocks:
Indexer = pt.terrier.J.BlockIndexer
Merger = pt.terrier.J.BlockStructureMerger
else:
Indexer = pt.terrier.J.BasicIndexer
Merger = pt.terrier.J.StructureMerger
elif self.type is IndexingType.MEMORY:
if self.blocks:
raise Exception("Memory indexing with positions not yet implemented")
else:
Indexer = pt.terrier.J.BasicMemoryIndexer
Merger = None
else:
raise Exception("Unknown indexer type")
assert Indexer is not None
return Indexer, Merger
def getIndexStats(self):
"""
Prints the index statistics
Note:
Does not work with notebooks at the moment
"""
pt.terrier.J.CLITool.main(["indexstats", "-I" + self.path])
def getIndexUtil(self, util):
"""
Utilities for displaying the content of an index
Note:
Does not work with notebooks at the moment
Args:
util: which util to print
Possible Utils:
util(string): possible values:
printbitentry
printlex
printlist
printlistentry
printmeta
printposting
printpostingfile
printterm
"""
if not util.startswith("-"):
util = "-" + util
pt.terrier.J.CLITool.main(["indexutil", "-I" + self.path, util])
class DFIndexUtils:
@staticmethod
def get_column_lengths(df):
meta2len = dict([(v, df[v].apply(lambda r: len(str(r)) if r is not None else 0).max())for v in df.columns.values])
# nan values can arise if df is empty. Here we take a metalength of 1 instead.
meta2len = {k : 1 if math.isnan(le) else le for k, le in meta2len.items()}
return meta2len
@staticmethod
@pt.java.required
def create_javaDocIterator(text, *args, **kwargs):
HashMap = pt.java.J.HashMap
TaggedDocument = pt.terrier.J.TaggedDocument
StringReader = pt.java.J.StringReader
tokeniser = pt.terrier.J.Tokeniser.getTokeniser()
all_metadata = {}
for i, arg in enumerate(args):
if isinstance(arg, pd.Series):
all_metadata[arg.name] = arg
assert len(arg) == len(text), "Length of metadata arguments needs to be equal to length of text argument"
elif isinstance(arg, pd.DataFrame):
for name, column in arg.items():
all_metadata[name] = column
assert len(column) == len(text), "Length of metadata arguments needs to be equal to length of text argument"
else:
raise ValueError(f"Non-keyword args need to be of type pandas.Series or pandas.DataFrame, argument {i} was {type(arg)}")
for key, value in kwargs.items():
if isinstance(value, (pd.Series, list, tuple)):
all_metadata[key] = value
assert len(value) == len(value), "Length of metadata arguments needs to be equal to length of text argument"
elif isinstance(value, pd.DataFrame):
for name, column in arg.items():
all_metadata[name] = column
assert len(column) == len(text), "Length of metadata arguments needs to be equal to length of text argument"
else:
raise ValueError("Keyword kwargs need to be of type pandas.Series, list or tuple")
if "docno" not in all_metadata:
raise ValueError('No docno column specified, while PyTerrier assumes a docno should exist. Found meta columns were %s' % str(list(all_metadata.keys())))
# this method creates the documents as and when needed.
def convertDoc(text_row, meta_column):
if text_row is None:
text_row = ""
hashmap = HashMap()
for column, value in meta_column[1].items():
if value is None:
value = ""
hashmap.put(column, value)
return TaggedDocument(StringReader(text_row), hashmap, tokeniser)
df = pd.DataFrame.from_dict(all_metadata, orient="columns")
lengths = DFIndexUtils.get_column_lengths(df)
return (
PythonListIterator(
text.values,
df.iterrows(),
convertDoc,
len(text.values),
),
lengths)
@deprecated(version='0.11.0', reason="use pt.terrier.IterDictIndexer().index(dataframe.to_dict(orient='records')) instead")
class DFIndexer(TerrierIndexer):
"""
Use this Indexer if you wish to index a pandas.Dataframe
"""
@pt.java.required
def index(self, text, *args, **kwargs):
"""
Index the specified
Args:
text(pd.Series): A pandas.Series(a column) where each row is the body of text for each document
*args: Either a pandas.Dataframe or pandas.Series.
If a Dataframe: All columns(including text) will be passed as metadata
If a Series: The Series name will be the name of the metadata field and the body will be the metadata content
**kwargs: Either a list, a tuple or a pandas.Series
The name of the keyword argument will be the name of the metadata field and the keyword argument contents will be the metadata content
"""
self.checkIndexExists()
# we need to prevent collectionIterator from being GCd, so assign to a variable that outlives the indexer
collectionIterator, meta_lengths = DFIndexUtils.create_javaDocIterator(text, *args, **kwargs)
# record the metadata key names and the length of the values
self.meta = meta_lengths
#Â make a Collection class for Terrier
javaDocCollection = pt.terrier.J.CollectionFromDocumentIterator(collectionIterator)
if self.verbose:
javaDocCollection = TQDMSizeCollection(javaDocCollection, len(text))
index = self.createIndexer()
index.index(pt.terrier.J.PTUtils.makeCollection(javaDocCollection))
global lastdoc
lastdoc = None
javaDocCollection.close()
self.index_called = True
collectionIterator = None
indexref = None
if self.type is IndexingType.MEMORY:
index = index.getIndex()
indexref = index.getIndexRef()
else:
indexref = pt.terrier.J.IndexRef.of(self.index_dir + "/data.properties")
if len(self.cleanup_hooks) > 0:
sindex = pt.terrier.J.Index
sindex.setIndexLoadingProfileAsRetrieval(False)
index = pt.terrier.IndexFactory.of(indexref)
for hook in self.cleanup_hooks:
hook(self, index)
sindex.setIndexLoadingProfileAsRetrieval(True)
return indexref
class _BaseIterDictIndexer(TerrierIndexer, pt.Indexer):
def __init__(self,
index_path: str,
*,
meta : Dict[str,int] = {'docno' : 20},
text_attrs : List[str] = ["text"],
meta_reverse : List[str] = ['docno'],
pretokenised : bool = False,
fields : bool = False,
threads : int = 1,
**kwargs):
"""
:param index_path: Directory to store index. Ignored for IndexingType.MEMORY.
:param meta: What metadata for each document to record in the index, and what length to reserve. Metadata values will be truncated to this length. Defaults to `{"docno" : 20}`.
:param text_attrs: List of columns of the input data that should be indexed. These are concatenated in the document representation. Defaults to `["text"]`.
:param meta_reverse: What metadata should we be able to resolve back to a docid. Defaults to `["docno"]`.
:param pretokenised: Whether to index pre-tokenized text, e.g., through a Learned Sparse encoder. If True, will ignore ``text_attrs`` and indstead index the dictionary contained in the ``toks`` column.
:param fields: Whether a fields-indexer should be used, i.e. whether the frequency in each attribute should be recorded separately in the Terrer index. This allows application of weighting models such as BM25F.
:param threads: Number of threads to use for indexing. Defaults to 1.
:param kwargs: Additional keyword arguments passed to TerrierIndexer.
"""
pt.Indexer.__init__(self)
TerrierIndexer.__init__(self, index_path, **kwargs)
assert pt.terrier.check_version("5.11"), "Terrier 5.11 is required"
self.threads = threads
self.text_attrs = text_attrs
self.meta = meta
self.meta_reverse = meta_reverse
self.pretokenised = pretokenised
self.fields = fields
if self.pretokenised:
if self.fields:
raise ValueError("pretokenised not supported for fields")
if self.text_attrs != ['text']: # user provided text_attrs, which is ignored when pretokenised=True
raise ValueError("pretokenised ignores text_attrs")
# we disable stemming and stopwords for pretokenised indices
self.stemmer = None
self.stopwords = None
self.text_attrs = ['toks'] # pretokenized always uses toks column
def _setup(self):
"""
Configure the indexing properties based on the current indexer configuration.
"""
self.checkIndexExists()
if not isinstance(self.meta, dict):
# user did not specify the lengths of the metadata columns, so set them all to a default value of 512.
# the ramifications of setting all lengths to a large value is an overhead in memory usage during decompression
# also increased reverse lookup file if reverse meta lookups are enabled.
self.meta = {k: '512' for k in self.meta}
if self.fields:
self.setProperties(**{
'metaindex.compressed.crop.long' : 'true',
'FlatJSONDocument.process' : ','.join(self.text_attrs), # index all these json columns
'FieldTags.process': ','.join(self.text_attrs), # each of them will be a field for the indexer
'FieldTags.casesensitive': 'true',
})
else:
self.setProperties(**{
'metaindex.compressed.crop.long' : 'true',
'FlatJSONDocument.process' : ','.join(self.text_attrs), # index all these json columns
'FieldTags.process': '', # but dont make them into fields
'FieldTags.casesensitive': 'true',
})
def _filter_iterable(self, it):
# Only include necessary columns: those that are indexed, metadata columns, and docno
# Also, check that the provided iterator is a suitable format
all_cols = {'docno'} | set(self.text_attrs) | set(self.meta.keys())
first_docs, it = more_itertools.spy(it) # peek at the first document and validate it
if len(first_docs) > 0: # handle empty input
self._validate_doc_dict(first_docs[0])
# important: return an iterator (i.e. using a generator expression) here, rather than make this
# function a generator, to be sure that the validation above happens when _filter_iterable is
# called, rather than on the first invocation of next()
return ({f: doc[f] for f in all_cols} for doc in it)
def _is_dict(self, obj):
return hasattr(obj, '__getitem__') and hasattr(obj, 'items')
def _validate_doc_dict(self, obj):
"""
Raise errors/warnings for common indexing mistakes
"""
if not self._is_dict(obj):
raise ValueError("Was passed %s while expected dict-like object" % (str(type(obj))))
if self.meta is not None:
for k in self.meta:
if k not in obj:
raise ValueError(f"Indexing meta key {k} not found in first document (keys {list(obj.keys())})")
if len(obj[k]) > int(self.meta[k]):
msg = f"Indexing meta key {k} length requested {self.meta[k]} but exceeded in first document (actual length {len(obj[k])}). " + \
f"Increase the length in the meta dict for the indexer, e.g., pt.IterDictIndexer(..., meta={ {k: len(obj[k])} })."
if k == 'docno':
# docnos that are truncated will cause major issues; raise an error
raise ValueError(msg)
else:
# Other fields may not matter as much; just show a warning
warn(msg)
def index_inputs(self) -> List[List[str]]:
return [list(set(list(self.meta.keys()) + self.text_attrs ))]
class _IterDictIndexer_nofifo(_BaseIterDictIndexer):
"""
Use this Indexer if you wish to index an iter of dicts (possibly with multiple fields).
This version is used for Windows -- which doesn't support the faster fifo implementation.
"""
@pt.java.required
def index(self, it, fields=None):
"""
Index the specified iter of dicts with the (optional) specified fields
:param it: an iter of document dicts to be indexed
"""
if fields is not None:
if len(fields) > 1:
raise ValueError("Use fields and text_attrs constructor kwargs")
raise ValueError("Specify the text attribute to index in the constructor.")
self._setup()
assert self.threads == 1, 'IterDictIndexer does not support multiple threads on Windows'
indexer = self.createIndexer()
if self.pretokenised:
assert not self.blocks, "pretokenised isnt compatible with blocks"
# we generate DocumentPostingList from a dictionary of pretokenised text, e.g.
# [
# {'docno' : 'd1', 'toks' : {'a' : 1, 'aa' : 2}}
# ]
iter_docs = DocListIterator(self._filter_iterable(it))
self.index_called = True
indexer.indexDocuments(iter_docs)
iter_docs = None
else:
# we need to prevent collectionIterator from being GCd
collectionIterator = FlatJSONDocumentIterator(self._filter_iterable(it))
javaDocCollection = pt.terrier.J.CollectionFromDocumentIterator(collectionIterator)
indexer.index(javaDocCollection)
global lastdoc
lastdoc = None
self.index_called = True
collectionIterator = None
indexref = None
if self.type is IndexingType.MEMORY:
index = indexer.getIndex()
indexref = index.getIndexRef()
else:
indexref = pt.terrier.J.IndexRef.of(self.index_dir + "/data.properties")
if len(self.cleanup_hooks) > 0:
sindex = pt.terrier.J.Index
sindex.setIndexLoadingProfileAsRetrieval(False)
index = pt.terrier.IndexFactory.of(indexref)
for hook in self.cleanup_hooks:
hook(self, index)
sindex.setIndexLoadingProfileAsRetrieval(True)
return indexref
class _IterDictIndexer_fifo(_BaseIterDictIndexer):
"""
Use this Indexer if you wish to index an iter of dicts (possibly with multiple fields).
This version is optimized by using multiple threads and POSIX fifos to transfer data,
which ends up being much faster.
"""
@pt.java.required
def index(self, it, fields=None):
"""
Index the specified iter of dicts with the (optional) specified fields
:param it: an iter of document dicts to be indexed
"""
CollectionFromDocumentIterator = pt.terrier.J.CollectionFromDocumentIterator
JsonlDocumentIterator = pt.terrier.J.JsonlDocumentIterator
if self.pretokenised:
JsonlTokenisedIterator = pt.terrier.J.JsonlPretokenisedIterator
ParallelIndexer = pt.terrier.J.ParallelIndexer
if fields is not None:
if len(fields) > 1:
raise ValueError("Use fields and text_attrs constructor kwargs")
raise ValueError("Specify the text attribute to index in the constructor.")
self._setup()
os.makedirs(self.index_dir, exist_ok=True) # ParallelIndexer expects the directory to exist
Indexer, Merger = self.indexerAndMergerClasses()
assert self.threads > 0, "threads must be positive"
if Indexer is pt.terrier.J.BasicMemoryIndexer:
assert self.threads == 1, 'IterDictIndexer does not support multiple threads for IndexingType.MEMORY'
if self.threads > 1:
warn(
'Using multiple threads results in a non-deterministic ordering of document in the index. For deterministic behavior, use threads=1')
# Document iterator
fifos = []
j_collections = []
with tempfile.TemporaryDirectory() as d:
# Make a POSIX FIFO with associated java collection for each thread to use
for i in range(self.threads):
fifo = f'{d}/docs-{i}.jsonl'
os.mkfifo(fifo)
if (self.pretokenised):
j_collections.append(JsonlTokenisedIterator(fifo))
else:
j_collections.append(CollectionFromDocumentIterator(JsonlDocumentIterator(fifo)))
fifos.append(fifo)
# Start dishing out the docs to the fifos
threading.Thread(target=self._write_fifos, args=(self._filter_iterable(it), fifos), daemon=True).start()
# Different process for memory indexer (still taking advantage of faster fifos)
if Indexer is pt.terrier.J.BasicMemoryIndexer:
indexer = Indexer()
if self.pretokenised:
indexer.indexDocuments(j_collections)
else:
indexer.index(j_collections)
for hook in self.cleanup_hooks:
hook(self, indexer.getIndex())
return indexer.getIndex().getIndexRef()
# Start the indexing threads
if self.pretokenised:
ParallelIndexer.buildParallelTokenised(j_collections, self.index_dir, Indexer, Merger)
else:
ParallelIndexer.buildParallel(j_collections, self.index_dir, Indexer, Merger)
indexref = None
indexref = pt.terrier.J.IndexRef.of(self.index_dir + "/data.properties")
if len(self.cleanup_hooks) > 0:
sindex = pt.terrier.J.Index
sindex.setIndexLoadingProfileAsRetrieval(False)
index = pt.terrier.IndexFactory.of(indexref)
sindex.setIndexLoadingProfileAsRetrieval(True)
for hook in self.cleanup_hooks:
hook(self, index)
return indexref
def _write_fifos(self, it, fifos):
with contextlib.ExitStack() as stack:
fifos = [stack.enter_context(open(f, 'wt')) for f in fifos]
ready = None
for doc in it:
if not ready: # either first iteration or deque is empty
if len(fifos) > 1:
# Not all the fifos may be ready yet for the next document. Rather than
# waiting for the next one to finish up, go ahead and can check wich are ready
# with the select syscall. This will block until at least one is ready. This
# optimization can actually have a pretty big impact-- on CORD19, indexing
# with 8 threads was 30% faster with this.
_, ready, _ = select.select([], fifos, [])
ready = deque(ready)
else:
# single threaded mode
ready = deque(fifos)
fifo = ready.popleft()
if 'toks' in doc:
doc['toks'] = {k: int(v) for k, v in doc['toks'].items()} # cast all values as ints
json.dump(doc, fifo)
fifo.write('\n')
# Windows doesn't support fifos -- so we have 2 versions.
# Choose which one to expose based on whether os.mkfifo exists.
IterDictIndexer: Type[Union[_IterDictIndexer_fifo, _IterDictIndexer_nofifo]]
if hasattr(os, 'mkfifo'):
#IterDictIndexer = _IterDictIndexer_nofifo
IterDictIndexer = _IterDictIndexer_fifo
else:
IterDictIndexer = _IterDictIndexer_nofifo
# trick sphinx into not using "alias of"
IterDictIndexer.__name__ = 'IterDictIndexer'
[docs]
class TRECCollectionIndexer(TerrierIndexer):
"""
Use this Indexer if you wish to index a TREC formatted collection
"""
def __init__(self,
index_path : str,
collection : str = "trec",
verbose : bool = False,
meta : Dict[str,int] = {"docno" : 20},
meta_reverse : List[str] = ["docno"],
meta_tags : Dict[str,str] = {},
**kwargs
):
"""
Init method
:param index_path: Directory to store index. Ignored for IndexingType.MEMORY.
:param blocks: Create indexer with blocks if true, else without blocks. Default is False.
:param overwrite: If index already present at `index_path`, True would overwrite it, False throws an Exception. Default is False.
:param type: the specific indexing procedure to use. Default is IndexingType.CLASSIC.
:param collection: name, or Class instance, or one of "trec", "trecweb", "warc"). Default is "trec".
:param meta: What metadata for each document to record in the index, and what length to reserve. Metadata fields will be truncated to this length. Defaults to `{"docno" : 20}`.
:param meta_reverse: What metadata shoudl we be able to resolve back to a docid. Defaults to `["docno"]`.
:param meta_tags: For collections formed using tagged data (e.g. HTML), which tags correspond to which metadata. This is useful for recording the text of documents for use in neural rankers - see :ref:`pt.text`.
"""
super().__init__(index_path, **kwargs)
if isinstance(collection, str):
if collection in type_to_class:
collection = type_to_class[collection]
self.collection = collection
self.verbose = verbose
self.meta = meta
self.meta_reverse = meta_reverse
self.meta_tags = meta_tags
[docs]
@pt.java.required
def index(self, files_path : Union[str,List[str]]):
"""
Index the specified TREC formatted files
Args:
files_path: can be a String of the path or a list of Strings of the paths for multiple files
"""
self.checkIndexExists()
index = self.createIndexer()
if not isinstance(files_path, list):
raise ValueError('files_path must be a list')
_TaggedDocumentSetup(self.meta, self.meta_tags)
colObj = createCollection(files_path, self.collection)
if self.verbose and isinstance(colObj, pt.terrier.J.MultiDocumentFileCollection):
colObj = pt.java.cast("org.terrier.indexing.MultiDocumentFileCollection", colObj)
colObj = TQDMCollection(colObj)
# remove once 5.7 is now the minimum version
if pt.terrier.check_version("5.7"):
index.index(colObj)
else:
index.index(pt.terrier.J.PTUtils.makeCollection(colObj))
global lastdoc
lastdoc = None
colObj.close()
self.index_called = True
for hook in self.cleanup_hooks:
hook(self, index.getIndex())
if self.type is IndexingType.MEMORY:
return index.getIndex().getIndexRef()
return pt.terrier.J.IndexRef.of(self.index_dir + "/data.properties")
[docs]
class FilesIndexer(TerrierIndexer):
'''
Use this Indexer if you wish to index a pdf, docx, txt etc files
Args:
index_path (str): Directory to store index. Ignored for IndexingType.MEMORY.
blocks (bool): Create indexer with blocks if true, else without blocks. Default is False.
type (IndexingType): the specific indexing procedure to use. Default is IndexingType.CLASSIC.
meta(Dict[str,int]): What metadata for each document to record in the index, and what length to reserve. Metadata fields will be truncated to this length. Defaults to `{"docno" : 20, "filename" : 512}`.
meta_reverse(List[str]): What metadata shoudl we be able to resolve back to a docid. Defaults to `["docno"]`,
meta_tags(Dict[str,str]): For collections formed using tagged data (e.g. HTML), which tags correspond to which metadata. Defaults to empty. This is useful for recording the text of documents for use in neural rankers - see :ref:`pt.text`.
'''
def __init__(self, index_path, *, meta={"docno" : 20, "filename" : 512}, meta_reverse=["docno"], meta_tags={}, **kwargs):
super().__init__(index_path, **kwargs)
self.meta = meta
self.meta_reverse = meta_reverse
self.meta_tags = meta_tags
[docs]
@pt.java.required
def index(self, files_path : Union[str,List[str]]):
"""
Index the specified files.
:param files_path: can be a String of the path or a list of Strings of the paths for multiple files
"""
self.checkIndexExists()
index = self.createIndexer()
asList = createAsList(files_path)
_TaggedDocumentSetup(self.meta, self.meta_tags)
_FileDocumentSetup(self.meta, self.meta_tags)
simpleColl = pt.terrier.J.SimpleFileCollection(asList, False)
# remove once 5.7 is now the minimum version
index.index(simpleColl if pt.terrier.check_version("5.7") else [simpleColl])
global lastdoc
lastdoc = None
self.index_called = True
indexref = None
if self.type is IndexingType.MEMORY:
index = index.getIndex()
indexref = index.getIndexRef()
else:
indexref = pt.terrier.J.IndexRef.of(self.index_dir + "/data.properties")
if len(self.cleanup_hooks) > 0:
sindex = pt.terrier.J.Index
sindex.setIndexLoadingProfileAsRetrieval(False)
index = pt.terrier.IndexFactory.of(indexref)
for hook in self.cleanup_hooks:
hook(self, index)
sindex.setIndexLoadingProfileAsRetrieval(True)
return indexref