def buildExamples(exampleBuilder, sentences, options):
print >> sys.stderr, "Defining predicted value range:",
sentenceElements = []
for sentence in sentences:
sentenceElements.append(sentence[0].sentenceElement)
exampleBuilder.definePredictedValueRange(sentenceElements, "entity")
print >> sys.stderr, exampleBuilder.getPredictedValueRange()
examples = []
if hasattr(exampleBuilder,
"styles") and "graph_kernel" in exampleBuilder.styles:
counter = ProgressCounter(len(sentences), "Build examples", 0)
else:
counter = ProgressCounter(len(sentences), "Build examples")
for sentence in sentences:
counter.update(
1, "Building examples (" + sentence[0].getSentenceId() + "): ")
sentence[1] = exampleBuilder.buildExamples(sentence[0])
examples.extend(sentence[1])
print >> sys.stderr, "Examples built:", len(examples)
print >> sys.stderr, "Features:", len(exampleBuilder.featureSet.getNames())
print >> sys.stderr, "Preprocessing examples:"
examples = exampleBuilder.preProcessExamples(examples)
# Save examples
# if options.output != None:
# print >> sys.stderr, "Saving examples to", options.output + "/examples.txt"
# commentLines = []
# commentLines.append("Input file: " + options.input)
# commentLines.append("Example builder: " + options.exampleBuilder)
# commentLines.append("Features:")
# commentLines.extend(exampleBuilder.featureSet.toStrings())
# Example.writeExamples(examples, options.output + "/examples.txt", commentLines)
#examples = filterFeatures(exampleBuilder.featureSet, examples)
#Example.normalizeFeatureVectors(examples)
return examples
def processCorpus(input, attrs=["text"]):
print attrs
print >> sys.stderr, "Loading corpus file", input
corpusRoot = ETUtils.ETFromObj(input).getroot()
documents = corpusRoot.findall("document")
counter = ProgressCounter(len(documents), "Documents")
countsByType = {}
interactors = {}
for document in documents:
entDict = {}
for entity in document.getiterator("entity"):
entDict[entity.get("id")] = entity
for interaction in document.getiterator("interaction"):
e1 = entDict[interaction.get("e1")]
e2 = entDict[interaction.get("e2")]
# form identifier tuples
e1Tuple = []
for attr in attrs:
e1Tuple.append(e1.get(attr))
e1Tuple = tuple(e1Tuple)
e2Tuple = []
for attr in attrs:
e2Tuple.append(e2.get(attr))
e2Tuple = tuple(e2Tuple)
interactors = [e1Tuple, e2Tuple]
#interactors.sort()
print interactors
def prepareDocuments(self,
corpusRoot,
files,
conllFormat=None,
counts=None):
print >> sys.stderr, "Generating document elements from the parses"
docNames = sorted(files.keys())
corpusName = corpusRoot.get("source", "CORPUS")
#parseExtensions = set(["ptb", "conll", "conllx", "conllu"])
counter = ProgressCounter(len(docNames), "Document Generation")
for i in range(len(docNames)):
docName = docNames[i]
counter.update(
1, "Making document element for document '" + str(docName) +
"': ")
#filePaths = files[docName]
extensions = sorted(files[docName].keys())
sentObjs = self.readParse(extensions[0],
files[docName][extensions[0]],
conllFormat)
sentTexts = []
for sentObj in sentObjs:
if "tokens" in sentObj:
sentTexts.append(" ".join(
[x["text"] for x in sentObj["tokens"]]))
docText = " ".join(sentTexts)
ET.SubElement(corpusRoot,
"document",
id=corpusName + ".d" + str(i),
origId=docName,
text=docText)
return [x for x in corpusRoot.findall("document")]
def processCorpus(inputFilename, outputFilename, rules):
print >> sys.stderr, "Deleting elements, rules =", rules
print >> sys.stderr, "Loading corpus file", inputFilename
corpusTree = ETUtils.ETFromObj(inputFilename)
corpusRoot = corpusTree.getroot()
for eType in rules.keys():
for attrRule in rules[eType].keys():
if type(rules[eType][attrRule]) in types.StringTypes:
rules[eType][attrRule] = rules[eType][attrRule].split("|")
documents = corpusRoot.findall("document")
counter = ProgressCounter(len(documents), "Documents")
countsByType = defaultdict(int)
for document in documents:
counter.update()
for sentence in document.findall("sentence"):
processSentence(sentence, rules, countsByType)
print >> sys.stderr, "Deleted elements"
for k in sorted(countsByType.keys()):
print >> sys.stderr, " " + k + ":", countsByType[k]
if outputFilename != None:
print >> sys.stderr, "Writing output to", outputFilename
ETUtils.write(corpusRoot, outputFilename)
return corpusTree
def fixAltOffsets(input, output=None):
print >> sys.stderr, "Loading corpus", input
corpusTree = ETUtils.ETFromObj(input)
print >> sys.stderr, "Corpus file loaded"
corpusRoot = corpusTree.getroot()
docCount = 0
sentencesCreated = 0
sentences = [x for x in corpusRoot.getiterator("sentence")]
counter = ProgressCounter(len(sentences), "FixAltOffsets")
fixCount = 0
# fix spans
for sentence in sentences:
counter.update(1, "Fixing AltOffsets for sentence ("+sentence.get("id")+"): ")
sentOffset = Range.charOffsetToSingleTuple(sentence.get("charOffset"))
for entity in sentence.findall("entity"):
altOffsetString = entity.get("altOffset")
if altOffsetString == None:
continue
#print altOffsetString
altOffsets = Range.charOffsetToTuples(altOffsetString)
assert len(altOffsets) == 1
for i in range(len(altOffsets)):
altOffset = altOffsets[i]
altOffsets[i] = (altOffset[0] - sentOffset[0], altOffset[1] - sentOffset[0])
entity.set("altOffset", Range.tuplesToCharOffset(altOffsets))
fixCount += 1
print >> sys.stderr, "Fixed", fixCount, "altOffsets"
if output != None:
print >> sys.stderr, "Writing output to", output
ETUtils.write(corpusRoot, output)
return corpusTree
def buildExamplesForDocuments(self,
documentSentences,
output,
idFileTag=None):
examples = []
counter = ProgressCounter(len(documentSentences), "Build examples")
#calculatePredictedRange(self, sentences)
outfile = open(output, "wt")
exampleCount = 0
for document in documentSentences:
counter.update(
1,
"Building examples (" + document[0].sentence.get("id") + "): ")
examples = self.buildExamples(document)
exampleCount += len(examples)
#examples = self.preProcessExamples(examples)
ExampleUtils.appendExamples(examples, outfile)
outfile.close()
print >> sys.stderr, "Examples built:", exampleCount
print >> sys.stderr, "Features:", len(self.featureSet.getNames())
#IF LOCAL
if self.exampleStats.getExampleCount() > 0:
self.exampleStats.printStats()
#ENDIF
# Save Ids
if idFileTag != None:
print >> sys.stderr, "Saving class names to", idFileTag + ".class_names"
self.classSet.write(idFileTag + ".class_names")
print >> sys.stderr, "Saving feature names to", idFileTag + ".feature_names"
self.featureSet.write(idFileTag + ".feature_names")
def processCorpora(EvaluatorClass, fromCorpus, toCorpus, target, classSets,
negativeClassId, entityMatchFunction):
entityExamples = []
entityPredictions = []
interactionExamples = []
interactionPredictions = []
eventExamples = []
eventPredictions = []
falseEntity = defaultdict(lambda: defaultdict(int))
counter = ProgressCounter(len(fromCorpus.sentences), "Corpus Processing")
# Loop through the sentences and collect all predictions
toCorpusSentences = None
if toCorpus != None:
toCorpusSentences = toCorpus.documentSentences
for i in range(len(fromCorpus.documentSentences)):
if len(fromCorpus.documentSentences[i]) > 0:
counter.update(
len(fromCorpus.documentSentences[i]),
fromCorpus.documentSentences[i][0].sentence.get("id").rsplit(
".", 1)[0])
if toCorpusSentences != None:
newEntityExPred, newInteractionExPred, newEventExPred, sentFalseEntity = processDocument(
fromCorpus.documentSentences[i], toCorpusSentences[i], target,
classSets, negativeClassId, entityMatchFunction)
else:
newEntityExPred, newInteractionExPred, newEventExPred, sentFalseEntity = processDocument(
fromCorpus.documentSentences[i], None, target, classSets,
negativeClassId, entityMatchFunction)
entityExamples.extend(newEntityExPred[0])
entityPredictions.extend(newEntityExPred[1])
interactionExamples.extend(newInteractionExPred[0])
interactionPredictions.extend(newInteractionExPred[1])
eventExamples.extend(newEventExPred[0])
eventPredictions.extend(newEventExPred[1])
for k, v in sentFalseEntity.iteritems():
falseEntity[k][0] += v[0]
falseEntity[k][1] += v[1]
# Process the predictions with an evaluator and print the results
evaluator = None
if len(entityPredictions) > 0:
evaluator = EvaluatorClass(entityExamples,
entityPredictions,
classSet=classSets["entity"])
print evaluator.toStringConcise(title="Entities")
if len(interactionPredictions) > 0:
evaluator = EvaluatorClass(interactionExamples,
interactionPredictions,
classSet=classSets["interaction"])
print evaluator.toStringConcise(title="Interactions")
#print "Interactions (fp ent->fp int, fn-ent->fn-int )"
#for key in sorted(falseEntity.keys()):
# print "", key, falseEntity[key][0], "/", falseEntity[key][1]
if len(eventPredictions) > 0:
evaluator = EvaluatorClass(eventExamples,
eventPredictions,
classSet=classSets["entity"])
print evaluator.toStringConcise(title="Events")
return evaluator
def run(cls,
fileIn,
fileOut=None,
tokenization="split-Charniak-Lease",
entityOffsetKey="charOffset",
includeNeg=False,
stem=False):
"""Builds the master gazzeteer.
fileIn: a string (ending with .xml or .xml.gz), an open input stream, an ElementTree or an Element
fileOut: a string or None. If given, the resulting gazzetteer will be written out
tokenization: name of the tokenization to be used
Produces a dictionary with...
"""
print >> sys.stderr, "Building gazetteer"
gztr = {} #key: token value: dictionary (key: className, value count)
root = ETUtils.ETFromObj(fileIn)
if not ET.iselement(root):
assert isinstance(root, ET.ElementTree)
root = root.getroot()
sentences = []
for sNode in root.getiterator("sentence"):
sentences.append(sNode)
counter = ProgressCounter(len(sentences), "Build gazetteer")
for sNode in sentences:
counter.update(
1, "Adding to gazetteer sentence " + sNode.get("id") + ", ")
for tokenizationNode in sNode.getiterator("tokenization"):
if tokenizationNode.get("tokenizer") == tokenization:
break
else:
assert False, "Did not find %s tokenization" % tokenization
tClasses = tokClasses(tokenizationNode, sNode, entityOffsetKey)
assert len(tClasses) == len(tokenizationNode)
for tokIdx, tokNode in enumerate(tokenizationNode):
gsClass = tClasses[tokIdx]
b, e = charOffStr2tuple(tokNode.get("charOffset"))
tokNodeTxt = tokTxt(b, e, sNode, stem).lower()
tokDict = gztr.setdefault(tokNodeTxt, {})
tokDict[gsClass] = tokDict.get(gsClass, 0) + 1
# for multi-part texts, add collapsed and last token versions
if tokNodeTxt.find("-") != -1:
# collapsed
text = tokNodeTxt.replace("-", "")
if text != "":
tokDict = gztr.setdefault(text, {})
tokDict[gsClass] = tokDict.get(gsClass, 0) + 1
# last part
text = tokNodeTxt.rsplit("-", 1)[-1]
if text != "":
tokDict = gztr.setdefault(text, {})
tokDict[gsClass] = tokDict.get(gsClass, 0) + 1
if fileOut:
Gazetteer.saveGztr(gztr, fileOut, includeNeg)
return gztr
def processCorpus(input, outDir, stem=None, tail=".xml", mergedSets=[], saveCombined=False, verbose=False):
newCorpora = {}
print >> sys.stderr, "Loading corpus file", input
corpusRoot = ETUtils.ETFromObj(input).getroot()
documents = corpusRoot.findall("document")
counter = ProgressCounter(len(documents), "Documents")
countsByType = {}
for document in documents:
counter.update()
docSet = document.get("set")
if docSet == None:
if verbose: print >> sys.stderr, "Warning, no set defined for document", document.get("id")
if not countsByType.has_key("No set"):
countsByType["No set"] = 0
countsByType["No set"] += 1
continue
elif not newCorpora.has_key(docSet):
newCorpora[docSet] = ET.Element("corpus")
for k, v in corpusRoot.attrib.iteritems():
newCorpora[docSet].set(k, v)
countsByType[docSet] = 0
newCorpora[docSet].append(document)
countsByType[docSet] += 1
# Make merged sets
for mergedSet in mergedSets:
tag = "-and-".join(sorted(mergedSet))
if not newCorpora.has_key(tag):
newCorpora[tag] = ET.Element("corpus")
for k, v in corpusRoot.attrib.iteritems():
newCorpora[tag].set(k, v)
countsByType[tag] = 0
for componentSet in mergedSet:
for element in newCorpora[componentSet].findall("document"):
newCorpora[tag].append(element)
countsByType[tag] += 1
print >> sys.stderr, "Documents per set"
for k in sorted(countsByType.keys()):
print >> sys.stderr, " " + str(k) + ":", countsByType[k]
if stem == None:
outDir, stem = os.path.dirname(outDir), os.path.basename(outDir)
if not os.path.exists(outDir):
os.makedirs(outDir)
print >> sys.stderr, "Writing output files to directory", outDir
if saveCombined:
print >> sys.stderr, "Saving combined input to", stem + tail
ETUtils.write(corpusRoot, stem + tail)
else:
print >> sys.stderr, "Combined input not saved"
for docSet in sorted(newCorpora.keys()):
outFilename = os.path.join(outDir, stem + "-" + docSet + tail)
print >> sys.stderr, "Writing set", docSet, "to", outFilename
ETUtils.write(newCorpora[docSet], outFilename)
def loadCorpus(corpus, parse, tokenization=None, removeNameInfo=False, removeIntersentenceInteractionsFromCorpusElements=True):
"""
Load an entire corpus through CorpusElements and add SentenceGraph-objects
to its SentenceElements-objects.
"""
import Utils.ElementTreeUtils as ETUtils
import sys
from Utils.ProgressCounter import ProgressCounter
from Utils.InteractionXML.CorpusElements import CorpusElements
# Corpus may be in file or not
if type(corpus) == types.StringType:
print >> sys.stderr, "Loading corpus file", corpus
corpusTree = ETUtils.ETFromObj(corpus)
corpusRoot = corpusTree.getroot()
# Use CorpusElements-class to access xml-tree
corpusElements = CorpusElements(corpusRoot, parse, tokenization, tree=corpusTree, removeNameInfo=removeNameInfo, removeIntersentenceInteractions=removeIntersentenceInteractionsFromCorpusElements)
print >> sys.stderr, str(len(corpusElements.documentsById)) + " documents, " + str(len(corpusElements.sentencesById)) + " sentences"
# Make sentence graphs
duplicateInteractionEdgesRemoved = 0
sentences = []
counter = ProgressCounter(len(corpusElements.sentences), "Make sentence graphs")
counter.showMilliseconds = True
for sentence in corpusElements.sentences[:]:
counter.update(1, "Making sentence graphs ("+sentence.sentence.get("id")+"): ")
# No tokens, no sentence. No also no dependencies = no sentence.
# Let's not remove them though, so that we don't lose sentences from input.
if len(sentence.tokens) == 0 or len(sentence.dependencies) == 0:
#corpusElements.sentences.remove(sentence)
sentence.sentenceGraph = None
continue
for pair in sentence.pairs:
# gif-xml defines two closely related element types, interactions and
# pairs. Pairs are like interactions, but they can also be negative (if
# interaction-attribute == False). Sometimes pair-elements have been
# (incorrectly) used without this attribute. To work around these issues
# we take all pair-elements that define interaction and add them to
# the interaction-element list.
isInteraction = pair.get("interaction")
if isInteraction == "True" or isInteraction == None:
sentence.interactions.append(pair) # add to interaction-elements
if pair.get("type") == None: # type-attribute must be explicitly defined
pair.set("type", "undefined")
# Construct the basic SentenceGraph (only syntactic information)
graph = SentenceGraph(sentence.sentence, sentence.tokens, sentence.dependencies)
# Add semantic information, i.e. the interactions
graph.mapInteractions(sentence.entities, sentence.interactions)
graph.interSentenceInteractions = sentence.interSentenceInteractions
duplicateInteractionEdgesRemoved += graph.duplicateInteractionEdgesRemoved
sentence.sentenceGraph = graph
graph.parseElement = sentence.parseElement
#graph.mapEntityHints()
print >> sys.stderr, "Skipped", duplicateInteractionEdgesRemoved, "duplicate interaction edges in SentenceGraphs"
return corpusElements
def processCorpus(input, output, wordVectorPath, tokenizerName="McCC", max_rank_mem=100000, max_rank=10000000):
print >> sys.stderr, "Making vocabulary"
print >> sys.stderr, "Loading corpus file", input
corpusTree = ETUtils.ETFromObj(input)
corpusRoot = corpusTree.getroot()
vocabulary = {"indices":{}, "vectors":[]}
print >> sys.stderr, "Loading word vectors from", wordVectorPath
print >> sys.stderr, "max_rank_mem", max_rank_mem
print >> sys.stderr, "max_rank", max_rank
max_rank_mem = int(max_rank_mem)
max_rank = int(max_rank)
wv = WV.load(wordVectorPath, max_rank_mem, max_rank)
dimVector = wv.vectors.shape[1]
print >> sys.stderr, "WordVector length", dimVector
#addVector("[out]", wv.w_to_normv("and").tolist(), vocabulary) #addVector("[out]", dimVector * [0.0] + [0.0, 1.0], vocabulary) # Outside sentence range
#addVector("[OoV]", wv.w_to_normv("and").tolist(), vocabulary) #addVector("[OoV]", dimVector * [0.0] + [1.0, 0.0], vocabulary) # Out of vocabulary
addVector("[out]", dimVector * [0.0] + [0.0, 1.0], vocabulary) # Outside sentence range
addVector("[OoV]", dimVector * [0.0] + [1.0, 0.0], vocabulary) # Out of vocabulary
documents = corpusRoot.findall("document")
counter = ProgressCounter(len(documents), "Documents")
counts = defaultdict(int)
for document in documents:
counter.update()
counts["document"] += 1
for sentence in document.findall("sentence"):
counts["sentence"] += 1
tokenization = IXMLUtils.getTokenizationElement(sentence, tokenizerName)
if tokenization != None:
counts["tokenization"] += 1
for token in tokenization.findall("token"):
counts["token"] += 1
text = token.get("text")
if text not in vocabulary["indices"]:
counts["token-unique"] += 1
vector = wv.w_to_normv(token.get("text").lower())
if vector is not None:
counts["vector"] += 1
vector = vector.tolist() + [0.0, 0.0]
addVector(text, vector, vocabulary)
else:
counts["no-vector"] += 1
print >> sys.stderr, "Counts:", dict(counts)
if output != None:
print >> sys.stderr, "Writing vectors to", output + "-vectors.json.gz"
with gzip.open(output + "-vectors.json.gz", "wt") as f:
json.dump(vocabulary, f)
print >> sys.stderr, "Writing indices to", output + "-indices.json.gz"
with gzip.open(output + "-indices.json.gz", "wt") as f:
json.dump({"indices":vocabulary["indices"], "vectors":None}, f)
return vocabulary
def compareDocuments(documentMap, targetFiles, options):
documentIds = sorted(documentMap.keys())
counter = ProgressCounter(len(documentIds))
stats = {}
eventStats = {
"Start Events": 0,
"End Events": 0,
"False Positive Trigger": 0
} #,
#"Cause FN":0,
#"Cause FP":0,
#"Theme FN":0,
#"Theme FP":0}
for docId in documentIds:
counter.update(1, "Processing: ") # document " + str(docId) + ": " )
for fileName in sorted(documentMap[docId]):
extension = fileName.split(".", 1)[-1]
addStat(stats, extension, "source")
if os.path.exists(os.path.join(options.output, fileName)):
addStat(stats, extension, "target")
if extension == "txt" or extension == "a1":
if compareByLine(fileName, options):
addStat(stats, extension, "identical")
else:
addStat(stats, extension, "different")
if options.verbose:
print >> sys.stderr, " ...in comparing", fileName
elif extension == "a2.t1":
if compareA2T1Files(fileName, options, eventStats):
addStat(stats, extension, "identical")
else:
addStat(stats, extension, "different")
if options.verbose:
print >> sys.stderr, " ...in comparing", fileName
print >> sys.stderr, "Files (source, target, identical, different):"
for key in sorted(stats.keys()):
print >> sys.stderr, " " + key + ":" + (10 - len(key)) * " " + "\t",
for value in stats[key]:
print >> sys.stderr, "\t" + str(value),
print >> sys.stderr
print >> sys.stderr, "Event stats:"
for key in sorted(eventStats.keys()):
print >> sys.stderr, " " + key + ": " + str(eventStats[key])
print >> sys.stderr, "Event extraction:"
eventsSource = eventStats["Start Events"]
events0 = 0
if eventStats.has_key("Error Level 0"):
events0 = eventStats["Error Level 0"]
if eventsSource == 0:
percent = 0
else:
percent = (100.0 * events0 / eventsSource)
print >> sys.stderr, " Exact:", events0, "/", eventsSource, "(%.2f" % percent + " %)"
def buildExamples(exampleBuilder, sentences, outfilename):
timer = Timer()
examples = []
if "graph_kernel" in exampleBuilder.styles:
counter = ProgressCounter(len(sentences), "Build examples", 0)
else:
counter = ProgressCounter(len(sentences), "Build examples")
calculatePredictedRange(exampleBuilder, sentences)
outfile = open(outfilename, "wt")
exampleCount = 0
for sentence in sentences:
counter.update(
1, "Building examples (" + sentence[0].getSentenceId() + "): ")
examples = exampleBuilder.buildExamples(sentence[0])
exampleCount += len(examples)
examples = exampleBuilder.preProcessExamples(examples)
Example.appendExamples(examples, outfile)
outfile.close()
print >> sys.stderr, "Examples built:", str(exampleCount)
print >> sys.stderr, "Features:", len(exampleBuilder.featureSet.getNames())
print >> sys.stderr, "Elapsed", timer.toString()
def process(input, output=None, preprocess=True, debug=False):
"""
Run MetaMap.
"""
counter = ProgressCounter(id="MetaMap")
# Create working directory
workdir = tempfile.mkdtemp()
outWriter = None
if output != None:
outWriter = ETUtils.ETWriter(output)
# Loop iteratively over elements
skip = False
for event, element in ETUtils.ETIteratorFromObj(input, ("start", "end")):
if event == "start": # element start message, element may not be fully read yet
if element.tag == "sentence":
sentence = element
counter.update(1, "Processing MetaMap ("+sentence.get("id")+"): ")
# Run metamap for the sentence element
elif element.tag == "metamap": # skip the metamap element to remove the original one
skip = True
if not skip and output != None:
outWriter.begin(element)
elif event == "end": # element is fully read in memory
if not skip and output != None:
outWriter.end(element)
if element.tag == "metamap":
skip = False # write elements again after this one
if preprocess:
element = convert(element, sentence)
outWriter.write(element) # insert the new metamap element into the output stream
if output != None:
print >> sys.stderr, "Writing output to", output
outWriter.close()
ETUtils.encodeNewlines(output)
if debug:
print >> sys.stderr, "Work directory preserved for debugging at", workdir
else:
shutil.rmtree(workdir)
return output
def readARFF(filename):
featureSet = IdSet(1)
classSet = IdSet(0)
f = open(filename,"rt")
inData = False
lines = f.readlines()
counter = ProgressCounter(len(lines),"ARFFLine")
examples = []
for line in lines:
counter.update(string="Processing line " + str(counter.current + 1) + ": ")
line = line.strip()
if len(line) == 0 or line[0] == "%":
continue
elif line[0] == "@":
#print line
category = line.split()[0].lower()
if category == "@attribute":
category, name, type = line.split()
assert(not inData)
if name.lower() == "class":
name = name.lower()
classNames = type[1:-1].split(",")
assert(len(classNames)==2)
classSet.defineId(classNames[0].strip(),1)
classSet.defineId(classNames[1].strip(),-1)
featureSet.getId(name)
elif category.lower() == "@relation":
assert(not inData)
elif category == "@data":
inData = True
else:
assert(inData)
count = 1
features = {}
for column in line.split(","):
if featureSet.getName(count) != "class":
features[count] = float(column)
else:
classId = classSet.getId(column, False)
assert(classId != None)
count += 1
exampleCount = str(len(examples))
exampleId = "BreastCancer.d" + exampleCount + ".s0.x0"
examples.append([exampleId,classId,features,{}])
return examples
def build(cls, input, output, parse, tokenization=None, includeNeg=False):
p = PathGazetteer(includeNeg)
sentences = cls.getSentences(input, parse, tokenization)
counter = ProgressCounter(len(sentences), "Build path gazetteer")
for sentence in sentences:
counter.update(
1, "Building path gazetteer (" + sentence[0].getSentenceId() +
"): ")
p.processSentence(sentence[0])
p.calculateFractions()
f = open(output, "wt")
for key in sorted(p.gazetteer.keys()):
v = p.gazetteer[key]
f.write(key + " " + str(v[0]) + " " + str(v[1]) + " " + str(v[2]) +
" " + str(v[3]) + "\n")
f.close()
def buildExamplesForSentences(self,
sentences,
goldSentences,
output,
idFileTag=None,
append=False):
examples = []
counter = ProgressCounter(len(sentences), "Build examples")
if append:
outfile = open(output, "at")
else:
outfile = open(output, "wt")
exampleCount = 0
for i in range(len(sentences)):
sentence = sentences[i]
goldSentence = [None]
if goldSentences != None:
goldSentence = goldSentences[i]
counter.update(
1, "Building examples (" + sentence[0].getSentenceId() + "): ")
examples = self.buildExamples(sentence[0],
goldSentence[0],
append=append)
exampleCount += len(examples)
examples = self.preProcessExamples(examples)
ExampleUtils.appendExamples(examples, outfile)
outfile.close()
print >> sys.stderr, "Examples built:", exampleCount
print >> sys.stderr, "Features:", len(self.featureSet.getNames())
#IF LOCAL
if self.exampleStats.getExampleCount() > 0:
self.exampleStats.printStats()
#ENDIF
# Save Ids
if idFileTag != None:
print >> sys.stderr, "Saving class names to", idFileTag + ".class_names"
self.classSet.write(idFileTag + ".class_names")
print >> sys.stderr, "Saving feature names to", idFileTag + ".feature_names"
self.featureSet.write(idFileTag + ".feature_names")
def polynomizeExamples(exampleFile, outFile, weightFeatures, idSet):
outFile = open(outFile, "wt")
addCount = 0
f = open(exampleFile)
numExamples = sum([1 for line in f])
f.close()
counter = ProgressCounter(numExamples, "Polynomize examples", step=0)
weightFeatureIds = {}
for weightFeature in weightFeatures:
wId = idSet.getId(weightFeature, False)
if wId == None:
sys.exit("Weight vector feature", weightFeature, "not in id file")
weightFeatureIds[weightFeature] = wId
print "Polynomizing", exampleFile
exampleCache = []
for example in ExampleUtils.readExamples(exampleFile):
counter.update(1, "Processing example (" + example[0] + "): ")
features = example[2]
for i in range(len(weightFeatures) - 1):
wI = weightFeatures[i]
wIid = weightFeatureIds[wI]
if not features.has_key(wIid):
continue
for j in range(i + 1, len(weightFeatures)):
wJ = weightFeatures[j]
wJid = weightFeatureIds[wJ]
if not features.has_key(wJid):
continue
# Make polynomial feature
features[idSet.getId(wI + "_AND_" + wJ)] = 1
addCount += 1
exampleCache.append(example)
if len(exampleCache) > 50:
ExampleUtils.appendExamples(exampleCache, outFile)
exampleCache = []
ExampleUtils.appendExamples(exampleCache, outFile)
outFile.close()
print "Added", addCount, "polynomial features"
def splitMergedElements(inputFilename, outputFilename=None):
print >> sys.stderr, "##### Split elements with merged types #####"
print >> sys.stderr, "Loading corpus", inputFilename
corpusTree = ETUtils.ETFromObj(inputFilename)
corpusRoot = corpusTree.getroot()
documents = corpusRoot.findall("document")
counter = ProgressCounter(len(documents), "Documents")
countsByType = {"entity": [0, 0], "interaction": [0, 0], "pair": [0, 0]}
for document in documents:
counter.update()
for sentence in document.findall("sentence"):
processSentence(sentence, countsByType)
print >> sys.stderr, "Results"
for k in sorted(countsByType.keys()):
print >> sys.stderr, " " + k + ": removed", countsByType[k][
0], "created", countsByType[k][1]
if outputFilename != None:
print >> sys.stderr, "Writing output to", outputFilename
ETUtils.write(corpusRoot, outputFilename)
return corpusTree
def findHeadsSyntactic(corpus, parse, tokenization):
"""
Determine the head token for a named entity or trigger. The head token is the token closest
to the root for the subtree of the dependency parse spanned by the text of the element.
@param entityElement: a semantic node (trigger or named entity)
@type entityElement: cElementTree.Element
@param verbose: Print selected head tokens on screen
@param verbose: boolean
"""
counts = [0, 0]
sentences = [x for x in corpus.getiterator("sentence")]
counter = ProgressCounter(len(sentences), "SYNTAX")
for sentence in sentences:
counter.update()
tokElement = ETUtils.getElementByAttrib(
sentence, "sentenceanalyses/tokenizations/tokenization",
{"tokenizer": tokenization})
parseElement = ETUtils.getElementByAttrib(
sentence, "sentenceanalyses/parses/parse", {"parser": parse})
if tokElement == None or parseElement == None:
print >> sys.stderr, "Warning, sentence", sentence.get(
"id"), "missing parse or tokenization"
tokens = tokElement.findall("token")
tokenHeadScores = getTokenHeadScores(
tokens,
parseElement.findall("dependency"),
sentenceId=sentence.get("id"))
for entity in sentence.findall("entity"):
if entity.get("headOffset") == None:
headToken = getEntityHeadToken(entity, tokens, tokenHeadScores)
# The ElementTree entity-element is modified by setting the headOffset attribute
entity.set("headOffset", headToken.get("charOffset"))
entity.set("headMethod", "Syntax")
entity.set("headString", headToken.get("text"))
counts[0] += 1
return counts
def processCorpus(self,
input,
output,
gold=None,
append=False,
allowNewIds=True,
structureAnalyzer=None):
# Create intermediate paths if needed
if os.path.dirname(output) != "" and not os.path.exists(
os.path.dirname(output)):
os.makedirs(os.path.dirname(output))
# Open output file
openStyle = "wt"
if append:
#print "Appending examples"
openStyle = "at"
if output.endswith(".gz"):
outfile = gzip.open(output, openStyle)
else:
outfile = open(output, openStyle)
# Build examples
self.exampleCount = 0
if type(input) in types.StringTypes: # Entered here - Mu
self.elementCounts = self.getElementCounts(input)
if self.elementCounts["sentences"] > 0: # Entered here, 1448 - Mu
self.progress = ProgressCounter(
self.elementCounts["sentences"], "Build examples")
else:
self.elementCounts = None
self.progress = ProgressCounter(None, "Build examples")
else:
self.elementCounts = None
self.progress = ProgressCounter(None, "Build examples")
# pdb.set_trace()
# This line generates log below:(getSentences function generates the first 2 lines)
# Making sentence graphs (GE09.d149.s5): 100.00 % (0:0:1.113)
# Skipped 381 duplicate interaction edges in SentenceGraphs
# Defining predicted value range: None - Mu
self.calculatePredictedRange(
self.getSentences(input, self.parse, self.tokenization)
) # self.parse: mccc; self.tokenization: None
removeIntersentenceInteractions = True
if "keep_intersentence" in self.styles and self.styles[
"keep_intersentence"]:
print >> sys.stderr, "Keeping intersentence interactions for input corpus"
removeIntersentenceInteractions = False # this is True - Mu
inputIterator = getCorpusIterator(
input,
None,
self.parse,
self.tokenization,
removeIntersentenceInteractions=removeIntersentenceInteractions)
# pdb.set_trace()
#goldIterator = []
if gold != None: # Entered here - Mu
removeGoldIntersentenceInteractions = True
if "keep_intersentence_gold" in self.styles and self.styles[
"keep_intersentence_gold"]:
print >> sys.stderr, "Keeping intersentence interactions for gold corpus"
removeGoldIntersentenceInteractions = False # this is False - Mu
goldIterator = getCorpusIterator(
gold,
None,
self.parse,
self.tokenization,
removeIntersentenceInteractions=
removeGoldIntersentenceInteractions)
for inputSentences, goldSentences in itertools.izip_longest(
inputIterator, goldIterator, fillvalue=None):
assert inputSentences != None
assert goldSentences != None
# pdb.set_trace()
# see the documentation of function processSentence() in this script
# inputSentences[1].sentence is the unmerged version
# inputSentences[1].sentenceGraph is the merged version, meaning that when generating sentenceGraph,
# duplicated intereactions are removed(actually skipped, not added to the graph, but not really removed) - Mu
self.processDocument(inputSentences,
goldSentences,
outfile,
structureAnalyzer=structureAnalyzer)
else:
for inputSentences in inputIterator:
self.processDocument(inputSentences,
None,
outfile,
structureAnalyzer=structureAnalyzer)
outfile.close()
self.progress.endUpdate()
# Show statistics
print >> sys.stderr, "Examples built:", self.exampleCount
print >> sys.stderr, "Features:", len(self.featureSet.getNames())
print >> sys.stderr, "Classes:", len(self.classSet.getNames())
print >> sys.stderr, "Style:", Utils.Parameters.toString(
self.getParameters(self.styles))
if self.exampleStats.getExampleCount() > 0:
self.exampleStats.printStats()
# Save Ids
if allowNewIds:
self.saveIds()
def mainFunc(input,
output=None,
parseName="McCC",
tokenizationName=None,
newParseName=None,
newTokenizationName=None,
logFileName=None,
removeOld=True):
print >> sys.stderr, "Protein Name Splitter"
if logFileName != None:
print >> sys.stderr, "Writing log to", logFileName
logFile = open(logFileName, "wt")
else:
logFile = None
#if input.endswith(".gz"):
# inFile = gzip.GzipFile(input)
#else:
# inFile = open(input)
tree = ETUtils.ETFromObj(input)
if tokenizationName == None:
tokenizationName = parseName
#tree = ElementTree.parse(inFile)
root = tree.getroot()
sentences = [x for x in root.getiterator("sentence")]
counter = ProgressCounter(len(sentences), "Split Protein Names")
counter.showMilliseconds = True
missingTokCount = 0
for sentence in sentences:
sId = sentence.get("id")
counter.update(1, "Splitting names (" + sId + "): ")
tok = getTokenization(tokenizationName,
sentence,
sId,
remove=removeOld)
if tok == None:
missingTokCount += 1
continue
assert tok is not None, "Missing tokenization '%s' in sentence %s!" % (
tokenizationName, sId)
parse = getParse(parseName,
tokenizationName,
sentence,
sId,
remove=removeOld)
assert parse is not None, "Missing parse '%s' in sentence %s!" % (
parseName, sId)
split = splitTokens(tok, sentence, logFile)
# Default names
if removeOld:
if newTokenizationName == None:
newTokenizationName = tok.get("tokenizer")
if newParseName == None:
newParseName = parse.get("parser")
else:
if newTokenizationName == None:
newTokenizationName = "split-" + tok.get("tokenizer")
if newParseName == None:
newParseName = "split-" + parse.get("parser")
# add a new tokenization with the split tokens.
splittok = addTokenization(newTokenizationName, sentence, sId)
addTokensToTree(split, splittok)
for a in tok.attrib:
if splittok.get(a) == None:
splittok.set(a, tok.get(a))
#splittok.set("split-")
# make a mapping from original to split token ids. Store the
# head token when given.
tokenIdMap = {}
for t in split:
if t.head:
head = t.head
# traverse
while head.head is not None:
assert head.head != t, "Cyclic heads"
head = head.head
# should match (nah, punctuation problems)
# assert t.origId not in tokenIdMap or tokenIdMap[t.origId] == head.id, "Head conflict"
tokenIdMap[t.origId] = head.id
else:
# only allow overwrite of existing entry if the current token
# is not punctuation.
if t.origId not in tokenIdMap or not t.isPunct():
tokenIdMap[t.origId] = t.id
# make a copy of the specified parse that refers to the split tokens
# instead of the originals.
newparse = addParse(newParseName, newTokenizationName, sentence, sId)
for a in parse.attrib:
if newparse.get(a) == None:
newparse.set(a, parse.get(a))
newparse.set("ProteinNameSplitter", "True")
splittok.set("ProteinNameSplitter", "True")
depSeqId = 0 #1
for d in parse.getiterator("dependency"):
t1, t2, dType = d.get("t1"), d.get("t2"), d.get("type")
assert t1 in tokenIdMap and t2 in tokenIdMap, "INTERNAL ERROR"
dep = ElementTree.SubElement(newparse, "dependency")
dep.set("t1", tokenIdMap[t1])
dep.set("t2", tokenIdMap[t2])
dep.set("type", dType)
dep.set("id", "sd_%d" % depSeqId)
depSeqId += 1
# Add in new dependencies between the split parts.
for t in [tok for tok in split if tok.head is not None]:
dep = ElementTree.SubElement(newparse, "dependency")
dep.set("t1", t.head.id)
dep.set("t2", t.id)
dep.set("type", t.depType)
dep.set("split", "PNS")
dep.set("id", "spd_%d" % depSeqId)
depSeqId += 1
for phrase in parse.getiterator("phrase"):
newparse.append(phrase)
# debugging
#print >> sys.stderr, "NEW DEP IN", sId
print >> sys.stderr, "Tokenization missing from", missingTokCount, "sentences"
#indent(root)
if logFile != None:
logFile.close()
# debugging
if output != None:
print >> sys.stderr, "Writing output to", output
ETUtils.write(tree, output)
return tree
def processCorpus(self,
input,
output,
gold=None,
append=False,
allowNewIds=True,
structureAnalyzer=None):
# Create intermediate paths if needed
if os.path.dirname(output) != "" and not os.path.exists(
os.path.dirname(output)):
os.makedirs(os.path.dirname(output))
# Open output file
openStyle = "wt"
if append:
#print "Appending examples"
openStyle = "at"
if output.endswith(".gz"):
outfile = gzip.open(output, openStyle)
else:
outfile = open(output, openStyle)
# Build examples
self.exampleCount = 0
if type(input) in types.StringTypes:
self.elementCounts = self.getElementCounts(input)
if self.elementCounts["sentences"] > 0:
self.progress = ProgressCounter(
self.elementCounts["sentences"], "Build examples")
else:
self.elementCounts = None
self.progress = ProgressCounter(None, "Build examples")
else:
self.elementCounts = None
self.progress = ProgressCounter(None, "Build examples")
self.calculatePredictedRange(
self.getSentences(input, self.parse, self.tokenization))
removeIntersentenceInteractions = True
if "keep_intersentence" in self.styles and self.styles[
"keep_intersentence"]:
print >> sys.stderr, "Keeping intersentence interactions for input corpus"
removeIntersentenceInteractions = False
inputIterator = getCorpusIterator(
input,
None,
self.parse,
self.tokenization,
removeIntersentenceInteractions=removeIntersentenceInteractions)
#goldIterator = []
if gold != None:
removeGoldIntersentenceInteractions = True
if "keep_intersentence_gold" in self.styles and self.styles[
"keep_intersentence_gold"]:
print >> sys.stderr, "Keeping intersentence interactions for gold corpus"
removeGoldIntersentenceInteractions = False
goldIterator = getCorpusIterator(
gold,
None,
self.parse,
self.tokenization,
removeIntersentenceInteractions=
removeGoldIntersentenceInteractions)
for inputSentences, goldSentences in itertools.izip_longest(
inputIterator, goldIterator, fillvalue=None):
assert inputSentences != None
assert goldSentences != None
self.processDocument(inputSentences,
goldSentences,
outfile,
structureAnalyzer=structureAnalyzer)
else:
for inputSentences in inputIterator:
self.processDocument(inputSentences,
None,
outfile,
structureAnalyzer=structureAnalyzer)
outfile.close()
self.progress.endUpdate()
# Show statistics
print >> sys.stderr, "Examples built:", self.exampleCount
print >> sys.stderr, "Features:", len(self.featureSet.getNames())
print >> sys.stderr, "Style:", Utils.Parameters.toString(
self.getParameters(self.styles))
if self.exampleStats.getExampleCount() > 0:
self.exampleStats.printStats()
# Save Ids
if allowNewIds:
self.saveIds()
def makeSentences(input,
output=None,
removeText=False,
postProcess=True,
debug=False):
"""
Run GENIA Sentence Splitter
Divide text in the "text" attributes of document and section
elements into sentence elements. These sentence elements are
inserted into their respective parent elements.
"""
global sentenceSplitterDir
print >> sys.stderr, "Loading corpus", input
corpusTree = ETUtils.ETFromObj(input)
print >> sys.stderr, "Corpus file loaded"
corpusRoot = corpusTree.getroot()
print >> sys.stderr, "Running GENIA Sentence Splitter", Settings.GENIA_SENTENCE_SPLITTER_DIR,
if postProcess:
print >> sys.stderr, "(Using post-processing)"
else:
print >> sys.stderr, "(No post-processing)"
docCount = 0
sentencesCreated = 0
redivideCount = 0
emptySentenceCount = 0
sourceElements = [x for x in corpusRoot.getiterator("document")
] + [x for x in corpusRoot.getiterator("section")]
counter = ProgressCounter(len(sourceElements), "GeniaSentenceSplitter")
counter.showMilliseconds = True
# Create working directory
workdir = tempfile.mkdtemp()
for document in sourceElements:
counter.update(1, "Splitting Documents (" + document.get("id") + "): ")
docId = document.get("id")
if docId == None:
docId = "CORPUS.d" + str(docCount)
docTag = "-" + str(docCount)
assert document.find("sentence") == None
text = document.get("text")
if text == None or text.strip() == "":
continue
#print type(text)
# Write text to workfile
#workdir = tempfile.mkdtemp()
workfile = codecs.open(
os.path.join(workdir, "sentence-splitter-input.txt" + docTag),
"wt", "utf-8")
# From http://themoritzfamily.com/python-encodings-and-unicode.html
# "You have to be careful with the codecs module. Whatever you pass to it must be a Unicode
# object otherwise it will try to automatically decode the byte stream as ASCII"
# However, the unicode errors here were simply due to STTools reading unicode ST-format as ASCII,
# thus creating an ASCII interaction XML, which then triggered here the unicode error. So, at this
# point we should be able to safely write(text), as the output file is unicode, and reading with
# the correct coded is taken care of earlier in the pipeline.
workfile.write(text) #.encode("utf-8"))
workfile.close()
# Run sentence splitter
assert os.path.exists(
Settings.GENIA_SENTENCE_SPLITTER_DIR +
"/run_geniass.sh"), Settings.GENIA_SENTENCE_SPLITTER_DIR
args = [
Settings.GENIA_SENTENCE_SPLITTER_DIR + "/run_geniass.sh",
os.path.join(workdir, "sentence-splitter-input.txt" + docTag),
os.path.join(workdir, "sentence-splitter-output.txt" + docTag),
Settings.RUBY_PATH
]
#p = subprocess.call(args)
p = subprocess.Popen(args,
stdout=subprocess.PIPE,
stderr=subprocess.PIPE)
stdout, stderr = p.communicate()
if stdout != "":
print >> sys.stderr, stdout
if stderr != 'Extracting events.roading model file.\nstart classification.\n':
print >> sys.stderr, stderr
#print "stdout<", p.stdout.readlines(), ">"
#print "stderr<", p.stderr.readlines(), ">"
if postProcess:
postProcessorPath = os.path.join(
os.path.dirname(os.path.abspath(__file__)),
"geniass-postproc.pl")
assert os.path.exists(postProcessorPath), postProcessorPath
ppIn = codecs.open(
os.path.join(workdir, "sentence-splitter-output.txt" + docTag),
"rt", "utf-8")
ppOut = codecs.open(
os.path.join(
workdir,
"sentence-splitter-output-postprocessed.txt" + docTag),
"wt", "utf-8")
perlReturnValue = subprocess.call(["perl", postProcessorPath],
stdin=ppIn,
stdout=ppOut)
assert perlReturnValue == 0, perlReturnValue
ppIn.close()
ppOut.close()
# Read split sentences
workfile = codecs.open(
os.path.join(
workdir,
"sentence-splitter-output-postprocessed.txt" + docTag),
"rt", "utf-8")
else:
workfile = codecs.open(
os.path.join(workdir, "sentence-splitter-output.txt" + docTag),
"rt", "utf-8")
start = 0 # sentences are consecutively aligned to the text for charOffsets
sentenceCount = 0
#text = text.replace("\n", " ") # should stop sentence splitter from crashing.
#text = text.replace(" ", " ") # should stop sentence splitter from crashing.
#alignmentText = text.replace("\n", " ").replace("\r", " ")
#docTokens = reWhiteSpace.split(text)
docIndex = 0
sentenceBeginIndex = -1
prevSentence = None
prevEndIndex = None
#emptySentenceCount = 0
prevText = None
for sText in workfile.readlines():
sText = sText.strip() # The text of the sentence
if sText == "":
emptySentenceCount += 1
continue
for i in range(len(sText)):
if sText[i].isspace():
assert sText[i] not in ["\n", "\r"]
continue
while text[docIndex].isspace():
if text[docIndex] in ["\n", "\r"
] and sentenceBeginIndex != -1:
redivideCount += 1
prevSentence = makeSentence(text, sentenceBeginIndex,
docIndex, prevSentence,
prevEndIndex)
prevSentence.set("id",
docId + ".s" + str(sentenceCount))
prevSentence.set("redevided", "True")
sentencesCreated += 1
sentenceCount += 1
prevEndIndex = docIndex - 1
sentenceBeginIndex = -1
document.append(prevSentence)
docIndex += 1
assert sText[i] == text[docIndex], (
text, sText, prevText, sText[i:i + 10],
text[docIndex:docIndex + 10], (i, docIndex),
sentenceBeginIndex) # tokens[i].isspace() == False
if sentenceBeginIndex == -1:
sentenceBeginIndex = docIndex
docIndex += 1
prevText = sText
if sentenceBeginIndex != -1:
prevSentence = makeSentence(text, sentenceBeginIndex, docIndex,
prevSentence, prevEndIndex)
prevSentence.set("id", docId + ".s" + str(sentenceCount))
prevEndIndex = docIndex - 1
sentenceBeginIndex = -1
sentencesCreated += 1
sentenceCount += 1
document.append(prevSentence)
# Add possible tail for last sentence
if prevEndIndex < len(text) - 1 and prevSentence != None:
assert prevSentence.get("tail") == None, prevSentence.get("tail")
prevSentence.set("tail", text[prevEndIndex + 1:])
#if emptySentenceCount > 0:
# print >> sys.stderr, "Warning,", emptySentenceCount, "empty sentences in", document.get("id")
# Remove original text
if removeText:
del document["text"]
# Move elements from document element to sentences
moveElements(document)
docCount += 1
print >> sys.stderr, "Sentence splitting created", sentencesCreated, "sentences"
print >> sys.stderr, "Redivided", redivideCount, "sentences"
if emptySentenceCount > 0:
print >> sys.stderr, "Warning,", emptySentenceCount, "empty sentences"
if debug:
print >> sys.stderr, "Work directory preserved for debugging at", workdir
else:
# Remove work directory
shutil.rmtree(workdir)
if output != None:
print >> sys.stderr, "Writing output to", output
ETUtils.write(corpusRoot, output)
return corpusTree
"--tokenization",
default=None,
dest="tokenization",
help="Tokenization element name")
optparser.add_option("-p",
"--parse",
default=None,
dest="parse",
help="Parse element name")
(options, args) = optparser.parse_args()
print >> sys.stderr, "Loading input file", options.input
corpusElements = SentenceGraph.loadCorpus(options.input, options.parse,
options.tokenization)
counter = ProgressCounter(len(corpusElements.sentences),
"Resolving chains")
tags = ["e1", "e2"]
for sentence in corpusElements.sentences:
counter.update(
1,
"Resolving chains for (" + sentence.sentence.attrib["id"] + "): ")
identityChainDict = {}
tokenHeadScores = sentence.sentenceGraph.getTokenHeadScores()
for interaction in sentence.interactions:
if interaction.attrib["type"] == "identity":
e1 = sentence.entitiesById[interaction.attrib["e1"]]
e2 = sentence.entitiesById[interaction.attrib["e2"]]
t1 = sentence.sentenceGraph.entityHeadTokenByEntity[e1]
t2 = sentence.sentenceGraph.entityHeadTokenByEntity[e2]
if tokenHeadScores[t2] > tokenHeadScores[t1]:
identityChainDict[
def makeSentences(input,
tokenizationPath,
output=None,
removeText=False,
escDict={},
ignoreErrors=False):
"""
Divide text in the "text" attributes of document and section
elements into sentence elements. These sentence elements are
inserted into their respective parent elements.
"""
print >> sys.stderr, "Loading corpus", input
corpusTree = ETUtils.ETFromObj(input)
print >> sys.stderr, "Corpus file loaded"
corpusRoot = corpusTree.getroot()
print >> sys.stderr, "Inserting tokenizations from", tokenizationPath
assert os.path.exists(tokenizationPath)
if tokenizationPath.find(".tar.gz") != -1:
tarFilePath, tokenizationPath = tokenizationPath.split(".tar.gz")
tarFilePath += ".tar.gz"
tarFile = tarfile.open(tarFilePath)
if tokenizationPath[0] == "/":
tokenizationPath = tokenizationPath[1:]
else:
tarFile = None
docCount = 0
docsWithSentences = 0
sentencesCreated = 0
sourceElements = [x for x in corpusRoot.getiterator("document")
] + [x for x in corpusRoot.getiterator("section")]
counter = ProgressCounter(len(sourceElements), "Sentence Splitting")
for document in sourceElements:
docCount += 1
origId = document.get("pmid")
if origId == None:
origId = document.get("origId")
origId = str(origId)
counter.update(
1, "Splitting Documents (" + document.get("id") + "/" + origId +
"): ")
docId = document.get("id")
if docId == None:
docId = "CORPUS.d" + str(docCount)
if document.find("sentence") == None: # no existing sentence split
text = document.get("text")
if text == None or text.strip() == "":
continue
newFile = os.path.join(tokenizationPath, origId + ".tok")
f = openFile(newFile, tarFile)
if f == None: # file with BioNLP'11 extension not found, try BioNLP'09 extension
oldFile = os.path.join(tokenizationPath, origId + ".tokenized")
f = openFile(oldFile, tarFile)
if f == None: # no tokenization found
continue
sentencesCreated += alignSentences(document,
f.readlines(),
escDict,
ignoreErrors=ignoreErrors)
f.close()
# Remove original text
if removeText:
del document["text"]
# Move elements from document element to sentences
moveElements(document)
docsWithSentences += 1
else:
docsWithSentences += 1
if tarFile != None:
tarFile.close()
print >> sys.stderr, "Sentence splitting created", sentencesCreated, "sentences"
print >> sys.stderr, docsWithSentences, "/", docCount, "documents have sentences"
if output != None:
print >> sys.stderr, "Writing output to", output
ETUtils.write(corpusRoot, output)
return corpusTree
def findHeads(input,
parse,
tokenization=None,
output=None,
removeExisting=True,
iterate=False):
if iterate:
from Utils.ProgressCounter import ProgressCounter
import InteractionXML.SentenceElements as SentenceElements
print >> sys.stderr, "Determining head offsets using parse", parse, "and tokenization", tokenization
print >> sys.stderr, "Removing existing head offsets"
removeCount = 0
counter = ProgressCounter(None, "Find heads")
counter.showMilliseconds = True
for sentences in SentenceElements.getCorpusIterator(
input, output, parse, tokenization):
for sentence in sentences:
if removeExisting:
for e in sentence.sentence.findall("entity"):
if e.get("headOffset") != None:
removeCount += 1
del e.attrib["headOffset"]
graph = SentenceGraph.SentenceGraph(sentence.sentence,
sentence.tokens,
sentence.dependencies)
graph.mapInteractions(sentence.entities, sentence.interactions)
# Make sure every parse gets head scores
#if graph.tokenHeadScores == None:
# graph.getTokenHeadScores()
counter.update(
len(sentences),
"Finding heads (" + sentences[-1].sentence.get("id") + "): ")
print >> sys.stderr, "Removed head offsets from", removeCount, "entities"
else:
xml = ETUtils.ETFromObj(input)
if removeExisting:
print >> sys.stderr, "Removing existing head offsets"
removeCount = 0
xml = ETUtils.ETFromObj(input)
for d in xml.getroot().findall("document"):
for s in d.findall("sentence"):
for e in s.findall("entity"):
if e.get("headOffset") != None:
removeCount += 1
del e.attrib["headOffset"]
print >> sys.stderr, "Removed head offsets from", removeCount, "entities"
# SentenceGraph automatically calculates head offsets and adds them to entities if they are missing
print >> sys.stderr, "Determining head offsets using parse", parse, "and tokenization", tokenization
corpusElements = SentenceGraph.loadCorpus(xml, parse, tokenization)
# Make sure every parse gets head scores
for sentence in corpusElements.sentences:
if sentence.sentenceGraph == None:
continue
if sentence.sentenceGraph.tokenHeadScores == None:
sentence.sentenceGraph.getTokenHeadScores()
if output != None:
print >> sys.stderr, "Writing output to", output
ETUtils.write(corpusElements.rootElement, output)
return xml
def insertParses(self,
parseDir,
input,
output=None,
parseName="McCC",
extensions=None,
subDirs=None,
debug=False,
skipParsed=False,
docMatchKeys=None,
conllFormat=None,
splitting=True,
unescapeFormats="AUTO",
tokenMerging=True,
extMap=None,
sdFailedFormat="empty",
origIdType=None,
posTags=None):
corpusTree, corpusRoot = self.getCorpus(input)
if not os.path.exists(parseDir):
raise Exception("Cannot find parse input '" + str(parseDir) + "'")
if not os.path.isdir(parseDir):
raise Exception("Parse input '" + str(parseDir) +
"' is not a directory")
if extensions == None:
extensions = self.allExt
elif isinstance(extensions, basestring):
extensions = extensions.split(",")
extensions = [x for x in extensions if x in self.allExt]
unescapeFormats = self.getUnescapeFormats(unescapeFormats)
if docMatchKeys == None:
docMatchKeys = ["origId", "pmid", "id"]
elif isinstance(docMatchKeys, basestring):
docMatchKeys = docMatchKeys.split(",")
print >> sys.stderr, "Inserting parses from file types:", extensions
counts = defaultdict(int)
files = self.getParseFiles(parseDir,
extensions,
subDirs,
counts,
extMap=extMap,
origIdType=origIdType)
typeCounts = {x: defaultdict(int) for x in extensions}
# Make document elements if needed
documents = [x for x in corpusRoot.findall("document")]
if len(documents) == 0:
typeCounts["document-generation"] = defaultdict(int)
documents = self.prepareDocuments(corpusRoot, files)
counter = ProgressCounter(len(files), "Parse Insertion")
# Insert parses and make sentence elements if needed
typeCounts["sentence-splitting"] = defaultdict(int)
print >> sys.stderr, "Inserting parses for", len(
files), "out of total", len(documents), "documents"
for document in documents:
counts["document"] += 1
matchFound = False
for docMatchValue in [
document.get(x) for x in docMatchKeys
if document.get(x) != None
]:
if docMatchValue in files:
if matchFound:
raise Exception(
"Multiple matching parses for document " +
str(document.attrib) + " using keys " +
str(docMatchKeys))
matchFound = True
counter.update(
1, "Inserting parses for (" + document.get("id") +
"/" + str(docMatchValue) + "): ")
counts["document-match"] += 1
for ext in extensions:
if ext not in files[docMatchValue]:
continue
counts[ext + "-match"] += 1
sentences = [
x for x in self.getSentences(document,
skipParsed=skipParsed)
]
self.insertParse(document,
sentences,
ext,
files[docMatchValue][ext],
parseName,
splitting,
typeCounts,
conllFormat,
unescapeFormats=unescapeFormats,
tokenMerging=tokenMerging,
sdFailedFormat=sdFailedFormat,
posTags=posTags)
if not matchFound:
counts["document-no-match"] += 1
if len(typeCounts["sentence-splitting"]) > 0:
print >> sys.stderr, "Sentence Splitting Counts", dict(
typeCounts["sentence-splitting"])
print >> sys.stderr, "Counts", dict(counts)
for ext in extensions:
if len(typeCounts[ext]) > 0:
print >> sys.stderr, "Counts for type '" + ext + "':", dict(
typeCounts[ext])
# Write the output XML file
if output != None:
print >> sys.stderr, "Writing output to", output
ETUtils.write(corpusRoot, output)
return corpusTree
def combine(inputA,
inputB,
inputGold,
outPath=None,
mode="OR",
skip=None,
logPath="AUTO"):
assert options.mode in ("AND", "OR")
if skip != None and isinstance(skip, basestring):
skip = set(skip.split(","))
if skip != None:
print "Skipping interaction types:", skip
if logPath == "AUTO":
if outPath != None:
logPath = os.path.join(
outPath.rstrip("/").rstrip("\\") + "-log.txt")
else:
logPath = None
if logPath != None:
if not os.path.exists(os.path.dirname(logPath)):
os.makedirs(os.path.dirname(logPath))
Stream.openLog(logPath)
print "Loading the Interaction XML files"
print "Loading A from", inputA
a = ETUtils.ETFromObj(inputA)
print "Loading B from", inputB
b = ETUtils.ETFromObj(inputB)
gold = None
if inputGold:
print "Loading gold from", inputGold
gold = ETUtils.ETFromObj(inputGold) if inputGold else None
print "Copying a as template"
template = copy.deepcopy(a)
print "Calculating confidence score ranges"
scoreRanges = {}
scoreRanges["a"] = getScoreRange(a, skip)
scoreRanges["b"] = getScoreRange(b, skip)
print scoreRanges
print "Combining"
counts = defaultdict(int)
counts["skipped"] = defaultdict(int)
counter = ProgressCounter(len([x for x in a.findall("document")]),
"Combine")
for docA, docB, docGold, docTemplate in itertools.izip_longest(
*[x.findall("document") for x in (a, b, gold, template)]):
counter.update()
assert len(
set([x.get("id")
for x in (docA, docB, docGold, docTemplate)])) == 1
for sentA, sentB, sentGold, sentTemplate in itertools.izip_longest(*[
x.findall("sentence")
for x in (docA, docB, docGold, docTemplate)
]):
assert len(
set([
x.get("id") for x in (sentA, sentB, sentGold, sentTemplate)
])) == 1
interactions = getInteractions(sentA, sentB, sentGold, skip,
counts["skipped"])
for interaction in sentTemplate.findall("interaction"):
sentTemplate.remove(interaction)
analyses = sentTemplate.find("analyses")
if analyses:
sentTemplate.remove(analyses)
for key in interactions:
interaction = getCombinedInteraction(interactions[key], mode,
counts, scoreRanges)
if interaction != None:
sentTemplate.append(copy.deepcopy(interaction))
if analyses:
sentTemplate.append(analyses)
counts["skipped"] = dict(counts["skipped"])
print "Counts:", dict(counts)
if gold != None:
print "****** Evaluating A ******"
evaluateChemProt(
a, gold
) #EvaluateIXML.run(AveragingMultiClassEvaluator, a, gold, "McCC")
print "****** Evaluating B ******"
evaluateChemProt(
b, gold
) #EvaluateIXML.run(AveragingMultiClassEvaluator, b, gold, "McCC")
print "****** Evaluating Combined ******"
evaluateChemProt(
template, gold
) #EvaluateIXML.run(AveragingMultiClassEvaluator, template, gold, "McCC")
if outPath != None:
print "Writing output to", outPath
if outPath.endswith(".tsv"):
Preprocessor(steps=["EXPORT_CHEMPROT"]).process(template, outPath)
else:
ETUtils.write(template, outPath)
if logPath != None:
Stream.closeLog(logPath)
def export(self, input, output, parseName, tokenizerName=None, toExport=["tok", "ptb", "sd"], inputSuffixes=None, clear=False, tokenIdOffset=0, exportIds=None, useSetDirs=False):
print >> sys.stderr, "##### Export Parse #####"
if toExport == None:
toExport = ["txt", "sentences", "tok", "ptb", "sd"]
print >> sys.stderr, "Exporting parse formats", toExport
if os.path.exists(output) and clear:
shutil.rmtree(output)
if not os.path.exists(output):
os.makedirs(output)
if inputSuffixes != None:
inputFileNames = []
for suffix in inputSuffixes:
inputFileNames.append(input + suffix)
else:
inputFileNames = [input]
for inputFileName in inputFileNames:
print >> sys.stderr, "Processing input file", inputFileName
corpusRoot = ETUtils.ETFromObj(inputFileName).getroot()
documents = corpusRoot.findall("document")
counter = ProgressCounter(len(documents), "Documents")
counts = {"corpus":defaultdict(int)}
for fileExt in toExport:
counts[fileExt] = defaultdict(int)
for document in documents:
counter.update()
counts["corpus"]["documents"] += 1
exportId = IXMLUtils.getExportId(document, exportIds)
# Open document output files
outfiles = {}
for fileExt in toExport:
#print output, exportId , fileExt
if useSetDirs:
outfilePath = os.path.join(output, document.get("set"), exportId + "." + fileExt)
else:
outfilePath = os.path.join(output, exportId + "." + fileExt)
if os.path.exists(outfilePath): # check for overlapping files
raise Exception("Export file '" + str(outfilePath) + "' already exists")
if not os.path.exists(os.path.dirname(outfilePath)):
os.makedirs(os.path.dirname(outfilePath))
outfiles[fileExt] = codecs.open(outfilePath, "wt", "utf-8")
# Export document text
if "txt" in outfiles and document.get("text") != None:
outfiles["txt"].write(document.get("text"))
if "txt" not in counts:
counts["txt"] = defaultdict(int)
counts["txt"]["documents"] += 1
# Process all the sentences in the document
sentenceCount = 0
for sentence in document.findall("sentence"):
counts["corpus"]["sentences"] += 1
parse = IXMLUtils.getParseElement(sentence, parseName)
tokenization = IXMLUtils.getTokenizationElement(sentence, tokenizerName)
if "sentences" in outfiles:
outfiles["sentences"].write(sentence.get("text").strip().replace("\n", " ").replace("\r", " ") + "\n")
counts["sentences"]["sentences"] += 1
if "ptb" in outfiles:
if self.exportPennTreeBank(parse, outfiles["ptb"]):
counts["ptb"]["sentences"] += 1
if tokenization != None:
if "tok" in outfiles:
if self.exportTokenization(tokenization, parse, sentence, outfiles["tok"]):
counts["tok"]["sentences"] += 1
if "sd" in outfiles:
if self.exportStanfordDependencies(parse, tokenization, outfiles["sd"], tokenIdOffset):
counts["sd"]["sentences"] += 1
for conllFormat in ("conll", "conllx", "conllu"):
if conllFormat in outfiles:
if self.exportCoNLL(tokenization, parse, outfiles[conllFormat], conllFormat, counts[conllFormat]):
counts[conllFormat]["sentences"] += 1
if "epe" in outfiles:
if self.exportEPE(tokenization, parse, sentence, sentenceCount, outfiles["epe"]):
counts["epe"]["sentences"] += 1
sentenceCount += 1
# Close document output files
for fileExt in outfiles:
outfiles[fileExt].close()
outfiles[fileExt] = None
print >> sys.stderr, "Parse export counts:"
for k in sorted(counts.keys()):
print >> sys.stderr, " " + str(k) + ":", dict(counts[k])
