コード例 #1
0
ファイル: Release.py プロジェクト: DUT-LiuYang/TEES
def getDDI13Result(output, numFolds=10, catenate=False):
    global mainTEESDir
    foldPaths = []
    scores = []
    matrix = defaultdict(lambda:defaultdict(int))
    for fold in range(numFolds):
        foldPath = os.path.join(output, "DDI13-fold" + str(fold), "classification-test", "test-pred.xml.gz")
        foldPaths.append(foldPath)
        
        logPath = os.path.join(output, "DDI13-fold" + str(fold), "log.txt")
        getDDI13ResultLine(logPath, "DDI13-fold" + str(fold), scores)
        
        foldPath = os.path.join(output, "DDI13-fold" + str(fold), "classification-test")
        classPath = os.path.join(output, "DDI13-fold" + str(fold), "model-test", "trigger-ids.classes")
        if not os.path.exists(classPath):
            classPath = os.path.join(output, "DDI13-fold" + str(fold), "model-test", "edge-ids.classes")
            addExamples(os.path.join(foldPath, "test-edge-examples.gz"), os.path.join(foldPath, "test-edge-classifications"), classPath, matrix)
        else:
            addExamples(os.path.join(foldPath, "test-trigger-examples.gz"), os.path.join(foldPath, "test-trigger-classifications"), classPath, matrix)
        
        #parameterPaths = [[":TRAIN:END-MODEL", "Selected parameters"]]
        #print "DDI13-fold" + str(fold) + ": " + getResultLine(logPath, parameterPaths)
    print "-----"
    for testSet in ["DDI13-test9.1", "DDI13-test9.2"]:
        logPath = os.path.join(output, testSet, "log.txt")
        getDDI13ResultLine(logPath, testSet)
        #parameterPaths = [[":TRAIN:END-MODEL", "Selected parameters"]]
        #print testSet + ": " + getResultLine(logPath, parameterPaths)
        
        predPath = os.path.join(output, testSet, "classification-test", "test-pred.xml.gz")
        DDITools.makeDDI13SubmissionFile(predPath, os.path.join(output, testSet + "-interactions.txt"), "interactions")
        DDITools.makeDDI13SubmissionFile(predPath, os.path.join(output, testSet + "-entities.txt"), "entities")
    print "-----"
    print "Avg-score: ", stats.mean(scores), "stdev", stats.stdev(scores)
    
    print "-----"
    print matrixToString(matrix)
    print matrixToString(matrix, True)
    
    if catenate and len(foldPaths) > 1:
        catPath = os.path.join(output, "DDI13-train-analyses.xml.gz")
        Catenate.catenate(foldPaths, catPath, fast=True)
        DDITools.makeDDI13SubmissionFile(catPath, os.path.join(output, "DDI13-train-interactions.txt"), "interactions")
        DDITools.makeDDI13SubmissionFile(catPath, os.path.join(output, "DDI13-train-entities.txt"), "entities")
コード例 #2
0
def getDDI13Result(output, numFolds=10, catenate=False):
    global mainTEESDir
    foldPaths = []
    scores = []
    matrix = defaultdict(lambda:defaultdict(int))
    for fold in range(numFolds):
        foldPath = os.path.join(output, "DDI13-fold" + str(fold), "classification-test", "test-pred.xml.gz")
        foldPaths.append(foldPath)
        
        logPath = os.path.join(output, "DDI13-fold" + str(fold), "log.txt")
        getDDI13ResultLine(logPath, "DDI13-fold" + str(fold), scores)
        
        foldPath = os.path.join(output, "DDI13-fold" + str(fold), "classification-test")
        classPath = os.path.join(output, "DDI13-fold" + str(fold), "model-test", "trigger-ids.classes")
        if not os.path.exists(classPath):
            classPath = os.path.join(output, "DDI13-fold" + str(fold), "model-test", "edge-ids.classes")
            addExamples(os.path.join(foldPath, "test-edge-examples.gz"), os.path.join(foldPath, "test-edge-classifications"), classPath, matrix)
        else:
            addExamples(os.path.join(foldPath, "test-trigger-examples.gz"), os.path.join(foldPath, "test-trigger-classifications"), classPath, matrix)
        
        #parameterPaths = [[":TRAIN:END-MODEL", "Selected parameters"]]
        #print "DDI13-fold" + str(fold) + ": " + getResultLine(logPath, parameterPaths)
    print "-----"
    for testSet in ["DDI13-test9.1", "DDI13-test9.2"]:
        logPath = os.path.join(output, testSet, "log.txt")
        getDDI13ResultLine(logPath, testSet)
        #parameterPaths = [[":TRAIN:END-MODEL", "Selected parameters"]]
        #print testSet + ": " + getResultLine(logPath, parameterPaths)
        
        predPath = os.path.join(output, testSet, "classification-test", "test-pred.xml.gz")
        DDITools.makeDDI13SubmissionFile(predPath, os.path.join(output, testSet + "-interactions.txt"), "interactions")
        DDITools.makeDDI13SubmissionFile(predPath, os.path.join(output, testSet + "-entities.txt"), "entities")
    print "-----"
    print "Avg-score: ", stats.mean(scores), "stdev", stats.stdev(scores)
    
    print "-----"
    print matrixToString(matrix)
    print matrixToString(matrix, True)
    
    if catenate and len(foldPaths) > 1:
        catPath = os.path.join(output, "DDI13-train-analyses.xml.gz")
        Catenate.catenate(foldPaths, catPath, fast=True)
        DDITools.makeDDI13SubmissionFile(catPath, os.path.join(output, "DDI13-train-interactions.txt"), "interactions")
        DDITools.makeDDI13SubmissionFile(catPath, os.path.join(output, "DDI13-train-entities.txt"), "entities")
コード例 #3
0
def getDDI13Result(output, numFolds=10):
    global mainTEESDir
    from batch import batch
    foldPaths = []
    for fold in range(numFolds):
        foldPath = os.path.join(output, "DDI13-fold" + str(fold),
                                "classification-test", "test-pred.xml.gz")
        logPath = os.path.join(output, "DDI13-fold" + str(fold), "log.txt")
        tagPaths = [[
            "------------ Test set classification ------------",
            "##### EvaluateInteractionXML #####", "Interactions", "micro p/n:"
        ]]
        print "DDI13-fold" + str(fold) + ": " + getResultLine(
            logPath, tagPaths)
        foldPaths.append(foldPath)
    if len(foldPaths) > 1:
        Catenate.catenate(foldPaths,
                          os.path.join(output, "DDI13-train-analyses.xml.gz"),
                          fast=True)
コード例 #4
0
ファイル: train.py プロジェクト: ninjin/TEES
def getTaskSettings(task, detector, processUnmerging, processModifiers, isSingleStage,
                    bioNLPSTParams, preprocessorParams, 
                    inputFiles, exampleStyles, classifierParameters):
    if task != None:
        print >> sys.stderr, "Determining training settings for task", task
        assert task.replace("-MINI", "") in ["GE09", "GE09.1", "GE09.2", "GE", "GE.1", "GE.2", "EPI", "ID", "BB", "BI", "BI-FULL", "CO", "REL", "REN", "DDI", "DDI-FULL"], task
    
        fullTaskId = task
        subTask = 2
        if "." in task:
            task, subTask = task.split(".")
            subTask = int(subTask)
        #dataPath = os.path.expanduser("~/biotext/BioNLP2011/data/main-tasks/")
        dataPath = Settings.CORPUS_DIR
        # Optional overrides for input files
        #if inputFiles["devel"] == None: inputFiles["devel"] = dataPath + task + "/" + task + "-devel.xml"
        #if inputFiles["train"] == None: inputFiles["train"] = dataPath + task + "/" + task + "-train.xml"
        #if inputFiles["test"] == None: inputFiles["test"] = dataPath + task + "/" + task + "-test.xml"
        if inputFiles["devel"] == None and inputFiles["devel"] != "None": 
            inputFiles["devel"] = os.path.join(dataPath, task.replace("-FULL", "") + "-devel.xml")
        if inputFiles["train"] == None and inputFiles["train"] != "None":
            if task == "ID": # add GE-task data to the ID training set
                inputFiles["train"] = Catenate.catenate([os.path.join(dataPath, "ID-train.xml"),
                                                         os.path.join(dataPath, "GE-devel.xml"),
                                                         os.path.join(dataPath, "GE-train.xml")], 
                                                        "training/ID-train-and-GE-devel-and-train.xml.gz", fast=True)
            else:
                inputFiles["train"] = os.path.join(dataPath, task.replace("-FULL", "") + "-train.xml")
        if inputFiles["test"] == None and inputFiles["test"] != "None": 
            inputFiles["test"] = os.path.join(dataPath, task.replace("-FULL", "") + "-test.xml")
        
        task = task.replace("-MINI", "")
        # Example generation parameters
        if detector == None:
            detector = "Detectors.EventDetector"
            if task == "CO":
                detector = "Detectors.CODetector"
            elif task in ["REN", "BI", "DDI"]:
                detector = "Detectors.EdgeDetector"
                isSingleStage = True
            print >> sys.stderr, "Detector undefined, using default '" + detector + "' for task", fullTaskId
        if bioNLPSTParams == None and task not in ["DDI", "DDI-FULL"]:
            bioNLPSTParams = "convert:evaluate:scores"
            if task == "BI-FULL":
                bioNLPSTParams = "convert:scores" # the shared task evaluator is not designed for predicted entities
            print >> sys.stderr, "BioNLP Shared Task parameters undefined, using default '" + bioNLPSTParams + "' for task", fullTaskId
        if preprocessorParams == None:
            preprocessorParams = ["intermediateFiles"]
            if task in ["BI", "BI-FULL", "BB", "DDI", "DDI-FULL"]:
                preprocessorParams += ["omitSteps=NER,DIVIDE-SETS"]
            else:
                preprocessorParams += ["omitSteps=DIVIDE-SETS"]
                preprocessorParams += ["PARSE.requireEntities"] # parse only sentences where BANNER found an entity
            preprocessorParams = ":".join(preprocessorParams)
            print >> sys.stderr, "Preprocessor parameters undefined, using default '" + preprocessorParams + "' for task", fullTaskId
        if processUnmerging == None and not isSingleStage:
            processUnmerging = True
            if task in ["CO", "REL", "BB", "BI-FULL", "DDI-FULL"]:
                processUnmerging = False
            print >> sys.stderr, "Unmerging undefined, using default", processUnmerging, "for task", fullTaskId
        if processModifiers == None:
            processModifiers = False
            if task in ["GE", "EPI", "ID"]: 
                processModifiers = True
            print >> sys.stderr, "Modifier prediction undefined, using default", processModifiers, " for task", fullTaskId
        if exampleStyles["examples"] == None and isSingleStage:
            if task == "REN":
                exampleStyles["examples"] = "trigger_features:typed:no_linear:entities:noMasking:maxFeatures:bacteria_renaming:maskTypeAsProtein=Gene"
            elif task == "BI":
                exampleStyles["examples"] = "trigger_features:typed:directed:no_linear:entities:noMasking:maxFeatures:bi_limits"
            elif task == "DDI":
                exampleStyles["examples"] = "trigger_features:typed:no_linear:entities:noMasking:maxFeatures:ddi_features:ddi_mtmx:filter_shortest_path=conj_and"
            print >> sys.stderr, "Single-stage examples style undefined, using default '" + exampleStyles["examples"] + "' for task", fullTaskId
        if exampleStyles["edge"] == None and not isSingleStage:
            print >> sys.stderr, "Edge example style undefined, using default for task", fullTaskId
            if task in ["GE09", "GE"]:
                exampleStyles["edge"]="trigger_features:typed:directed:no_linear:entities:genia_limits:noMasking:maxFeatures" #,multipath"
                if subTask == 1:
                    exampleStyles["edge"] += ":genia_task1"
            elif task in ["BB"]:
                exampleStyles["edge"]="trigger_features:typed:directed:no_linear:entities:bb_limits:noMasking:maxFeatures"
            elif task == "EPI":
                exampleStyles["edge"]="trigger_features:typed:directed:no_linear:entities:epi_limits:noMasking:maxFeatures"
            elif task == "ID":
                exampleStyles["edge"]="trigger_features:typed:directed:no_linear:entities:id_limits:noMasking:maxFeatures"
            elif task == "REL":
                exampleStyles["edge"]="trigger_features:typed:directed:no_linear:entities:noMasking:maxFeatures:rel_limits:rel_features"
            elif task == "CO":
                exampleStyles["edge"]="trigger_features:typed:directed:no_linear:entities:noMasking:maxFeatures:co_limits"
            elif task == "BI-FULL":
                exampleStyles["edge"] = "trigger_features:typed:directed:no_linear:entities:noMasking:maxFeatures:bi_limits"
            elif task == "DDI-FULL":
                exampleStyles["edge"] = "trigger_features:typed:no_linear:entities:noMasking:maxFeatures:ddi_features:filter_shortest_path=conj_and"
            else:
                exampleStyles["edge"]="trigger_features:typed:directed:no_linear:entities:noMasking:maxFeatures"
        if exampleStyles["trigger"] == None and not isSingleStage:
            print >> sys.stderr, "Trigger example style undefined, using default for task", fullTaskId
            if task in ["GE09", "GE"] and subTask == 1:
                exampleStyles["trigger"] = "genia_task1"
            elif task == "EPI":
                exampleStyles["trigger"] = "epi_merge_negated"
            elif task == "BB":
                exampleStyles["trigger"] = "bb_features:build_for_nameless:wordnet"
            elif task == "REL":
                exampleStyles["trigger"] = "rel_features"
            elif task == "CO":
                options.triggerExampleBuilder = "PhraseTriggerExampleBuilder"
            elif task in ["BI-FULL", "DDI-FULL"]:
                exampleStyles["trigger"] = "build_for_nameless:names"
        if exampleStyles["unmerging"] == None and not isSingleStage:
           exampleStyles["unmerging"] = "trigger_features:typed:directed:no_linear:entities:genia_limits:noMasking:maxFeatures"
           #if task == "ID": # Do not use catenated GE for unmerging examples
           #    exampleStyles["unmerging"] += ":sentenceLimit=id.ID"
        # Classifier parameters
        if classifierParameters["examples"] == None and isSingleStage:
            print >> sys.stderr, "Classifier parameters for single-stage examples undefined, using default for task", fullTaskId
            if task == "REN":
                classifierParameters["examples"] = "10,100,1000,2000,3000,4000,4500,5000,5500,6000,7500,10000,20000,25000,28000,50000,60000"
            elif task == "BI":
                classifierParameters["examples"] = "10,100,1000,2500,5000,7500,10000,20000,25000,28000,50000,60000,65000,80000,100000,150000"
            elif task == "DDI":
                classifierParameters["examples"] = "c=10,100,1000,2500,4000,5000,6000,7500,10000,20000,25000,50000:TEES.threshold"
        if classifierParameters["trigger"] == None and not isSingleStage:
            print >> sys.stderr, "Classifier parameters for trigger examples undefined, using default for task", fullTaskId
            classifierParameters["trigger"] = "1000,5000,10000,20000,50000,80000,100000,150000,180000,200000,250000,300000,350000,500000,1000000"
        if classifierParameters["recall"] == None and not isSingleStage:
            print >> sys.stderr, "Recall adjust parameter undefined, using default for task", fullTaskId
            classifierParameters["recall"] = "0.5,0.6,0.65,0.7,0.85,1.0,1.1,1.2"
            if task == "CO":
                classifierParameters["recall"] = "0.8,0.9,0.95,1.0"
        if classifierParameters["edge"] == None and not isSingleStage:
            print >> sys.stderr, "Classifier parameters for edge examples undefined, using default for task", fullTaskId
            classifierParameters["edge"] = "5000,7500,10000,20000,25000,27500,28000,29000,30000,35000,40000,50000,60000,65000"
            if task in ["REL", "CO"]:
                classifierParameters["edge"] = "10,100,1000,5000,7500,10000,20000,25000,28000,50000,60000,65000,100000,500000,1000000"
        if classifierParameters["unmerging"] == None and not isSingleStage:
            print >> sys.stderr, "Classifier parameters for unmerging examples undefined, using default for task", fullTaskId
            classifierParameters["unmerging"] = "1,10,100,500,1000,1500,2500,5000,10000,20000,50000,80000,100000"
        if classifierParameters["modifiers"] == None and not isSingleStage:
            print >> sys.stderr, "Classifier parameters for modifier examples undefined, using default for task", fullTaskId
            classifierParameters["modifiers"] = "5000,10000,20000,50000,100000"
    
    if isSingleStage and exampleStyles["examples"] != None and "names" in exampleStyles["examples"]:
        removeNamesFromEmpty = True
    elif (not isSingleStage) and exampleStyles["trigger"] != None and "names" in exampleStyles["trigger"]:
        removeNamesFromEmpty = True
    else:
        removeNamesFromEmpty = False
    return detector, processUnmerging, processModifiers, isSingleStage, bioNLPSTParams, preprocessorParams, exampleStyles, classifierParameters, removeNamesFromEmpty
コード例 #5
0
ファイル: train.py プロジェクト: ayoshiaki/TEES
def getTaskSettings(task, detector, bioNLPSTParams, preprocessorParams, 
                    inputFiles, exampleStyles, classifierParameters):
    if task != None:
        print >> sys.stderr, "*** Defining training settings for task", task, "***"
        fullTaskId = task
        subTask = 2
        if "." in task:
            task, subTask = task.split(".")
            subTask = int(subTask)
        dataPath = Settings.CORPUS_DIR
        for dataset in ["devel", "train", "test"]:
            if inputFiles[dataset] == None and inputFiles[dataset] != "None":
                inputFiles[dataset] = os.path.join(dataPath, task.replace("-FULL", "") + "-"+dataset+".xml")
            if task == "ID11" and dataset == "train":
                inputFiles[dataset] = Catenate.catenate([os.path.join(dataPath, "ID11-train.xml"), os.path.join(dataPath, "GE11-devel.xml"),
                                                         os.path.join(dataPath, "GE11-train.xml")], "training/ID11-train-and-GE11-devel-and-train.xml.gz", fast=True)
            if inputFiles[dataset] == "None":
                inputFiles[dataset] = None
            if inputFiles[dataset] != None and not os.path.exists(inputFiles[dataset]):
                inputFiles[dataset] = None
                print >> sys.stderr, "Input file", inputFiles[dataset], "for set '" + dataset + "' does not exist, skipping."
        assert inputFiles["train"] != None # at least training set must exist
        # Example generation parameters
        if task == "CO11":
            detector = "Detectors.CODetector"
        elif task in ["BI11-FULL", "DDI11-FULL"]:
            detector = "Detectors.EventDetector"
        
        # BioNLP Shared Task and preprocessing parameters
        if task == "BI11-FULL":
            bioNLPSTParams = Parameters.cat(bioNLPSTParams, "convert:scores", "BioNLP Shared Task / " + fullTaskId, ["default"]) # the shared task evaluator is not designed for predicted entities
        elif task == "REL11":
            bioNLPSTParams = Parameters.cat(bioNLPSTParams, "convert:evaluate:scores:a2Tag=rel", "BioNLP Shared Task / " + fullTaskId, ["default"])
        elif task not in ["DDI11", "DDI11-FULL", "DDI13"]:
            bioNLPSTParams = Parameters.cat(bioNLPSTParams, "convert:evaluate:scores", "BioNLP Shared Task / " + fullTaskId, ["default"])
        
        # Preprocessing parameters
        if task in ["BI11", "BI11-FULL", "BB11", "DDI11", "DDI11-FULL"]:
            Parameters.cat("intermediateFiles:omitSteps=NER,DIVIDE-SETS", preprocessorParams, "Preprocessor /" + fullTaskId, ["default"])
        else: # parse only sentences where BANNER found an entity
            Parameters.cat("intermediateFiles:omitSteps=DIVIDE-SETS:PARSE.requireEntities", preprocessorParams, "Preprocessor /" + fullTaskId, ["default"])
        
        # Example style parameters for single-stage tasks
        if task == "REN11":
            exampleStyles["examples"] = Parameters.cat("undirected:bacteria_renaming:maskTypeAsProtein=Gene", exampleStyles["examples"], "Single-stage example style / " + fullTaskId)
        elif task == "DDI11":
            exampleStyles["examples"] = Parameters.cat("drugbank_features:ddi_mtmx:filter_shortest_path=conj_and", exampleStyles["examples"], "Single-stage example style / " + fullTaskId)
        elif task == "DDI13":
            exampleStyles["examples"] = Parameters.cat("keep_neg:drugbank_features:filter_shortest_path=conj_and", exampleStyles["examples"], "Single-stage example style / " + fullTaskId)
        elif task == "BI11":
            exampleStyles["edge"] = Parameters.cat("bi_features", exampleStyles["edge"], "Edge example style / " + fullTaskId)
        # Edge style
        if task in ["GE09", "GE11", "GE13"] and subTask == 1:
            exampleStyles["edge"] = Parameters.cat("genia_features:genia_task1", exampleStyles["edge"])
        elif task in ["GE09", "GE11", "GE13"]:
            exampleStyles["edge"] = Parameters.cat("genia_features", exampleStyles["edge"])
        elif task == "REL11":
            exampleStyles["edge"] = Parameters.cat("rel_features", exampleStyles["edge"], "Edge example style / " + fullTaskId)
        elif task == "DDI11-FULL":
            exampleStyles["edge"] = Parameters.cat("drugbank_features:filter_shortest_path=conj_and", exampleStyles["edge"], "Edge example style / " + fullTaskId)
        elif task == "CO11":
            exampleStyles["edge"] = Parameters.cat("co_features", exampleStyles["edge"], "Edge example style / " + fullTaskId)
        elif task == "BI11-FULL":
            exampleStyles["edge"] = Parameters.cat("bi_features", exampleStyles["edge"], "Edge example style / " + fullTaskId)
        # Trigger style
        if task in ["GE09", "GE11", "GE13"] and subTask == 1:
            exampleStyles["trigger"] = Parameters.cat("genia_task1", exampleStyles["trigger"], "Trigger example style / " + fullTaskId)
        elif task in ["EPI11", "PC13"]:
            exampleStyles["trigger"] = Parameters.cat("epi_merge_negated", exampleStyles["trigger"], "Trigger example style / " + fullTaskId)
        elif task == "BB11": # "bb_features:build_for_nameless:wordnet"
            exampleStyles["trigger"] = Parameters.cat("bb_features:build_for_nameless", exampleStyles["trigger"], "Trigger example style / " + fullTaskId)
        elif task == "BB13T3": # "bb_features:build_for_nameless:wordnet"
            exampleStyles["trigger"] = Parameters.cat("bb_features:build_for_nameless", exampleStyles["trigger"], "Trigger example style / " + fullTaskId)
        elif task == "REL11":
            exampleStyles["trigger"] = Parameters.cat("rel_features", exampleStyles["trigger"], "Trigger example style / " + fullTaskId)
        elif task in ["BI11-FULL", "DDI11-FULL"]:
            exampleStyles["trigger"] = "build_for_nameless:names"        
        # Classifier parameters
        if task == "DDI11":
            classifierParameters["examples"] = Parameters.cat("c=10,100,1000,2500,4000,5000,6000,7500,10000,20000,25000,50000:TEES.threshold", classifierParameters["examples"], "Classifier parameters for single-stage examples" + fullTaskId)
        #elif task == "DDI13":
        #    classifierParameters["examples"] = Parameters.cat("c=10,100,1000,2500,4000,5000,6000,7500,10000,20000,25000,50000:TEES.threshold", classifierParameters["examples"], "Classifier parameters for single-stage examples" + fullTaskId)
        elif task == "CO11":
            classifierParameters["edge"] = Parameters.cat("c=1000,4500,5000,7500,10000,20000,25000,27500,28000,29000,30000,35000,40000,50000,60000,65000", classifierParameters["examples"], "Classifier parameters for edges / " + fullTaskId)
            classifierParameters["trigger"] = Parameters.cat("c=1000,5000,10000,20000,50000,80000,100000,150000,180000,200000,250000,300000,350000,500000,1000000", classifierParameters["examples"], "Classifier parameters for triggers / " + fullTaskId)
            classifierParameters["recall"] = Parameters.cat("0.8,0.9,0.95,1.0", classifierParameters["recall"], "Recall adjust / " + fullTaskId)
    
    return detector, bioNLPSTParams, preprocessorParams
コード例 #6
0
ファイル: train.py プロジェクト: MaximumEntropy/UPSITE
def getTaskSettings(task, detector, bioNLPSTParams, preprocessorParams,
                    inputFiles, exampleStyles, classifierParameters):
    if task != None:
        print >> sys.stderr, "*** Defining training settings for task", task, "***"
        fullTaskId = task
        subTask = 2
        if "." in task:
            task, subTask = task.split(".")
            subTask = int(subTask)
        dataPath = Settings.CORPUS_DIR
        for dataset in ["devel", "train", "test"]:
            if inputFiles[dataset] == None and inputFiles[dataset] != "None":
                inputFiles[dataset] = os.path.join(
                    dataPath,
                    task.replace("-FULL", "") + "-" + dataset + ".xml")
            if task == "ID11" and dataset == "train":
                inputFiles[dataset] = Catenate.catenate(
                    [
                        os.path.join(dataPath, "ID11-train.xml"),
                        os.path.join(dataPath, "GE11-devel.xml"),
                        os.path.join(dataPath, "GE11-train.xml")
                    ],
                    "training/ID11-train-and-GE11-devel-and-train.xml.gz",
                    fast=True)
            if inputFiles[dataset] == "None":
                inputFiles[dataset] = None
            if inputFiles[dataset] != None and not os.path.exists(
                    inputFiles[dataset]):
                inputFiles[dataset] = None
                print >> sys.stderr, "Input file", inputFiles[
                    dataset], "for set '" + dataset + "' does not exist, skipping."
        assert inputFiles["train"] != None  # at least training set must exist
        # Example generation parameters
        if task == "CO11":
            detector = "Detectors.CODetector"
        elif task in ["BI11-FULL", "DDI11-FULL"]:
            detector = "Detectors.EventDetector"

        # BioNLP Shared Task and preprocessing parameters
        if task == "BI11-FULL":
            bioNLPSTParams = Parameters.cat(
                bioNLPSTParams, "convert:scores",
                "BioNLP Shared Task / " + fullTaskId, ["default"]
            )  # the shared task evaluator is not designed for predicted entities
        elif task == "REL11":
            bioNLPSTParams = Parameters.cat(
                bioNLPSTParams, "convert:evaluate:scores:a2Tag=rel",
                "BioNLP Shared Task / " + fullTaskId, ["default"])
        elif task not in ["DDI11", "DDI11-FULL", "DDI13"]:
            bioNLPSTParams = Parameters.cat(
                bioNLPSTParams, "convert:evaluate:scores",
                "BioNLP Shared Task / " + fullTaskId, ["default"])

        # Preprocessing parameters
        if task in ["BI11", "BI11-FULL", "BB11", "DDI11", "DDI11-FULL"]:
            Parameters.cat("intermediateFiles:omitSteps=NER,DIVIDE-SETS",
                           preprocessorParams, "Preprocessor /" + fullTaskId,
                           ["default"])
        else:  # parse only sentences where BANNER found an entity
            Parameters.cat(
                "intermediateFiles:omitSteps=DIVIDE-SETS:PARSE.requireEntities",
                preprocessorParams, "Preprocessor /" + fullTaskId, ["default"])

        # Example style parameters for single-stage tasks
        if task == "REN11":
            exampleStyles["examples"] = Parameters.cat(
                "undirected:bacteria_renaming:maskTypeAsProtein=Gene",
                exampleStyles["examples"],
                "Single-stage example style / " + fullTaskId)
        elif task == "DDI11":
            exampleStyles["examples"] = Parameters.cat(
                "drugbank_features:ddi_mtmx:filter_shortest_path=conj_and",
                exampleStyles["examples"],
                "Single-stage example style / " + fullTaskId)
        elif task == "DDI13":
            exampleStyles["examples"] = Parameters.cat(
                "keep_neg:drugbank_features:filter_shortest_path=conj_and",
                exampleStyles["examples"],
                "Single-stage example style / " + fullTaskId)
        elif task == "BI11":
            exampleStyles["edge"] = Parameters.cat(
                "bi_features", exampleStyles["edge"],
                "Edge example style / " + fullTaskId)
        # Edge style
        if task in ["GE09", "GE11", "GE13"] and subTask == 1:
            exampleStyles["edge"] = Parameters.cat(
                "genia_features:genia_task1", exampleStyles["edge"])
        elif task in ["GE09", "GE11", "GE13"]:
            exampleStyles["edge"] = Parameters.cat("genia_features",
                                                   exampleStyles["edge"])
        elif task == "REL11":
            exampleStyles["edge"] = Parameters.cat(
                "rel_features", exampleStyles["edge"],
                "Edge example style / " + fullTaskId)
        elif task == "DDI11-FULL":
            exampleStyles["edge"] = Parameters.cat(
                "drugbank_features:filter_shortest_path=conj_and",
                exampleStyles["edge"], "Edge example style / " + fullTaskId)
        elif task == "CO11":
            exampleStyles["edge"] = Parameters.cat(
                "co_features", exampleStyles["edge"],
                "Edge example style / " + fullTaskId)
        elif task == "BI11-FULL":
            exampleStyles["edge"] = Parameters.cat(
                "bi_features", exampleStyles["edge"],
                "Edge example style / " + fullTaskId)
        # Trigger style
        if task in ["GE09", "GE11", "GE13"] and subTask == 1:
            exampleStyles["trigger"] = Parameters.cat(
                "genia_task1", exampleStyles["trigger"],
                "Trigger example style / " + fullTaskId)
        elif task in ["EPI11", "PC13"]:
            exampleStyles["trigger"] = Parameters.cat(
                "epi_merge_negated", exampleStyles["trigger"],
                "Trigger example style / " + fullTaskId)
        elif task == "BB11":  # "bb_features:build_for_nameless:wordnet"
            exampleStyles["trigger"] = Parameters.cat(
                "bb_features:build_for_nameless", exampleStyles["trigger"],
                "Trigger example style / " + fullTaskId)
        elif task == "BB13T3":  # "bb_features:build_for_nameless:wordnet"
            exampleStyles["trigger"] = Parameters.cat(
                "bb_features:build_for_nameless", exampleStyles["trigger"],
                "Trigger example style / " + fullTaskId)
        elif task == "REL11":
            exampleStyles["trigger"] = Parameters.cat(
                "rel_features", exampleStyles["trigger"],
                "Trigger example style / " + fullTaskId)
        elif task in ["BI11-FULL", "DDI11-FULL"]:
            exampleStyles["trigger"] = "build_for_nameless:names"
        # Classifier parameters
        if task == "DDI11":
            classifierParameters["examples"] = Parameters.cat(
                "c=10,100,1000,2500,4000,5000,6000,7500,10000,20000,25000,50000:TEES.threshold",
                classifierParameters["examples"],
                "Classifier parameters for single-stage examples" + fullTaskId)
        #elif task == "DDI13":
        #    classifierParameters["examples"] = Parameters.cat("c=10,100,1000,2500,4000,5000,6000,7500,10000,20000,25000,50000:TEES.threshold", classifierParameters["examples"], "Classifier parameters for single-stage examples" + fullTaskId)
        elif task == "CO11":
            classifierParameters["edge"] = Parameters.cat(
                "c=1000,4500,5000,7500,10000,20000,25000,27500,28000,29000,30000,35000,40000,50000,60000,65000",
                classifierParameters["examples"],
                "Classifier parameters for edges / " + fullTaskId)
            classifierParameters["trigger"] = Parameters.cat(
                "c=1000,5000,10000,20000,50000,80000,100000,150000,180000,200000,250000,300000,350000,500000,1000000",
                classifierParameters["examples"],
                "Classifier parameters for triggers / " + fullTaskId)
            classifierParameters["recall"] = Parameters.cat(
                "0.8,0.9,0.95,1.0", classifierParameters["recall"],
                "Recall adjust / " + fullTaskId)

    return detector, bioNLPSTParams, preprocessorParams
コード例 #7
0
ファイル: train.py プロジェクト: sbnlp/2017BioNLPEvaluation
def getTaskSettings(task,
                    detector,
                    bioNLPSTParams,
                    preprocessorParams,
                    inputFiles,
                    exampleStyles,
                    classifierParameters,
                    folds,
                    corpusDir=None):
    if task != None:
        print >> sys.stderr, "*** Defining training settings for task", task, "***"
        fullTaskId = task
        subTask = 2
        if "." in task:
            task, subTask = task.split(".")
            subTask = int(subTask)
        if corpusDir == None:
            corpusDir = Settings.CORPUS_DIR
        for dataset in ["devel", "train", "test"]:
            if inputFiles[dataset] == None and inputFiles[dataset] != "None":
                if task.startswith("DDI13"):
                    if dataset in ["devel", "train"]:
                        inputFiles[dataset] = os.path.join(
                            corpusDir, "DDI13-train.xml")
                    elif dataset == "test":
                        if task.endswith("T91"):
                            inputFiles[dataset] = os.path.join(
                                corpusDir, "DDI13-test-task9.1.xml")
                        elif task.endswith("T92") or task.endswith("FULL"):
                            inputFiles[dataset] = os.path.join(
                                corpusDir, "DDI13-test-task9.2.xml")
                elif task == "ID11" and dataset == "train":
                    inputFiles[dataset] = Catenate.catenate(
                        [
                            os.path.join(corpusDir, "ID11-train.xml"),
                            os.path.join(corpusDir, "GE11-devel.xml"),
                            os.path.join(corpusDir, "GE11-train.xml")
                        ],
                        "training/ID11-train-and-GE11-devel-and-train.xml.gz",
                        fast=True)
                else:
                    inputFiles[dataset] = os.path.join(
                        corpusDir,
                        task.replace("-FULL", "") + "-" + dataset + ".xml")

            if inputFiles[dataset] == "None":
                inputFiles[dataset] = None
            if inputFiles[dataset] != None and not os.path.exists(
                    inputFiles[dataset]):
                fullPath = os.path.join(Settings.CORPUS_DIR,
                                        inputFiles[dataset])
                if os.path.exists(fullPath):
                    inputFiles[dataset] = fullPath
                else:
                    inputFiles[dataset] = None
                    print >> sys.stderr, "Input file", inputFiles[
                        dataset], "for set '" + dataset + "' does not exist, skipping."
        assert inputFiles["train"] != None  # at least training set must exist
        # Example generation parameters
        if task == "CO11":
            detector = "Detectors.CODetector"
        elif task in [
                "BI11-FULL", "DDI11-FULL", "DDI13-FULL", "BB_EVENT_16-FULL"
        ]:
            detector = "Detectors.EventDetector"
        elif task.startswith("DDI13"):
            if task.endswith("T91"):
                detector = "Detectors.EntityDetector"
            elif task.endswith("T92"):
                detector = "Detectors.EdgeDetector"

        #######################################################################
        # BioNLP Shared Task and preprocessing parameters
        #######################################################################
        if task == "BI11-FULL":
            bioNLPSTParams = Parameters.cat(
                bioNLPSTParams, "convert:scores",
                "BioNLP Shared Task / " + fullTaskId, ["default"]
            )  # the shared task evaluator is not designed for predicted entities
        elif task == "REL11":
            bioNLPSTParams = Parameters.cat(
                bioNLPSTParams, "convert:evaluate:scores:a2Tag=rel",
                "BioNLP Shared Task / " + fullTaskId, ["default"])
        elif task in ("BB_EVENT_16", "BB_EVENT_16-FULL", "BB_EVENT_NER_16",
                      "SDB16"):
            bioNLPSTParams = Parameters.cat(
                bioNLPSTParams, "convert=zip",
                "BioNLP Shared Task / " + fullTaskId, ["default"])
        elif task not in [
                "DDI11", "DDI11-FULL", "DDI13T91", "DDI13T92", "DDI13-FULL"
        ]:
            bioNLPSTParams = Parameters.cat(
                bioNLPSTParams, "convert:evaluate:scores",
                "BioNLP Shared Task / " + fullTaskId, ["default"])

        #######################################################################
        # Preprocessing parameters
        #######################################################################
        if task in [
                "BI11", "BI11-FULL", "BB11", "DDI11", "DDI11-FULL", "DDI13T91",
                "DDI13T92", "DDI13-FULL"
        ]:
            Parameters.cat("intermediateFiles:omitSteps=NER,DIVIDE-SETS",
                           preprocessorParams, "Preprocessor /" + fullTaskId,
                           ["default"])
        else:  # parse only sentences where BANNER found an entity
            Parameters.cat(
                "intermediateFiles:omitSteps=DIVIDE-SETS:PARSE.requireEntities",
                preprocessorParams, "Preprocessor /" + fullTaskId, ["default"])

        #######################################################################
        # Example style parameters
        #######################################################################
        # Example style parameters for single-stage tasks #####################
        msg = "Single-stage example style / " + fullTaskId
        if task == "REN11":
            exampleStyles["examples"] = Parameters.cat(
                "undirected:bacteria_renaming:maskTypeAsProtein=Gene",
                exampleStyles["examples"], msg)
        elif task == "DDI11":
            exampleStyles["examples"] = Parameters.cat(
                "drugbank_features:ddi_mtmx:filter_shortest_path=conj_and",
                exampleStyles["examples"], msg)
        elif task.startswith("DDI13"):
            if task.endswith("T91"):
                exampleStyles["examples"] = Parameters.cat(
                    "names:build_for_nameless:ddi13_features:drugbank_features",
                    exampleStyles["examples"], msg)
            elif task.endswith("T92"):
                exampleStyles["examples"] = Parameters.cat(
                    "keep_neg:drugbank_features:filter_shortest_path=conj_and",
                    exampleStyles["examples"], msg)
        elif task == "BI11":
            exampleStyles["examples"] = Parameters.cat(
                "bi_features", exampleStyles["examples"], msg)
        elif task == "BB_EVENT_16":
            exampleStyles["examples"] = Parameters.cat(
                "keep_neg", exampleStyles["examples"], msg
            )  #exampleStyles["examples"] = Parameters.cat("linear_features:keep_neg", exampleStyles["examples"], msg)
        elif task == "SDB16":
            exampleStyles["examples"] = Parameters.cat(
                "sdb_merge:sdb_features", exampleStyles["examples"], msg)
        # Edge style ##########################################################
        msg = "Edge example style / " + fullTaskId
        if task in ["GE09", "GE11", "GE13"] and subTask == 1:
            exampleStyles["edge"] = Parameters.cat(
                "genia_features:genia_task1", exampleStyles["edge"], msg)
        elif task in ["GE09", "GE11", "GE13"]:
            exampleStyles["edge"] = Parameters.cat("genia_features",
                                                   exampleStyles["edge"], msg)
        elif task == "REL11":
            exampleStyles["edge"] = Parameters.cat("rel_features",
                                                   exampleStyles["edge"], msg)
        elif task == "DDI11-FULL":
            exampleStyles["edge"] = Parameters.cat(
                "drugbank_features:filter_shortest_path=conj_and",
                exampleStyles["edge"], msg)
        elif task == "DDI13-FULL":
            exampleStyles["edge"] = Parameters.cat(
                "keep_neg:drugbank_features:filter_shortest_path=conj_and",
                exampleStyles["edge"], msg)
        elif task == "CO11":
            exampleStyles["edge"] = Parameters.cat("co_features",
                                                   exampleStyles["edge"], msg)
        elif task == "BI11-FULL":
            exampleStyles["edge"] = Parameters.cat("bi_features",
                                                   exampleStyles["edge"], msg)
        # Trigger style #######################################################
        msg = "Trigger example style / " + fullTaskId
        if task in ["GE09", "GE11", "GE13"] and subTask == 1:
            exampleStyles["trigger"] = Parameters.cat("genia_task1",
                                                      exampleStyles["trigger"],
                                                      msg)
        elif task in ["EPI11", "PC13"]:
            exampleStyles["trigger"] = Parameters.cat("epi_merge_negated",
                                                      exampleStyles["trigger"],
                                                      msg)
        elif task == "BB11":  # "bb_features:build_for_nameless:wordnet"
            exampleStyles["trigger"] = Parameters.cat("bb_features",
                                                      exampleStyles["trigger"],
                                                      msg)
        elif task == "BB13T3":  # "bb_features:build_for_nameless:wordnet"
            exampleStyles["trigger"] = Parameters.cat("bb_features",
                                                      exampleStyles["trigger"],
                                                      msg)
        elif task == "REL11":
            exampleStyles["trigger"] = Parameters.cat("rel_features",
                                                      exampleStyles["trigger"],
                                                      msg)
        elif task in ["BI11-FULL", "DDI11-FULL"]:
            exampleStyles["trigger"] = "names:build_for_nameless"
        elif task == "DDI13-FULL":
            exampleStyles[
                "trigger"] = "names:build_for_nameless:ddi13_features:drugbank_features"
        elif task == "BB_EVENT_16-FULL":
            exampleStyles["trigger"] = Parameters.cat(
                "bb_spans:bb_features:ontobiotope_features:build_for_nameless:all_tokens:only_types=Bacteria,Habitat,Geographical",
                exampleStyles["trigger"], msg)
        elif task in "BB_EVENT_NER_16":
            exampleStyles["trigger"] = Parameters.cat(
                "bb_spans:bb_features:ontobiotope_features:build_for_nameless:all_tokens",
                exampleStyles["trigger"], msg)

        #######################################################################
        # Classifier parameters
        #######################################################################
        if task == "DDI11":
            classifierParameters["examples"] = Parameters.cat(
                "c=10,100,1000,2500,4000,5000,6000,7500,10000,20000,25000,50000:TEES.threshold",
                classifierParameters["examples"],
                "Classifier parameters for single-stage examples" + fullTaskId)
        #elif task == "DDI13":
        #    classifierParameters["examples"] = Parameters.cat("c=10,100,1000,2500,4000,5000,6000,7500,10000,20000,25000,50000:TEES.threshold", classifierParameters["examples"], "Classifier parameters for single-stage examples" + fullTaskId)
        elif task == "CO11":
            classifierParameters["edge"] = Parameters.cat(
                "c=1000,4500,5000,7500,10000,20000,25000,27500,28000,29000,30000,35000,40000,50000,60000,65000",
                classifierParameters["edge"],
                "Classifier parameters for edges / " + fullTaskId)
            classifierParameters["trigger"] = Parameters.cat(
                "c=1000,5000,10000,20000,50000,80000,100000,150000,180000,200000,250000,300000,350000,500000,1000000",
                classifierParameters["trigger"],
                "Classifier parameters for triggers / " + fullTaskId)
            classifierParameters["recall"] = Parameters.cat(
                "0.8,0.9,0.95,1.0", classifierParameters["recall"],
                "Recall adjust / " + fullTaskId)
        elif task == "BB_EVENT_16":
            classifierParameters["examples"] = Parameters.cat(
                "c=10,20,30,40,50,60,70,80,100,110,115,120,125,130,140,150,200,500,1000,2000,3000,4000,4500,5000,7500,10000,20000,25000,27500,28000,29000,30000,35000,40000,50000,60000,65000",
                classifierParameters["examples"],
                "Classifier parameters for edges / " + fullTaskId)
        elif task in ("BB_EVENT_16-FULL", "BB_EVENT_NER_16"):
            classifierParameters["edge"] = Parameters.cat(
                "c=10,20,50,80,100,110,115,120,125,130,140,150,200,500,1000",
                classifierParameters["edge"],
                "Classifier parameters for edges / " + fullTaskId)
        elif task == "SDB16":
            classifierParameters["examples"] = Parameters.cat(
                "c=1000,4500,5000,7500,10000,20000,25000,27500,28000,29000,30000,35000,40000,50000,60000,65000,80000,100000,150000",
                classifierParameters["examples"],
                "Classifier parameters for single-stage examples / " +
                fullTaskId)
        # Training fold parameters ############################################
        if task.startswith("DDI13"):
            folds["devel"] = ["train1", "train2", "train3", "train4"]
            folds["train"] = ["train5", "train6", "train7", "train8", "train9"]

    return detector, bioNLPSTParams, preprocessorParams, folds
コード例 #8
0
ファイル: train.py プロジェクト: jbjorne/TEES
def getTaskSettings(task, detector, bioNLPSTParams, preprocessorParams, 
                    inputFiles, exampleStyles, classifierParameters, folds, corpusDir=None, useKerasDetector=False):
    if task != None:
        print >> sys.stderr, "*** Defining training settings for task", task, "***"
        fullTaskId = task
        task, subTask = getSubTask(task)
        if corpusDir == None:
            corpusDir = Settings.CORPUS_DIR
        print >> sys.stderr, "Loading corpus", task, "from", corpusDir
        for dataset in ["devel", "train", "test"]:
            if inputFiles[dataset] == None:
                if task.startswith("DDI13") and task != "DDI13":
                    if dataset in ["devel", "train"]:
                        inputFiles[dataset] = os.path.join(corpusDir, "DDI13-train.xml")
                    elif dataset == "test":
                        if task.endswith("T91"):
                            inputFiles[dataset] = os.path.join(corpusDir, "DDI13-test-task9.1.xml")
                        elif task.endswith("T92") or task.endswith("FULL"):
                            inputFiles[dataset] = os.path.join(corpusDir, "DDI13-test-task9.2.xml")
                elif task == "ID11" and dataset == "train":
                    inputFiles[dataset] = Catenate.catenate([os.path.join(corpusDir, "ID11-train.xml"), os.path.join(corpusDir, "GE11-devel.xml"),
                                                             os.path.join(corpusDir, "GE11-train.xml")], "training/ID11-train-and-GE11-devel-and-train.xml.gz", fast=True)
                else:
                    inputFiles[dataset] = os.path.join(corpusDir, task.replace("-FULL", "") + "-"+dataset+".xml")
                
            if inputFiles[dataset] == "skip":
                inputFiles[dataset] = None
            if inputFiles[dataset] != None and not os.path.exists(inputFiles[dataset]):
                fullPath = os.path.join(Settings.CORPUS_DIR, inputFiles[dataset])
                if os.path.exists(fullPath):
                    inputFiles[dataset] = fullPath
                else:
                    inputFiles[dataset] = None
                    print >> sys.stderr, "Input file", inputFiles[dataset], "for set '" + dataset + "' does not exist, skipping."
        assert inputFiles["train"] != None # at least training set must exist
        # Example generation parameters
        if detector == None:
            if task == "CO11":
                detector = "Detectors.CODetector"
            elif task in ["BI11-FULL", "DDI11-FULL", "DDI13-FULL", "BB_EVENT_16-FULL"]:
                detector = "Detectors.EventDetector"
            elif task.startswith("DDI13"):
                if task.endswith("T91"):
                    detector = "Detectors.EntityDetector"
                elif task.endswith("T92") or task == "DDI13":
                    detector = "Detectors.EdgeDetector"
        
        #######################################################################
        # BioNLP Shared Task and preprocessing parameters
        #######################################################################
        if task == "BI11-FULL":
            bioNLPSTParams = Parameters.cat(bioNLPSTParams, "convert:scores", "BioNLP Shared Task / " + fullTaskId, ["default"]) # the shared task evaluator is not designed for predicted entities
        elif task == "REL11":
            bioNLPSTParams = Parameters.cat(bioNLPSTParams, "convert:evaluate:scores:a2Tag=rel", "BioNLP Shared Task / " + fullTaskId, ["default"])
        elif task in ("BB_EVENT_16", "BB_EVENT_16-FULL", "BB_EVENT_NER_16", "SDB16"):
            bioNLPSTParams = Parameters.cat(bioNLPSTParams, "convert=zip", "BioNLP Shared Task / " + fullTaskId, ["default"])
        elif task not in ["DDI11", "DDI11-FULL", "DDI13T91", "DDI13T92", "DDI13-FULL", "DDI13", "CP17", "SEMEVAL10T8"]:
            bioNLPSTParams = Parameters.cat(bioNLPSTParams, "convert:evaluate:scores", "BioNLP Shared Task / " + fullTaskId, ["default"])
        else:
            bioNLPSTParams = "skip"
        
        #######################################################################
        # Preprocessing parameters
        #######################################################################
        if task in ["BI11", "BI11-FULL", "BB11", "DDI11", "DDI11-FULL", "DDI13T91", "DDI13T92", "DDI13-FULL", "DDI13"]:
            Parameters.cat("intermediateFiles:omitSteps=NER,DIVIDE-SETS", preprocessorParams, "Preprocessor /" + fullTaskId, ["default"])
        else: # parse only sentences where BANNER found an entity
            Parameters.cat("intermediateFiles:omitSteps=DIVIDE-SETS:PARSE.requireEntities", preprocessorParams, "Preprocessor /" + fullTaskId, ["default"])
        
        #######################################################################
        # Example style parameters
        #######################################################################
        if not useKerasDetector:
            # Example style parameters for single-stage tasks #####################
            msg = "Single-stage example style / " + fullTaskId
            if task == "REN11":
                exampleStyles["examples"] = Parameters.cat("undirected:bacteria_renaming:maskTypeAsProtein=Gene", exampleStyles["examples"], msg)
            elif task == "DDI11":
                exampleStyles["examples"] = Parameters.cat("drugbank_features:ddi_mtmx:filter_shortest_path=conj_and", exampleStyles["examples"], msg)
            elif task.startswith("DDI13"):
                if task.endswith("T91"):
                    exampleStyles["examples"] = Parameters.cat("names:build_for_nameless:ddi13_features:drugbank_features", exampleStyles["examples"], msg)
                elif task.endswith("T92") or task == "DDI13":
                    exampleStyles["examples"] = Parameters.cat("keep_neg:drugbank_features:filter_shortest_path=conj_and", exampleStyles["examples"], msg)
            elif task == "BI11":
                exampleStyles["examples"] = Parameters.cat("bi_features", exampleStyles["examples"], msg)
            elif task == "BB_EVENT_16":
                exampleStyles["examples"] = Parameters.cat("keep_neg", exampleStyles["examples"], msg) #exampleStyles["examples"] = Parameters.cat("linear_features:keep_neg", exampleStyles["examples"], msg)
            elif task == "SDB16":
                exampleStyles["examples"] = Parameters.cat("sdb_merge:sdb_features", exampleStyles["examples"], msg)
            # Edge style ##########################################################
            msg = "Edge example style / " + fullTaskId
            if task in ["GE09", "GE11", "GE13"] and subTask == 1:
                exampleStyles["edge"] = Parameters.cat("genia_features:genia_task1", exampleStyles["edge"], msg)
            elif task in ["GE09", "GE11", "GE13"]:
                exampleStyles["edge"] = Parameters.cat("genia_features", exampleStyles["edge"], msg)
            elif task == "REL11":
                exampleStyles["edge"] = Parameters.cat("rel_features", exampleStyles["edge"], msg)
            elif task == "DDI11-FULL":
                exampleStyles["edge"] = Parameters.cat("drugbank_features:filter_shortest_path=conj_and", exampleStyles["edge"], msg)
            elif task == "DDI13-FULL":
                exampleStyles["edge"] = Parameters.cat("keep_neg:drugbank_features:filter_shortest_path=conj_and", exampleStyles["edge"], msg)
            elif task == "CO11":
                exampleStyles["edge"] = Parameters.cat("co_features", exampleStyles["edge"], msg)
            elif task == "BI11-FULL":
                exampleStyles["edge"] = Parameters.cat("bi_features", exampleStyles["edge"], msg)
            # Trigger style #######################################################
            msg = "Trigger example style / " + fullTaskId
            if task in ["GE09", "GE11", "GE13"] and subTask == 1:
                exampleStyles["trigger"] = Parameters.cat("genia_task1", exampleStyles["trigger"], msg)
            elif task in ["EPI11", "PC13"]:
                exampleStyles["trigger"] = Parameters.cat("epi_merge_negated", exampleStyles["trigger"], msg)
            elif task == "BB11": # "bb_features:build_for_nameless:wordnet"
                exampleStyles["trigger"] = Parameters.cat("bb_features", exampleStyles["trigger"], msg)
            elif task == "BB13T3": # "bb_features:build_for_nameless:wordnet"
                exampleStyles["trigger"] = Parameters.cat("bb_features", exampleStyles["trigger"], msg)
            elif task == "REL11":
                exampleStyles["trigger"] = Parameters.cat("rel_features", exampleStyles["trigger"], msg)
            elif task in ["BI11-FULL", "DDI11-FULL"]:
                exampleStyles["trigger"] = "names:build_for_nameless"
            elif task == "DDI13-FULL":
                exampleStyles["trigger"] = "names:build_for_nameless:ddi13_features:drugbank_features"
            elif task == "BB_EVENT_16-FULL":
                exampleStyles["trigger"] = Parameters.cat("bb_spans:bb_features:ontobiotope_features:build_for_nameless:all_tokens:only_types=Bacteria,Habitat,Geographical", exampleStyles["trigger"], msg)
            elif task in "BB_EVENT_NER_16":
                exampleStyles["trigger"] = Parameters.cat("bb_spans:bb_features:ontobiotope_features:build_for_nameless:all_tokens", exampleStyles["trigger"], msg)            
                
            #######################################################################
            # Classifier parameters
            #######################################################################
            if task == "DDI11":
                classifierParameters["examples"] = Parameters.cat("c=10,100,1000,2500,4000,5000,6000,7500,10000,20000,25000,50000:TEES.threshold", classifierParameters["examples"], "Classifier parameters for single-stage examples" + fullTaskId)
            #elif task == "DDI13":
            #    classifierParameters["examples"] = Parameters.cat("c=10,100,1000,2500,4000,5000,6000,7500,10000,20000,25000,50000:TEES.threshold", classifierParameters["examples"], "Classifier parameters for single-stage examples" + fullTaskId)
            elif task == "CO11":
                classifierParameters["edge"] = Parameters.cat("c=1000,4500,5000,7500,10000,20000,25000,27500,28000,29000,30000,35000,40000,50000,60000,65000", classifierParameters["edge"], "Classifier parameters for edges / " + fullTaskId)
                classifierParameters["trigger"] = Parameters.cat("c=1000,5000,10000,20000,50000,80000,100000,150000,180000,200000,250000,300000,350000,500000,1000000", classifierParameters["trigger"], "Classifier parameters for triggers / " + fullTaskId)
                classifierParameters["recall"] = Parameters.cat("0.8,0.9,0.95,1.0", classifierParameters["recall"], "Recall adjust / " + fullTaskId)
            elif task == "BB_EVENT_16":
                classifierParameters["examples"] = Parameters.cat("c=10,20,30,40,50,60,70,80,100,110,115,120,125,130,140,150,200,500,1000,2000,3000,4000,4500,5000,7500,10000,20000,25000,27500,28000,29000,30000,35000,40000,50000,60000,65000", classifierParameters["examples"], "Classifier parameters for edges / " + fullTaskId)
            elif task in ("BB_EVENT_16-FULL", "BB_EVENT_NER_16"):
                classifierParameters["edge"] = Parameters.cat("c=10,20,50,80,100,110,115,120,125,130,140,150,200,500,1000", classifierParameters["edge"], "Classifier parameters for edges / " + fullTaskId)
            elif task == "SDB16":
                classifierParameters["examples"] = Parameters.cat("c=1000,4500,5000,7500,10000,20000,25000,27500,28000,29000,30000,35000,40000,50000,60000,65000,80000,100000,150000", classifierParameters["examples"], "Classifier parameters for single-stage examples / " + fullTaskId)
        
        # Training fold parameters ############################################
        if task.startswith("DDI13") and task != "DDI13":
            #folds["devel"]=["train1", "train2", "train3", "train4"]
            #folds["train"]=["train5", "train6", "train7", "train8", "train9"]
            folds["devel"]=["train1", "train2", "train3"]
            folds["train"]=["train4", "train5", "train6", "train7", "train8", "train9"]
        
    return detector, bioNLPSTParams, preprocessorParams, folds