def train(training_list, model_path, format, is_crf=True, grid=False):
# Read the data into a Note object
notes = []
for txt, con in training_list:
note_tmp = Note(format) # Create Note
note_tmp.read(txt, con) # Read data into Note
notes.append(note_tmp) # Add the Note to the list
# file names
if not notes:
print 'Error: Cannot train on 0 files. Terminating train.'
return 1
# Create a Machine Learning model
model = Model(is_crf=is_crf)
# Train the model using the Note's data
model.train(notes, grid)
# Pickle dump
print 'pickle dump'
with open(model_path, "wb") as m_file:
pickle.dump(model, m_file)
def find_note(self, name):
note = None
aux = self.db.get_note(name)
self.db.close_db()
print type(aux), aux
if aux:
note = Note(aux[0][1], aux[0][3])
note.id = aux[0][0]
note.creation_date = aux[0][2]
return note
return note
def train(training_list,
model_path,
format,
is_crf=True,
grid=False,
third=False,
disambiguate=False):
"""
train()
Purpose: Train a model for given clinical data.
@param training_list list of (txt,con) file path tuples (training instances)
@param model_path string filename of where to pickle model object
@param format concept file data format (ex. i2b2, semeval)
@param is_crf whether first pass should use CRF classifier
@param grid whether second pass should perform grid search
@param third whether to perform third/clustering pass
"""
# Read the data into a Note object
notes = []
for txt, con in training_list:
note_tmp = Note(format) # Create Note
note_tmp.read(txt, con) # Read data into Note
notes.append(note_tmp) # Add the Note to the list
# file names
if not notes:
print 'Error: Cannot train on 0 files. Terminating train.'
return 1
# Create a Machine Learning model
model = Model(is_crf=is_crf)
# disambiguation
if format == "semeval" and disambiguate is True and enabled["UMLS"] != None:
model.set_cui_freq(cui_disambiguation.calcFreqOfCuis(training_list))
# Train the model using the Note's data
model.train(notes, grid, do_third=third)
# Pickle dump
print '\nserializing model to %s\n' % model_path
with open(model_path, "wb") as m_file:
pickle.dump(model, m_file)
# return trained model
return model
def predict(files, model_path, output_dir, format):
# Must specify output format
if format not in Note.supportedFormats():
print >>sys.stderr, '\n\tError: Must specify output format'
print >>sys.stderr, '\tAvailable formats: ', ' | '.join(Note.supportedFormats())
print >>sys.stderr, ''
exit(1)
# Load model
model = Model.load(model_path)
# Tell user if not predicting
if not files:
print >>sys.stderr, "\n\tNote: You did not supply any input files\n"
exit()
# For each file, predict concept labels
n = len(files)
for i,txt in enumerate(sorted(files)):
# Read the data into a Note object
note = Note(format)
note.read(txt)
print '-' * 30
print '\n\t%d of %d' % (i+1,n)
print '\t', txt, '\n'
# Predict concept labels
labels = model.predict(note)
# Get predictions in proper format
extension = note.getExtension()
output = note.write(labels)
#print output
# Output file
fname = os.path.splitext(os.path.basename(txt))[0] + '.' + extension
out_path = os.path.join(output_dir, fname)
# Output the concept predictions
print '\n\nwriting to: ', out_path
with open(out_path, 'w') as f:
print >>f, output
print
def train(training_list, model_path, format, is_crf=True, grid=False):
# Read the data into a Note object
notes = []
for txt, con in training_list:
note_tmp = Note(format) # Create Note
note_tmp.read(txt, con) # Read data into Note
notes.append(note_tmp) # Add the Note to the list
# file names
if not notes:
print 'Error: Cannot train on 0 files. Terminating train.'
return 1
# Create a Machine Learning model
model = Model(is_crf=is_crf)
# Train the model using the Note's data
model.train(notes, grid)
# Pickle dump
print 'pickle dump'
with open(model_path, "wb") as m_file:
pickle.dump(model, m_file)
def main():
parser = argparse.ArgumentParser()
parser.add_argument("-i",
dest = "input",
help = "The input files to predict",
default = os.path.join(os.getenv('CLINER_DIR'), 'data/test_data/*')
)
parser.add_argument("-o",
dest = "output",
help = "The directory to write the output",
default = os.path.join(os.getenv('CLINER_DIR'), 'data/test_predictions')
)
parser.add_argument("-m",
dest = "model",
help = "The model to use for prediction",
default = os.path.join(os.getenv('CLINER_DIR'), 'models/run.model')
)
parser.add_argument("-f",
dest = "format",
help = "Data format ( " + ' | '.join(Note.supportedFormats()) + " )",
default = 'i2b2'
)
parser.add_argument("-crf",
dest = "with_crf",
help = "Specify where to find crfsuite",
default = None
)
args = parser.parse_args()
# Parse arguments
files = glob.glob(args.input)
helper.mkpath(args.output)
format = args.format
# Predict
predict(files, args.model, args.output, format=format)
def main():
parser = argparse.ArgumentParser()
parser.add_argument("-t",
dest = "txt",
help = "The files that contain the training examples",
default = os.path.join(os.getenv('CLINER_DIR'), 'data/train/txt/*')
)
parser.add_argument("-c",
dest = "con",
help = "The files that contain the labels for the training examples",
default = os.path.join(os.getenv('CLINER_DIR'), 'data/train/con/*')
)
parser.add_argument("-m",
dest = "model",
help = "Path to the model that should be generated",
default = os.path.join(os.getenv('CLINER_DIR'), 'models/run.model')
)
parser.add_argument("-f",
dest = "format",
help = "Data format ( " + ' | '.join(Note.supportedFormats()) + " )",
default = 'i2b2'
)
parser.add_argument("-g",
dest = "grid",
help = "A flag indicating whether to perform a grid search",
action = "store_true"
)
parser.add_argument("-no-crf",
dest = "nocrf",
help = "A flag indicating whether to use crfsuite for pass one.",
action = "store_true"
)
# Parse the command line arguments
args = parser.parse_args()
is_crf = not args.nocrf
# A list of text file paths
# A list of concept file paths
txt_files = glob.glob(args.txt)
con_files = glob.glob(args.con)
# data format
format = args.format
# Must specify output format
if format not in Note.supportedFormats():
print >>sys.stderr, '\n\tError: Must specify output format'
print >>sys.stderr, '\tAvailable formats: ', ' | '.join(Note.supportedFormats())
print >>sys.stderr, ''
exit(1)
# Collect training data file paths
txt_files_map = helper.map_files(txt_files) # ex. {'record-13': 'record-13.con'}
con_files_map = helper.map_files(con_files)
training_list = [] # ex. training_list = [ ('record-13.txt', 'record-13.con') ]
for k in txt_files_map:
if k in con_files_map:
training_list.append((txt_files_map[k], con_files_map[k]))
# display file names (for user to see data was properly located)
print '\n', training_list, '\n'
# Train the model
train(training_list, args.model, format, is_crf=is_crf, grid=args.grid)
def main():
# Argument Parser
parser = argparse.ArgumentParser()
parser.add_argument(
"-txt",
dest="txt",
help="The files that contain the training examples",
)
parser.add_argument(
"-annotations",
dest="annotations",
help="The files that contain the labels for the training examples",
)
parser.add_argument(
"-out",
dest="out",
default=None,
help="Directory to output data",
)
parser.add_argument(
"-format",
dest="format",
help="Output format (%s)" % str(' or '.join(Note.supportedFormats())),
)
# Parse the command line arguments
args = parser.parse_args()
# Parse arguments
txt = args.txt
annotations = args.annotations
out_file = args.out
format = args.format
# Ensure annotations are specified
if not txt:
print >> sys.stderr, '\n\tError: Must supply text file'
print >> sys.stderr
exit(1)
elif not os.path.exists(txt):
print >> sys.stderr, '\n\tError: Given text file does not exist'
print >> sys.stderr
exit(1)
# Ensure annotations are specified
extensions = Note.supportedFormatExtensions()
if not annotations:
print >> sys.stderr, '\n\tError: Must supply annotations'
print >> sys.stderr
exit(2)
elif not os.path.exists(txt):
print >> sys.stderr, '\n\tError: Given annotation file does not exist'
print >> sys.stderr
exit(2)
elif os.path.splitext(annotations)[1][1:] not in extensions:
print >> sys.stderr, '\n\tError: annotation must be a supported format'
print >> sys.stderr, '\t\t(.%s)' % str(' or .'.join(extensions))
print >> sys.stderr
exit(2)
# Ensure output format is specified
if (not format) or (format not in Note.supportedFormats()):
print >> sys.stderr, '\n\tError: Must specify supported output format'
print >> sys.stderr, '\t\t(%s)' % str(' or '.join(
Note.supportedFormats()))
print >> sys.stderr
exit(3)
# Automatically find the input file format
in_extension = os.path.splitext(annotations)[1][1:]
for f, ext in Note.dictOfFormatToExtensions().items():
if ext == in_extension:
in_format = f
# Read input data into note object
in_note = Note(in_format)
in_note.read(txt, annotations)
# Convert data to standard format
internal_output = in_note.write_standard()
os_handle, tmp_file = tempfile.mkstemp(dir=tmp_dir, suffix="format_temp")
with open(tmp_file, 'w') as f:
f.write(internal_output)
os.close(os_handle)
#print internal_output
# Read internal standard data into new file with given output format
out_note = Note(format)
out_note.read_standard(txt, tmp_file)
# Output data
out = out_note.write()
if out_file:
with open(out_file, 'w') as out_f:
out_f.write(out)
else:
sys.stdout.write(out)
# Clean up
os.remove(tmp_file)
if out_file:
out_f.close()
def main():
parser = argparse.ArgumentParser()
parser.add_argument(
"-t",
help="Text files that were used to generate predictions",
dest="txt",
default=os.path.join(os.getenv('CLINER_DIR'), 'data/test_data/*'))
parser.add_argument(
"-c",
help=
"The directory that contains predicted concept files organized into subdirectories for svm, lin, srf",
dest="con",
default=os.path.join(os.getenv('CLINER_DIR'), 'data/predictions/'))
parser.add_argument(
"-r",
help=
"The directory that contains reference gold standard concept files",
dest="ref",
default=os.path.join(
os.getenv('CLINER_DIR'),
'data/reference_standard_for_test_data/concepts/'))
parser.add_argument(
"-f",
dest="format",
help="Data format ( " + ' | '.join(Note.supportedFormats()) + " )",
)
parser.add_argument(
"-o",
help="Write the evaluation to a file rather than STDOUT",
dest="output",
default=None)
# Parse command line arguments
args = parser.parse_args()
if args.format:
format = args.format
else:
print '\n\tERROR: must provide "format" argument\n'
exit()
# Is output destination specified?
if args.output:
args.output = open(args.output, "w")
else:
args.output = sys.stdout
# Must specify output format
if format not in Note.supportedFormats():
print >> sys.stderr, '\n\tError: Must specify output format'
print >> sys.stderr, '\tAvailable formats: ', ' | '.join(
Note.supportedFormats())
print >> sys.stderr, ''
exit(1)
# List of medical text
txt_files = glob.glob(args.txt)
txt_files_map = helper.map_files(txt_files)
wildcard = '*.' + Note.dictOfFormatToExtensions()[format]
# List of gold data
ref_files = glob.glob(os.path.join(args.ref, wildcard))
ref_files_map = helper.map_files(ref_files)
# List of predictions
pred_files = glob.glob(os.path.join(args.con, wildcard))
pred_files_map = helper.map_files(pred_files)
# Grouping of text, predictions, gold
files = []
for k in txt_files_map:
if k in pred_files_map and k in ref_files_map:
files.append(
(txt_files_map[k], pred_files_map[k], ref_files_map[k]))
# txt <- medical text
# annotations <- predictions
# gold <- gold standard
truePositivesExactSpan = 0
falseNegativesExactSpan = 0
falsePositivesExactSpan = 0
truePositivesInexactSpan = 0
falseNegativesInexactSpan = 0
falsePositivesInexactSpan = 0
confusion = [[0] * len(labels) for e in labels]
confusionMatrixExactSpan = deepcopy(confusion)
confusionMatrixInexactSpan = deepcopy(confusion)
if len(files) == 0:
exit("No files to be evaluated")
for txt, annotations, gold in files:
# Read predictions and gols standard data
cnote = Note(format)
rnote = Note(format)
cnote.read(txt, annotations)
rnote.read(txt, gold)
referenceSpans = getConceptSpans(rnote.getIOBLabels(), rnote.conlist())
predictedSpans = getConceptSpans(cnote.getIOBLabels(), cnote.conlist())
#TO DO: i need to generate a cumulative total accross all of the files
#modify my functions slightly and have it return the number of true positive and etc...
#then call generate results
exactResults = evaluate(deepcopy(referenceSpans),
deepcopy(predictedSpans),
exactMatch=True,
reportSeperately=False)
inexactResults = evaluate(deepcopy(referenceSpans),
deepcopy(predictedSpans),
exactMatch=False,
reportSeperately=False)
truePositivesExactSpan += exactResults["True Positives"]
falseNegativesExactSpan += exactResults["False Negatives"]
falsePositivesExactSpan += exactResults["False Positives"]
inexactResults = evaluate(deepcopy(referenceSpans),
deepcopy(predictedSpans),
exactMatch=False,
reportSeperately=False)
truePositivesInexactSpan += inexactResults["True Positives"]
falseNegativesInexactSpan += inexactResults["False Negatives"]
falsePositivesInexactSpan += inexactResults["False Positives"]
MatrixInexactSpan = evaluate(deepcopy(referenceSpans),
deepcopy(predictedSpans),
exactMatch=False,
reportSeperately=True)
for sublist1, sublist2 in zip(confusionMatrixInexactSpan,
MatrixInexactSpan):
for i, int2 in enumerate(sublist2):
sublist1[i] += int2
MatrixExactSpan = evaluate(deepcopy(referenceSpans),
deepcopy(predictedSpans),
exactMatch=True,
reportSeperately=True)
for sublist1, sublist2 in zip(confusionMatrixExactSpan,
MatrixExactSpan):
for i, int2 in enumerate(sublist2):
sublist1[i] += int2
print "\nResults for exact span for concepts together.\n"
print "True Positives: ", truePositivesExactSpan
print "False Negatives: ", falseNegativesExactSpan
print "False Positives: ", falsePositivesExactSpan
exactSpan = generateResultsForExactSpans(truePositivesExactSpan,
falseNegativesExactSpan,
falsePositivesExactSpan)
print "Recall: ", exactSpan["Recall"]
print "Precision: ", exactSpan["Precision"]
print "F Measure: ", exactSpan["F Score"]
inexactSpan = generateResultsForExactSpans(truePositivesInexactSpan,
falseNegativesInexactSpan,
falsePositivesInexactSpan)
print "\nResults for inexact span for concepts together.\n"
print "True Positives: ", truePositivesInexactSpan
print "False Negatives: ", falseNegativesInexactSpan
print "False Positives: ", falsePositivesInexactSpan
print "Recall: ", inexactSpan["Recall"]
print "Precision: ", inexactSpan["Precision"]
print "F Measure: ", inexactSpan["F Score"]
#TO DO: ENSURE NUMBER OF FP,FN,TP is equal to number of predicted spans
#TO DO: number of FP, FN, TP is not same between exact and inexact.
#LEFT OFF HERE. FIX DISPLAY FUNCTION
displayMatrix(args.output, 'Exact', confusionMatrixExactSpan)
displayMatrix(args.output, 'Inexact', confusionMatrixInexactSpan)
#print evaluate(deepcopy(referenceSpans), deepcopy(predictedSpans), exactMatch=False, reportSeperately=True)
return
def main():
parser = argparse.ArgumentParser()
parser.add_argument(
"-t",
dest="txt",
help="The files that contain the training examples",
)
parser.add_argument(
"-c",
dest="con",
help="The files that contain the labels for the training examples",
)
parser.add_argument(
"-m",
dest="model",
help="Path to the model that should be generated",
)
parser.add_argument(
"-f",
dest="format",
help="Data format ( " + ' | '.join(Note.supportedFormats()) + " )",
)
parser.add_argument(
"-g",
dest="grid",
help="A flag indicating whether to perform a grid search",
action="store_true")
parser.add_argument(
"-no-crf",
dest="nocrf",
help="A flag indicating whether to use crfsuite for pass one.",
action="store_true")
parser.add_argument(
"-discontiguous_spans",
dest="third",
help="A flag indicating whether to have third/clustering pass",
action="store_true")
parser.add_argument(
"-umls_disambiguation",
dest="umls_disambiguation",
action="store_true",
help=
"A flag indicating wheter to disambiguate CUI id for detected entities in semeval format",
)
"""
parser.add_argument("-unlabeled",
dest = "unlabeled",
help = "Path to dir containing unlabelled data used for unsupervised methods",
)
"""
# Parse the command line arguments
args = parser.parse_args()
is_crf = not args.nocrf
third = args.third
# Error check: Ensure that file paths are specified
if not args.txt:
print >> sys.stderr, '\n\tError: Must provide text files'
print >> sys.stderr, ''
exit(1)
if not args.con:
print >> sys.stderr, '\n\tError: Must provide annotations for text files'
print >> sys.stderr, ''
exit(1)
if not args.model:
print >> sys.stderr, '\n\tError: Must provide valid path to store model'
print >> sys.stderr, ''
exit(1)
modeldir = os.path.dirname(args.model)
if (not os.path.exists(modeldir)) and (modeldir != ''):
print >> sys.stderr, '\n\tError: Model dir does not exist: %s' % modeldir
print >> sys.stderr, ''
exit(1)
if "PY4J_DIR_PATH" not in os.environ and args.third is True:
exit(
"please set environ var PY4J_DIR_PATH to the dir of the folder containg py4j<version>.jar"
)
# A list of text file paths
# A list of concept file paths
txt_files = glob.glob(args.txt)
con_files = glob.glob(args.con)
# data format
if args.format:
format = args.format
else:
print '\n\tERROR: must provide "format" argument\n'
exit()
if third is True and args.format == "i2b2":
exit("i2b2 formatting does not support disjoint spans")
# Must specify output format
if format not in Note.supportedFormats():
print >> sys.stderr, '\n\tError: Must specify output format'
print >> sys.stderr, '\tAvailable formats: ', ' | '.join(
Note.supportedFormats())
print >> sys.stderr, ''
exit(1)
# Collect training data file paths
txt_files_map = helper.map_files(
txt_files) # ex. {'record-13': 'record-13.con'}
con_files_map = helper.map_files(con_files)
training_list = [
] # ex. training_list = [ ('record-13.txt', 'record-13.con') ]
for k in txt_files_map:
if k in con_files_map:
training_list.append((txt_files_map[k], con_files_map[k]))
# Train the model
train(training_list,
args.model,
format,
is_crf=is_crf,
grid=args.grid,
third=third,
disambiguate=args.umls_disambiguation)
def main():
parser = argparse.ArgumentParser()
parser.add_argument("-t",
help = "Text files that were used to generate predictions",
dest = "txt",
default = os.path.join(os.getenv('CLINER_DIR'), 'data/test_data/*')
)
parser.add_argument("-c",
help = "The directory that contains predicted concept files organized into subdirectories for svm, lin, srf",
dest = "con",
default = os.path.join(os.getenv('CLINER_DIR'), 'data/predictions/')
)
parser.add_argument("-r",
help = "The directory that contains reference gold standard concept files",
dest = "ref",
default = os.path.join(os.getenv('CLINER_DIR'), 'data/reference_standard_for_test_data/concepts/')
)
parser.add_argument("-f",
dest = "format",
help = "Data format ( " + ' | '.join(Note.supportedFormats()) + " )",
default = 'i2b2'
)
parser.add_argument("-o",
help = "Write the evaluation to a file rather than STDOUT",
dest = "output",
default = None
)
# Parse command line arguments
args = parser.parse_args()
format = args.format
# Is output destination specified?
if args.output:
args.output = open(args.output, "w")
else:
args.output = sys.stdout
# Must specify output format
if format not in Note.supportedFormats():
print >>sys.stderr, '\n\tError: Must specify output format'
print >>sys.stderr, '\tAvailable formats: ', ' | '.join(Note.supportedFormats())
print >>sys.stderr, ''
exit(1)
# List of medical text
txt_files = glob.glob(args.txt)
txt_files_map = helper.map_files(txt_files)
wildcard = '*.' + Note.dictOfFormatToExtensions()[format]
# List of gold data
ref_files = glob.glob( os.path.join(args.ref, wildcard) )
ref_files_map = helper.map_files(ref_files)
# List of predictions
pred_files = glob.glob( os.path.join(args.con, wildcard) )
pred_files_map = helper.map_files(pred_files)
# Grouping of text, predictions, gold
files = []
for k in txt_files_map:
if k in pred_files_map and k in ref_files_map:
files.append((txt_files_map[k], pred_files_map[k], ref_files_map[k]))
# txt <- medical text
# annotations <- predictions
# gold <- gold standard
truePositivesExactSpan = 0
falseNegativesExactSpan = 0
falsePositivesExactSpan = 0
truePositivesInexactSpan = 0
falseNegativesInexactSpan = 0
falsePositivesInexactSpan = 0
confusion = [[0] * len(labels) for e in labels]
confusionMatrixExactSpan = deepcopy(confusion)
confusionMatrixInexactSpan = deepcopy(confusion)
for txt, annotations, gold in files:
# Read predictions and gols standard data
cnote = Note(format)
rnote = Note(format)
cnote.read(txt, annotations)
rnote.read(txt, gold)
referenceSpans = getConceptSpans(rnote.getIOBLabels(), rnote.conlist())
predictedSpans = getConceptSpans(cnote.getIOBLabels(), cnote.conlist())
#TO DO: i need to generate a cumulative total accross all of the files
#modify my functions slightly and have it return the number of true positive and etc...
#then call generate results
exactResults = evaluate(deepcopy(referenceSpans), deepcopy(predictedSpans), exactMatch=True, reportSeperately=False)
inexactResults = evaluate(deepcopy(referenceSpans), deepcopy(predictedSpans), exactMatch=False, reportSeperately=False)
truePositivesExactSpan += exactResults["True Positives"]
falseNegativesExactSpan += exactResults["False Negatives"]
falsePositivesExactSpan += exactResults["False Positives"]
inexactResults = evaluate(deepcopy(referenceSpans), deepcopy(predictedSpans), exactMatch=False, reportSeperately=False)
truePositivesInexactSpan += inexactResults["True Positives"]
falseNegativesInexactSpan += inexactResults["False Negatives"]
falsePositivesInexactSpan += inexactResults["False Positives"]
MatrixInexactSpan = evaluate(deepcopy(referenceSpans), deepcopy(predictedSpans), exactMatch=False, reportSeperately=True)
for sublist1, sublist2 in zip(confusionMatrixInexactSpan, MatrixInexactSpan):
for i,int2 in enumerate(sublist2):
sublist1[i] += int2
MatrixExactSpan = evaluate(deepcopy(referenceSpans), deepcopy(predictedSpans), exactMatch=True, reportSeperately=True)
for sublist1, sublist2 in zip(confusionMatrixExactSpan, MatrixExactSpan):
for i,int2 in enumerate(sublist2):
sublist1[i] += int2
print "\nResults for exact span for concepts together.\n"
print "True Positives: ", truePositivesExactSpan
print "False Negatives: ", falseNegativesExactSpan
print "False Positives: ", falsePositivesExactSpan
exactSpan = generateResultsForExactSpans(truePositivesExactSpan,
falseNegativesExactSpan,
falsePositivesExactSpan)
print "Recall: ", exactSpan["Recall"]
print "Precision: ", exactSpan["Precision"]
print "F Measure: ", exactSpan["F Score"]
inexactSpan = generateResultsForExactSpans(truePositivesInexactSpan,
falseNegativesInexactSpan,
falsePositivesInexactSpan)
print "\nResults for inexact span for concepts together.\n"
print "True Positives: ", truePositivesInexactSpan
print "False Negatives: ", falseNegativesInexactSpan
print "False Positives: ", falsePositivesInexactSpan
print "Recall: ", inexactSpan["Recall"]
print "Precision: ", inexactSpan["Precision"]
print "F Measure: ", inexactSpan["F Score"]
#TO DO: ENSURE NUMBER OF FP,FN,TP is equal to number of predicted spans
#TO DO: number of FP, FN, TP is not same between exact and inexact.
#LEFT OFF HERE. FIX DISPLAY FUNCTION
displayMatrix(args.output, 'Exact' , confusionMatrixExactSpan)
displayMatrix(args.output, 'Inexact', confusionMatrixInexactSpan)
#print evaluate(deepcopy(referenceSpans), deepcopy(predictedSpans), exactMatch=False, reportSeperately=True)
return
def main():
parser = argparse.ArgumentParser()
parser.add_argument("-t",
dest="txt",
help="The files that contain the training examples",
default=os.path.join(os.getenv('CLINER_DIR'),
'data/train/txt/*'))
parser.add_argument(
"-c",
dest="con",
help="The files that contain the labels for the training examples",
default=os.path.join(os.getenv('CLINER_DIR'), 'data/train/con/*'))
parser.add_argument("-m",
dest="model",
help="Path to the model that should be generated",
default=os.path.join(os.getenv('CLINER_DIR'),
'models/run.model'))
parser.add_argument("-f",
dest="format",
help="Data format ( " +
' | '.join(Note.supportedFormats()) + " )",
default='i2b2')
parser.add_argument(
"-g",
dest="grid",
help="A flag indicating whether to perform a grid search",
action="store_true")
parser.add_argument(
"-no-crf",
dest="nocrf",
help="A flag indicating whether to use crfsuite for pass one.",
action="store_true")
# Parse the command line arguments
args = parser.parse_args()
is_crf = not args.nocrf
# A list of text file paths
# A list of concept file paths
txt_files = glob.glob(args.txt)
con_files = glob.glob(args.con)
# data format
format = args.format
# Must specify output format
if format not in Note.supportedFormats():
print >> sys.stderr, '\n\tError: Must specify output format'
print >> sys.stderr, '\tAvailable formats: ', ' | '.join(
Note.supportedFormats())
print >> sys.stderr, ''
exit(1)
# Collect training data file paths
txt_files_map = helper.map_files(
txt_files) # ex. {'record-13': 'record-13.con'}
con_files_map = helper.map_files(con_files)
training_list = [
] # ex. training_list = [ ('record-13.txt', 'record-13.con') ]
for k in txt_files_map:
if k in con_files_map:
training_list.append((txt_files_map[k], con_files_map[k]))
# display file names (for user to see data was properly located)
print '\n', training_list, '\n'
# Train the model
train(training_list, args.model, format, is_crf=is_crf, grid=args.grid)
def main():
# Argument Parser
parser = argparse.ArgumentParser()
parser.add_argument("-t",
dest = "txt",
help = "The files that contain the training examples",
)
parser.add_argument("-a",
dest = "annotations",
help = "The files that contain the labels for the training examples",
)
parser.add_argument("-o",
dest = "out",
default = None,
help = "Directory to output data",
)
parser.add_argument("-f",
dest = "format",
help = "Output format (%s)"%str(' or '.join(Note.supportedFormats())),
)
# Parse the command line arguments
args = parser.parse_args()
# Parse arguments
txt = args.txt
annotations = args.annotations
out_file = args.out
format = args.format
# Ensure annotations are specified
if not txt:
print >>sys.stderr, '\n\tError: Must supply text file'
print >>sys.stderr
exit(1)
elif not os.path.exists(txt):
print >>sys.stderr, '\n\tError: Given text file does not exist'
print >>sys.stderr
exit(1)
# Ensure annotations are specified
extensions = Note.supportedFormatExtensions()
if not annotations:
print >>sys.stderr, '\n\tError: Must supply annotations'
print >>sys.stderr
exit(2)
elif not os.path.exists(txt):
print >>sys.stderr, '\n\tError: Given annotation file does not exist'
print >>sys.stderr
exit(2)
elif os.path.splitext(annotations)[1][1:] not in extensions:
print >>sys.stderr, '\n\tError: annotation must be a supported format'
print >>sys.stderr, '\t\t(.%s)' %str(' or .'.join(extensions) )
print >>sys.stderr
exit(2)
# Ensure output format is specified
if (not format) or (format not in Note.supportedFormats()):
print >>sys.stderr, '\n\tError: Must specify supported output format'
print >>sys.stderr, '\t\t(%s)' %str(' or '.join(Note.supportedFormats()))
print >>sys.stderr
exit(3)
# Automatically find the input file format
in_extension = os.path.splitext(annotations)[1][1:]
for f,ext in Note.dictOfFormatToExtensions().items():
if ext == in_extension:
in_format = f
# Read input data into note object
in_note = Note(in_format)
in_note.read(txt,annotations)
# Convert data to standard format
internal_output = in_note.write_standard()
os_handle,tmp_file = tempfile.mkstemp(dir=tmp_dir, suffix="format_temp")
with open(tmp_file, 'w') as f:
f.write(internal_output)
os.close(os_handle)
#print internal_output
# Read internal standard data into new file with given output format
out_note = Note(format)
out_note.read_standard(txt,tmp_file)
# Output data
out = out_note.write()
if out_file:
with open(out_file, 'w') as out_f:
out_f.write(out)
else:
sys.stdout.write(out)
# Clean up
os.remove(tmp_file)
if out_file:
out_f.close()
def main():
parser = argparse.ArgumentParser()
parser.add_argument(
"-i",
dest="input",
help="The input files to predict",
)
parser.add_argument(
"-o",
dest="output",
help="The directory to write the output",
)
parser.add_argument(
"-m",
dest="model",
help="The model to use for prediction",
)
parser.add_argument(
"-f",
dest="format",
help="Data format ( " + ' | '.join(Note.supportedFormats()) + " )",
)
parser.add_argument("-crf",
dest="with_crf",
help="Specify where to find crfsuite",
default=None)
parser.add_argument(
"-discontiguous_spans",
dest="third",
help="A flag indicating whether to have third/clustering pass",
action="store_true")
parser.add_argument(
"-umls_disambiguation",
dest="disambiguate",
help=
"A flag indicating whether to disambiguate CUI ID for identified entities in semeval",
action="store_true")
args = parser.parse_args()
# Error check: Ensure that file paths are specified
if not args.input:
print >> sys.stderr, '\n\tError: Must provide text files\n'
exit(1)
if not args.output:
print >> sys.stderr, '\n\tError: Must provide output directory\n'
exit(1)
if not args.model:
print >> sys.stderr, '\n\tError: Must provide path to model\n'
exit(1)
if not os.path.exists(args.model):
print >> sys.stderr, '\n\tError: Model does not exist: %s\n' % args.model
exit(1)
# Parse arguments
files = glob.glob(args.input)
helper.mkpath(args.output)
third = args.third
if args.format:
format = args.format
else:
print '\n\tERROR: must provide "format" argument\n'
exit()
if third is True and args.format == "i2b2":
exit("i2b2 formatting does not support disjoint spans")
# Tell user if not predicting
if not files:
print >> sys.stderr, "\n\tNote: You did not supply any input files\n"
exit()
# Predict
predict(files,
args.model,
args.output,
format=format,
third=third,
disambiguate=args.disambiguate)
def predict(files,
model_path,
output_dir,
format,
third=False,
disambiguate=False):
# Must specify output format
if format not in Note.supportedFormats():
print >> sys.stderr, '\n\tError: Must specify output format'
print >> sys.stderr, '\tAvailable formats: ', ' | '.join(
Note.supportedFormats())
print >> sys.stderr, ''
exit(1)
# Load model
model = Model.load(model_path)
# Tell user if not predicting
if not files:
print >> sys.stderr, "\n\tNote: You did not supply any input files\n"
exit()
if enabled["UMLS"] is not None and disambiguate is True:
from disambiguation import cui_disambiguation
# For each file, predict concept labels
n = len(files)
for i, txt in enumerate(sorted(files)):
note = Note(format)
note.read(txt)
# Output file
extension = note.getExtension()
fname = os.path.splitext(os.path.basename(txt))[0] + '.' + extension
out_path = os.path.join(output_dir, fname)
#if os.path.exists(out_path):
# print '\tWARNING: prediction file already exists (%s)' % out_path
# continue
if format == "semevaL":
note.setFileName(os.path.split(txt)[-1])
# Predict concept labels
labels = model.predict(note, third)
# Get predictions in proper format
output = note.write(labels)
# TODO: make a flag to enable or disable looking up concept ids.
if format == "semeval":
print "\nencoding concept ids"
if enabled["UMLS"] is not None and disambiguate is True:
output = cui_disambiguation.disambiguate(
output, txt, model.get_cui_freq())
# Output the concept predictions
print '\n\nwriting to: ', out_path
with open(out_path, 'w') as f:
print >> f, output
print
