def buildDoc2VecModel():
reports = preprocess.getProcessedReports()
# construct sentences from reports
taggedDocuments = []
for i in range(len(reports)):
taggedDocument = gensim.models.doc2vec.TaggedDocument(words= reports[i], tags= [i])
taggedDocuments.append(taggedDocument)
# model = gensim.models.Doc2Vec(taggedDocuments)
model = gensim.models.Doc2Vec(size=300, min_count=5, workers=16,dm=1, dbow_words=1,negative=20)
model.build_vocab(taggedDocuments)
model.alpha = 0.025 # learning rate
for epoch in range(10):
print(epoch)
model.train(taggedDocuments)
model.alpha -= 0.001
model.min_alpha = model.alpha
model.save("./model_files/reports.doc2vec_model")
def buildDoc2VecModel():
reports = preprocess.getProcessedReports()
# construct sentences from reports
taggedDocuments = []
for i in range(len(reports)):
taggedDocument = gensim.models.doc2vec.TaggedDocument(words=reports[i],
tags=[i])
taggedDocuments.append(taggedDocument)
# model = gensim.models.Doc2Vec(taggedDocuments)
model = gensim.models.Doc2Vec(size=300,
min_count=5,
workers=16,
dm=1,
dbow_words=1,
negative=20)
model.build_vocab(taggedDocuments)
model.alpha = 0.025 # learning rate
for epoch in range(10):
print(epoch)
model.train(taggedDocuments)
model.alpha -= 0.001
model.min_alpha = model.alpha
model.save("./model_files/reports.doc2vec_model")
def buildDictionary(fileType):
reports = preprocess.getProcessedReports(fileType)
print("files loaded")
# build dictionary
dictionary = gensim.corpora.Dictionary(reports)
# dictionary.filter_extremes(no_below=3)
dictionary.save('../model_files/reports.dict')
print(dictionary)
print("dictionary finished")
# build corpus
corpus = [dictionary.doc2bow(report) for report in reports]
gensim.corpora.MmCorpus.serialize('../model_files/reports.mm', corpus)
# print(corpus)
print("corpus finished")
def buildWord2VecModel():
reports = preprocess.getProcessedReports()
model = gensim.models.Word2Vec(reports, min_count=3)
model.init_sims(replace=True)
model.save("./model_files/reports.word2vec_model")
print(model)
# model = gensim.models.Word2Vec.load("zzmodel")
print("----------------------------------similarity test")
print(model.similarity("head","brain"))
print("----------------------------------raw numpy vector of word")
print(model["age"])
print("----------------------------------remove outlier")
print(model.doesnt_match("hours four age".split()))
print("----------------------------------similar words")
print(model.most_similar("haem"))
print("script finished")
def buildDictionary():
reports = preprocess.getProcessedReports()
print("files loaded")
# build dictionary
dictionary = gensim.corpora.Dictionary(reports)
# dictionary.filter_extremes(no_below=3)
dictionary.save('./model_files/reports.dict')
print(dictionary)
print("dictionary finished")
# build corpus
corpus = [dictionary.doc2bow(report) for report in reports]
gensim.corpora.MmCorpus.serialize('./model_files/reports.mm', corpus)
# print(corpus)
print("corpus finished")
def buildWord2VecModel():
reports = preprocess.getProcessedReports()
model = gensim.models.Word2Vec(reports, min_count=3)
model.init_sims(replace=True)
model.save("./model_files/reports.word2vec_model")
print(model)
# model = gensim.models.Word2Vec.load("zzmodel")
print("----------------------------------similarity test")
print(model.similarity("head", "brain"))
print("----------------------------------raw numpy vector of word")
print(model["age"])
print("----------------------------------remove outlier")
print(model.doesnt_match("hours four age".split()))
print("----------------------------------similar words")
print(model.most_similar("haem"))
print("script finished")
def labelClassificationD2V():
model = gensim.models.Doc2Vec.load("./model_files/reports.doc2vec_model")
reports = preprocess.getReports()
processedReports = preprocess.getProcessedReports()
numFolds = 5 # number of folds for cross validation
directory = "label_classification/" + datetime.datetime.now().strftime(
'%m_%d_%H_%M') + "/"
if not os.path.exists(directory):
os.makedirs(directory)
with open(directory + "labelClassification.csv", 'w') as writeFile:
writer = csv.writer(writeFile)
writer.writerow(["score", "output label", "expected label", "report"])
for j in range(len(REPORT_FILES_LABELLED)):
writer.writerow("")
writer.writerow("")
writer.writerow([DIAGNOSES[j]])
# initialise figure and plot
name = DIAGNOSES[j] + " ROC"
plt.figure(name)
plt.xlabel("False Positive")
plt.ylabel("True Positive")
plt.title(DIAGNOSES[j] + " ROC")
# fetch corpus and labels
labelledReports = []
labelledCorpus = list()
# The labeled data is at the start of the data set
# Get the ids in the corpus of these first labeled examples for each class
for i in range(
preprocess.getNumReports(REPORT_FILES[:j]),
preprocess.getNumReports(REPORT_FILES[:j]) +
preprocess.getNumReports([REPORT_FILES_LABELLED[j]])):
labelledReports.append(reports[i])
labelledCorpus.append(model.infer_vector(processedReports[i]))
labels = np.asarray(preprocess.getData([REPORT_FILES_LABELLED[j]
]))[:, 2]
corpusList = [list(x) for x in labelledCorpus]
############### THIS CODE BLOCK REMOVES THE NUMBER OF NEGATIVE LABELS TO EQUALISE THE DISTRIBUTION OF CLASS LABELS. TO BE REMOVED IN FUTURE.
count = 0
deletes = []
for x in range(len(labels)):
if (labels[x] == "negative"):
count = count + 1
deletes.append(x)
if (count == (len(labels) -
(list(labels).count("positive")) * 2)):
break
labelledCorpus = np.delete(labelledCorpus, deletes, axis=0)
labels = np.delete(labels, deletes)
##################
numData = len(labels) # size of the labelled data set
dataPerFold = int(math.ceil(numData / numFolds))
for n in range(0, numFolds):
# split training and test data
train_labelledCorpus, test_labelledCorpus, train_labels, test_labels = train_test_split(
labelledCorpus, labels, test_size=0.13)
# build classifier
classifier = svm.SVC(kernel='linear').fit(
train_labelledCorpus, train_labels)
# compute output label and corresponding score
output_test = classifier.predict(test_labelledCorpus)
output_train = classifier.predict(train_labelledCorpus)
output_scores_test = classifier.decision_function(
test_labelledCorpus)
output_scores_train = classifier.decision_function(
train_labelledCorpus)
# sort scores and labels in order
sortList = list(
zip(output_scores_test, output_test, test_labels,
test_labelledCorpus))
sortList.sort()
output_scores_test, output_test, test_labels, test_labelledCorpus = zip(
*sortList)
# build roc curve and plot
fp_test, tp_test, _ = roc_curve(test_labels,
output_scores_test,
pos_label="positive")
fp_train, tp_train, _ = roc_curve(train_labels,
output_scores_train,
pos_label="positive")
plt.plot(fp_test,
tp_test,
'r',
label="train" if n == 0 else "")
plt.plot(fp_train,
tp_train,
'b',
label="test" if n == 0 else "")
plt.legend(loc='lower right')
plt.savefig(directory + name)
# save result to file
for r in range(len(test_labels)):
reportIdx = corpusList.index(list(test_labelledCorpus[r]))
writer.writerow("")
writer.writerow([
output_scores_test[r], output_test[r], test_labels[r]
])
writer.writerow([labelledReports[reportIdx]])
# plt.show()
writeFile.close()
def labelClassificationD2V():
model = gensim.models.Doc2Vec.load("./model_files/reports.doc2vec_model")
reports = preprocess.getReports()
processedReports = preprocess.getProcessedReports()
numFolds = 5 # number of folds for cross validation
directory = "label_classification/" + datetime.datetime.now().strftime('%m_%d_%H_%M') +"/"
if not os.path.exists(directory):
os.makedirs(directory)
with open(directory+"labelClassification.csv",'w') as writeFile:
writer = csv.writer(writeFile)
writer.writerow(["score","output label","expected label","report"])
for j in range(len(REPORT_FILES_LABELLED)):
writer.writerow("")
writer.writerow("")
writer.writerow([DIAGNOSES[j]])
# initialise figure and plot
name = DIAGNOSES[j] + " ROC"
plt.figure(name)
plt.xlabel("False Positive")
plt.ylabel("True Positive")
plt.title(DIAGNOSES[j] + " ROC")
# fetch corpus and labels
labelledReports = []
labelledCorpus = list()
# The labeled data is at the start of the data set
# Get the ids in the corpus of these first labeled examples for each class
for i in range(preprocess.getNumReports(REPORT_FILES[:j]),preprocess.getNumReports(REPORT_FILES[:j])+preprocess.getNumReports([REPORT_FILES_LABELLED[j]])):
labelledReports.append(reports[i])
labelledCorpus.append(model.infer_vector(processedReports[i]))
labels = np.asarray(preprocess.getData([REPORT_FILES_LABELLED[j]]))[:,2]
corpusList = [list(x) for x in labelledCorpus]
############### THIS CODE BLOCK REMOVES THE NUMBER OF NEGATIVE LABELS TO EQUALISE THE DISTRIBUTION OF CLASS LABELS. TO BE REMOVED IN FUTURE.
count = 0
deletes = []
for x in range(len(labels)):
if (labels[x] == "negative"):
count = count + 1
deletes.append(x)
if (count == (len(labels)-(list(labels).count("positive"))*2)):
break
labelledCorpus = np.delete(labelledCorpus,deletes,axis=0)
labels = np.delete(labels,deletes)
##################
numData = len(labels) # size of the labelled data set
dataPerFold = int(math.ceil(numData/numFolds))
for n in range(0,numFolds):
# split training and test data
train_labelledCorpus,test_labelledCorpus,train_labels,test_labels = train_test_split(labelledCorpus,labels,test_size=0.13)
# build classifier
classifier = svm.SVC(kernel='linear').fit(train_labelledCorpus,train_labels)
# compute output label and corresponding score
output_test = classifier.predict(test_labelledCorpus)
output_train = classifier.predict(train_labelledCorpus)
output_scores_test = classifier.decision_function(test_labelledCorpus)
output_scores_train = classifier.decision_function(train_labelledCorpus)
# sort scores and labels in order
sortList = list(zip(output_scores_test,output_test,test_labels,test_labelledCorpus))
sortList.sort()
output_scores_test,output_test,test_labels,test_labelledCorpus = zip(*sortList)
# build roc curve and plot
fp_test,tp_test,_ = roc_curve(test_labels,output_scores_test,pos_label="positive")
fp_train,tp_train,_ = roc_curve(train_labels,output_scores_train,pos_label="positive")
plt.plot(fp_test,tp_test,'r',label="train" if n == 0 else "")
plt.plot(fp_train,tp_train,'b',label="test" if n == 0 else "")
plt.legend(loc='lower right')
plt.savefig(directory+name)
# save result to file
for r in range(len(test_labels)):
reportIdx = corpusList.index(list(test_labelledCorpus[r]))
writer.writerow("")
writer.writerow([output_scores_test[r],output_test[r],test_labels[r]])
writer.writerow([labelledReports[reportIdx]])
# plt.show()
writeFile.close()
