def analyze_classifiers(region_key, classifiers, x_train, y_train, x_test, y_test, out_file, preprocessor=text_preprocessor):
printers.printsf('{0}Analysis for {1} ear region{0}'.format(40*'-', region_key), out_file)
for key,value in classifiers.items():
clf = value[0] #the classifier
usa = value[1] #use spare array
ubf = value[2] #use binary features (this is to support NB)
parameters = value[3]
vectorizer = CountVectorizer(input='content', decode_error='ignore', preprocessor=preprocessor, binary=ubf)
pipeline = (Pipeline(steps=[('vect', vectorizer),('clf',clf)]) if usa
else Pipeline(steps=[('vect', vectorizer),('sa',sklx.SparseToArray()),('clf',clf)]))
gs = sklx.grid_analysis(pipeline,parameters, x_train, y_train)
printers.print_grid_search_results(gs,key,out_file,x_test,y_test)
#create empty patient array to hold predicted values
num_patients = len(test_labels)
patients = np.empty((num_patients,), dtype=[('pid','S7'),('inner','i4'),('middle','i4'),('outer','i4'),('mastoid','i4')])
#initialize patients array
for k in region_keys:
patients[k] = 0
#get patient values based on icd9 codes
cnt = 0
for _ , row in test_labels.iterrows():
pid = row['pid']
patients['pid'][cnt] = pid
report = test_reports[cnt]
for region in region_keys:
for keyword in keywords[region]:
if keyword in report: patients[region][cnt] = 1
cnt += 1
#compare predicted and actual
for k in region_keys:
printers.printsf('{0}Analysis for {1} ear region{0}'.format(40*'-', k), standard_out_file)
y_pred = patients[k]
y_act = test_labels[k]
pm = PerformanceMetrics(y_act, y_pred)
printers.printsfPerformanceMetrics(pm, standard_out_file)
cm = sklearn.metrics.confusion_matrix(y_act, y_pred)
printers.printTwoClassConfusion(cm, standard_out_file)
if __name__ == '__main__':
# training/test data and output files
label_file = 'data/input/SDS_PV2_combined/SDS_PV2_class_labels.txt'
report_path = 'data/input/SDS_PV2_combined/reports'
output_path = 'data/output/{0}'
label_data = pd.read_csv(label_file)
region_keys = label_data.columns[2:6]
standard_out_file = output_path.format('Model_Persistence_Report.txt')
if not os.path.exists(os.path.dirname(standard_out_file)):
os.makedirs(os.path.dirname(standard_out_file))
now = time.localtime()
(printers.printsf(
'{6}{0}-{1}-{2} {3}:{4}:{5}{6}'.format(now.tm_year, now.tm_mon,
now.tm_mday, now.tm_hour,
now.tm_min, now.tm_sec,
40 * '-'), standard_out_file,
'a', False))
# set the numpy random seed so results are reproducible
rs = RandomState(__seed__)
# partition the data
pos_cases, neg_cases = wrangle.partion(label_data['doc_norm'] == 1,
label_data,
ratios=[0.8, 0.2])
train_mask = np.concatenate((pos_cases[0], neg_cases[0]))
test_mask = np.concatenate((pos_cases[1], neg_cases[1]))
rs.shuffle(train_mask)
rs.shuffle(test_mask)
train_labels = label_data.iloc[train_mask]
(printers.
printsf('{6}{0}-{1}-{2} {3}:{4}:{5}{6}'.
format(now.tm_year, now.tm_mon, now.tm_mday, now.tm_hour, now.tm_min, now.tm_sec, 40*'-'),
standard_out_file, 'a',False))
# partition the data
pos_cases, neg_cases = wrangle.partion(label_data['doc_norm']==1, label_data, ratios=[0.8,0.2])
train_mask = np.concatenate((pos_cases[0], neg_cases[0]))
test_mask = np.concatenate((pos_cases[1], neg_cases[1]))
rs.shuffle(train_mask)
rs.shuffle(test_mask)
train_labels = label_data.iloc[train_mask]
test_labels = label_data.iloc[test_mask]
# print partition stats
printers.printsf('{0}Data Partition Stats{0}'.format(40*'-'), standard_out_file)
printers.print_data_stats(train_labels['doc_norm'], test_labels['doc_norm'], '{0}Document{0}'.format(40*'-'),standard_out_file)
for key in region_keys:
printers.print_data_stats(train_labels[key], test_labels[key],
'{0}{1}{0}'.format(40*'-', key),standard_out_file)
# read in the text reports
train_reports = [load_report('{0}/{1}{2}'.format(report_path, pid, file_suffix)) for pid in train_labels['pid']]
test_reports = [load_report('{0}/{1}{2}'.format(report_path, pid, file_suffix)) for pid in test_labels['pid']]
#------------------------------ BASELINE ANALYSIS -----------------------------------------------------------------
if analyze_baseline:
clf = linear_model.LogisticRegression(C=1000)
#clf = BernoulliNB(alpha=1.0, binarize=None, fit_prior=True, class_prior=None)
standard_out_file, 'a',False))
# set the numpy random seed so results are reproducible
rs = RandomState(__seed__)
# partition the data
pos_cases, neg_cases = wrangle.partion(label_data['doc_norm']==1, label_data, ratios=[0.8,0.2])
train_mask = np.concatenate((pos_cases[0], neg_cases[0]))
test_mask = np.concatenate((pos_cases[1], neg_cases[1]))
rs.shuffle(train_mask)
rs.shuffle(test_mask)
train_labels = label_data.iloc[train_mask]
test_labels = label_data.iloc[test_mask]
# print partition stats
printers.printsf('{0}Data Partition Stats{0}'.format(40*'-'), standard_out_file)
for key in region_keys:
printers.print_data_stats(train_labels[key], test_labels[key],
'{0}{1}{0}'.format(40*'-', key),standard_out_file)
# read in the text reports
print 'loading reports ..............'
train_reports = [load_report('{0}/{1}.txt'.format(report_path,pid)) for pid in train_labels['pid']]
test_reports = [load_report('{0}/{1}.txt'.format(report_path,pid)) for pid in test_labels['pid']]
print 'loading configuration file .........'
# load the configuration
config_path = 'resources/config/models.ini'
configuration = ConfigParser.ConfigParser()
configuration.read(config_path)