import multiprocessing
if __name__ == '__main__':
multiprocessing.set_start_method('forkserver')
import sys
from sklearn.feature_selection import f_classif
# from sklearn.linear_model import LogisticRegression
from p_values_for_logreg import LogisticReg
from sklearn.preprocessing import StandardScaler
import pdb
import numpy as np
from read_file import read_file
dataset = sys.argv[1]
save_file = sys.argv[2]
random_state = int(sys.argv[3])
rare = eval(sys.argv[4])
longitudinal=False
features, labels, pt_ids, feature_names, zfile = read_file(dataset,longitudinal,rare)
scores=[]
# scale the data
features = StandardScaler().fit_transform(features)
# create the classifier
for i in np.arange(features.shape[1]):
print('fitting feature',feature_names[i],i,'of',features.shape[1])
est = LogisticReg(solver='saga',
C = 1000,
random_state=random_state)
est.fit(features[:,i].reshape(-1,1),labels)
print('pvalue:',est.p_values[0],
'weight:',np.abs(est.model.coef_.flatten()[0]))
if est.p_values[0] < 0.05:
scores.append(np.abs(est.model.coef_.flatten()[0]))
else:
scores.append(0)
# save file
out_text=''
param_string = ','.join(['{}={}'.format(p, v)
for p,v in est.model.get_params().items()])
# algorithm seed feature score
for i,c in enumerate(scores):
out_text += '\t'.join(['Univariate LR',
param_string,
str(random_state),
feature_names[i],
str(c)])+'\n'
import os
if os.path.exists(save_file.split('.csv')[0]+'.imp_score'):
os.remove(save_file.split('.csv')[0]+'.imp_score')
if os.path.exists(save_file.split('.csv')[0]+'.roc'):
os.remove(save_file.split('.csv')[0]+'.roc')
ext = '.univariate_score'
with open(save_file.split('.csv')[0] + ext,'w') as out:
out.write('algorithm\talg-parameters\tseed\tfeature\tscore\n')
out.write(out_text)
