fullTPOT.py

from tpot import TPOTRegressor
import math
import pandas as pd


def strToNum(st):
 try:
  f = float(st)
  if math.isnan(f):
   return -10000
  return f
 except:
  sta = [ord(x) for x in st]
  cs = 0
  for i in sta:
   cs += i
  return cs
 





if __name__ == '__main__':
 dataset = pd.read_csv('../homecredit_results/feature_matrix.csv')
 training = dataset[dataset['TARGET'].isnull() == False]
 test = dataset[dataset['TARGET'].isnull()]
 target = training['TARGET']
 training.drop(columns = 'TARGET', axis = 1 , inplace = True)
 test.drop(columns = 'TARGET', axis = 1 , inplace = True)
 for column in training.columns: # mask all string values and nan's
  training.loc[:, column] = pd.Series(training[column].apply(strToNum))
  test.loc[:, column] = pd.Series(test[column].apply(strToNum))
 tpot = TPOTRegressor(generations = 5, population_size = 50, verbosity = 2)
 tpot.fit(training.values, target.values)
 tpot.export('tpot_homecredit_best_pipeline.py')
 result = tpot.predict(test.values)
 submission_file = open('submission.csv','w')
 submission_file.write('SK_ID_CURR,TARGET\n')
 sk_id_curr = test['SK_ID_CURR'].values
 for i in range(len(sk_id_curr)):
  new_line = str(sk_id_curr[i]) + ',' + str(result[i]) + '\n'
  submission_file.write(new_line)
 submission_file.close()