The problem is possibly because there were null values which were replaced by fillna() method. But then even after running the method with integer 0/0.0 as parameter to the method, the problem persists.

I've tried doing this, and I'm successful in combining the two. But is it allowed in case of continuous or discrete coarse mapping? X_train_pre['empexp:0'] = sum([np.where((X_train_pre['emp_length_int'] == 0.0), 1, 0),np.where((X_train_pre['emp_length_int'] == '0'), 1, 0)])

Follow this topic