diff --git a/datamodel.py b/datamodel.py
new file mode 100644
index 0000000000000000000000000000000000000000..61b8507a2617f2fc22af62f1ea8b5e07b464eb70
--- /dev/null
+++ b/datamodel.py
@@ -0,0 +1,107 @@
+import pandas as pd
+import numpy as np
+from sklearn.preprocessing import OrdinalEncoder, OneHotEncoder
+from sklearn.preprocessing import MinMaxScaler
+from sklearn.feature_selection import VarianceThreshold
+
+class datamodel:
+ '''
+ Used for pre-processing.
+ '''
+
+ def __init__(self,x_train,y_train,x_test,y_test,target_col,pos_class,neg_class,sensitive_col_name, sensitive_val1,sensitive_val2,sensitive_train_col,sensitive_test_col):
+ self.x_train=x_train
+ self.x_test=x_test
+ self.y_train=y_train
+ self.y_test=y_test
+ self.target_col=target_col
+ self.pos_class=pos_class
+ self.neg_class=neg_class
+ #the sensitive data is tracked by remembering the sensitive column
+ self.sensitive_col_name=sensitive_col_name
+ self.sensitive_train_col=sensitive_train_col
+ self.sensitive_test_col=sensitive_test_col
+ self.sensitive_val1=sensitive_val1
+ self.sensitive_val2=sensitive_val2
+
+ def encode_columns(self,ordinal_cols,nominal_cols):
+ if (ordinal_cols!=[]):
+ encoder = OrdinalEncoder(handle_unknown='use_encoded_value',unknown_value=-1)
+ self.x_train[ordinal_cols]=encoder.fit_transform(self.x_train[ordinal_cols])
+ self.x_test[ordinal_cols]=encoder.transform(self.x_test[ordinal_cols])
+ if (nominal_cols!=[]):
+ encoder = OneHotEncoder(handle_unknown='ignore')
+ nominal_x_train=encoder.fit_transform(self.x_train[nominal_cols]).toarray()
+ nominal_x_test = encoder.transform(self.x_test[nominal_cols]).toarray()
+ categories = np.concatenate(encoder.categories_)
+ self.x_train[categories]=nominal_x_train
+ self.x_train = self.x_train.drop(columns=nominal_cols)
+ self.x_test[categories]=nominal_x_test
+ self.x_test = self.x_test.drop(columns=nominal_cols)
+
+
+ def remove_sensitive_feature(self):
+ if self.sensitive_col_name in self.x_train.columns:
+ self.x_train =self.x_train.drop(columns=self.sensitive_col_name)
+ self.x_test = self.x_test.drop(columns=self.sensitive_col_name)
+
+
+
+ def feature_scaler(self):
+ scaler = MinMaxScaler()
+ self.x_train[self.x_train.columns] = scaler.fit_transform(self.x_train)
+ self.x_test[self.x_test.columns] = scaler.transform(self.x_test)
+
+
+ def select_features_by_variance(self,binaries):
+ '''
+ Binary attributes that consist of 80% zeros or ones are removed
+ '''
+ sel = VarianceThreshold(threshold= .8 * (1 - .8))
+ non_binaries = list(set(self.x_train.columns)-set(binaries))
+ sel_x_train=sel.fit_transform(self.x_train[binaries])
+ sel_x_train = pd.DataFrame(sel_x_train,columns=sel.get_feature_names_out(),index=self.x_train.index)
+ sel_x_test = pd.DataFrame(sel.transform(self.x_test[binaries]),columns=sel.get_feature_names_out(),index=self.x_test.index)
+ self.x_train = pd.concat([sel_x_train,self.x_train[non_binaries]],axis=1)
+ self.x_test = pd.concat([sel_x_test,self.x_test[non_binaries]],axis=1)
+
+ def get_sensitive_train_data(self):
+ '''
+ Used to extract the data for pre-processing of group-specific classifiers
+ '''
+ sens_1_x_train = self.x_train[self.sensitive_train_col==self.sensitive_val1]
+ sens_1_y_train = self.y_train[self.sensitive_train_col==self.sensitive_val1]
+ sens_2_x_train = self.x_train[self.sensitive_train_col==self.sensitive_val2]
+ sens_2_y_train = self.y_train[self.sensitive_train_col==self.sensitive_val2]
+ return sens_1_x_train,sens_1_y_train,sens_2_x_train,sens_2_y_train
+
+ def get_sensitive_test_data(self):
+ '''
+ Used to extract the data group-specific classifiers + group-filtered pre-processed data of the single classifier
+ '''
+ sens_1_x_test = self.x_test[self.sensitive_test_col==self.sensitive_val1]
+ sens_1_y_test = self.y_test[self.sensitive_test_col==self.sensitive_val1]
+ sens_2_x_test = self.x_test[self.sensitive_test_col==self.sensitive_val2]
+ sens_2_y_test = self.y_test[self.sensitive_test_col==self.sensitive_val2]
+ return sens_1_x_test,sens_1_y_test,sens_2_x_test,sens_2_y_test
+
+
+ def create_contingency_table(self,attr_label1,attr_label2):
+ '''
+ Visualises the class and group distribution
+ '''
+ train_data = pd.concat([self.x_train,self.y_train],axis=1)
+ pos_class = train_data.query(''+self.target_col+'== @self.pos_class').shape[0]
+ neg_class = train_data.query(''+self.target_col+'== @self.neg_class').shape[0]
+ val1_total = train_data.query(''+self.sensitive_col_name+'== @self.sensitive_val1').shape[0]
+ val2_total = train_data.query(''+self.sensitive_col_name+'== @self.sensitive_val2').shape[0]
+ val1_pos = train_data.query(''+self.target_col+'== @self.pos_class and '+self.sensitive_col_name+'== @self.sensitive_val1').shape[0]
+ val1_neg = train_data.query(''+self.target_col+'== @self.neg_class and '+self.sensitive_col_name+'== @self.sensitive_val1').shape[0]
+ val2_pos = train_data.query(''+self.target_col+'== @self.pos_class and '+self.sensitive_col_name+'== @self.sensitive_val2').shape[0]
+ val2_neg = train_data.query(''+self.target_col+'== @self.neg_class and '+self.sensitive_col_name+'== @self.sensitive_val2').shape[0]
+ total = train_data.shape[0]
+ contingency_table = pd.DataFrame(index=[attr_label1,attr_label2,'total'],data={
+ 'positive class':[val1_pos/total,val2_pos/total,pos_class/total],
+ 'negative class':[val1_neg/total,val2_neg/total,neg_class/total],
+ 'total':[val1_total/total,val2_total/total,(val1_total+val2_total)/total]})
+ return contingency_table
\ No newline at end of file