def model_training_xgb(training_data,testing_data):
  # splits train and validation set
  X = training_data.drop(labels=['msno','is_churn'],axis=1)
  Y = training_data['is_churn']
  X_train, X_val, Y_train, Y_val = train_test_split(X, Y, test_size=0.2,random_state = 2)
  # model
  xgb_watchlist = [(X_train, Y_train), (X_val, Y_val)]
  model = xgb.XGBClassifier(learning_rate=0.08, max_depth=4,n_estimators=300,\
                 subsample=0.5, seed=2,missing=-1)
  model.fit(X_train, Y_train,eval_set=xgb_watchlist,eval_metric='logloss',
            early_stopping_rounds=20,verbose=70)
  # caculating E_val

  model_probs = model.predict_proba(X_val)
    # [:,1] to show the prob to is_churn = 1
  model_val_score = log_loss(Y_val,model_probs[:,1])

  # predict on testing set
  model_pred_testing_set = model.predict_proba(testing_data.drop(labels=['msno','is_churn'],axis=1))
  model_pred_testing_set = model_pred_testing_set[:,1] # take out the prob if is_churn = 1
  submission = pd.DataFrame({"msno": testing_data.msno})
  submission.insert(1,column='is_churn',value=model_pred_testing_set)

  return model, model_val_score, submission