def predict_recurse(dataset, test, model, features_to_impute=['Target_L1D', 'Target_Diff7D', 'Target_Diff14D'], last_feature='Target_L6D'):
    n_steps = len(test)
    merged_data = pd.concat([dataset[-14:], test], axis=0)
    all_index = merged_data.index
    X_test = test.drop(columns="Target")
    sd = -6 # Starting point for filling next value

    # For each step, get the predictions
    for i in range(n_steps-1):
        pred = final_model.predict(X_test)[i]
        # For the three features needed, compute the new value
        X_test.loc[all_index[sd+i], features_to_impute[0]] = pred
        X_test.loc[all_index[sd+i], features_to_impute[1]] = pred - merged_data.loc[all_index[sd+i-7], features_to_impute[1]]
        X_test.loc[all_index[sd+i], features_to_impute[2]] = pred - merged_data.loc[all_index[sd+i-14], features_to_impute[2]]

        # In the last iteration compute the Lag6D value
        if i == 5:
            X_test.loc[all_index[sd+i], last_feature] = pred - merged_data.loc[all_index[sd+i-6], last_feature]


    final_preds = final_model.predict(X_test)
    return final_preds