Actually, if the code flow go to in LSTM or TBAT (functions_timeseries.py) the following code will go to exception:
print ("el ganador es " + str(winner)) because winner is not created in this code flow. I also hope you can share the reference for each method because of reliability. As the auto_arima.py, I don't know why you choose the df_aler = df_aler[(df_aler['anomaly_score']> 2)], the threshold as 2 need some references.