How to choose the final experiment result?
Hessen525 opened this issue · comments
Hi, Dr.Zhong @MingjunZhong
How do you choose the final experiment result? In same model and parameters, the predicted values of same appliance are different. So, I am so confusion, how do you choose the final result? I have read many NILM papers, but none of them clearly show how they choose the final experimental results.
I am looking forward to your reply,
Thanks you so much!
Are you using seq2point model? If so, once the model is trained, the prediction of a window should be the same and so I can't see why you had different predictions.
Are you using seq2point model? If so, once the model is trained, the prediction of a window should be the same and so I can't see why you had different predictions.
@MingjunZhong
Thanks for your reply! I mean how to choose the final result because each trial has different result.
EX： I train seq2point for dishwasher 5 times, the prediction of them are different. So, how do you choose the final result for dishwasher?
@Hessen525 That's a good questions indeed, but I do not know because each initialization represents one model. So you have produced multiple prediction results from different models. The easiest approach is to choose the model from the best validation results and the apply that model for prediction. You could also average them.
Thank you so much，Appreciate！