A question of data input

[packer-c]

Hi!
Could you please tell me that when the traffic flow of 60 minutes will be predicted, the first step is input from t1-t12, and the traffic flow at t13-t24 are predicted (that means the forecast is one hour); so then the second step is input t25-t36 and to predict t37-t48? And so on.
I don't know if I understand correctly. Looking forward to your reply, thank you!

[alvinzhaowei]

Hi there, thanks for your question. Based on your example, if the first step is getting t1-t12 input and predicting t13-t24, the second step gonna be getting t2-t13 input and predicting t14-t25. In the training process, in this way, we can get more training data points. But for testing, it just affects the way to statistic the errors.

[packer-c]

Thank you for your kind reply!
If the input data is like this, how do we select the predicted values that might overlap?
For example, as you have said, the input is t1-t12 and predicts t13-t24; The second step is input t2-t13 and predicts t14-t25. Then they will all be predicted at the t14 point, and may have different values. So how to choose the value of t14? I guess the average or one of them?

[alvinzhaowei]

You can calculate the errors twice.

…

On Fri, 25 Nov 2022, 6:58 pm packer-c, ***@***.***> wrote: Thank you for your kind reply! If the input data is like this, how do we select the predicted values that might overlap? For example, as you have said, the input is t1-t12 and predicts t13-t24; The second step is input t2-t13 and predicts t14-t25. Then they will all be predicted at the t14 point, and may have different values. So how to choose the value of t14? I guess the average or one of them? — Reply to this email directly, view it on GitHub <#3 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AEI7WJYOCKPIEOOJYPNQHTTWKBWUDANCNFSM6AAAAAASKDZ3WY> . You are receiving this because you commented.Message ID: ***@***.***>

[packer-c]

Do you mean to calculate the two errors, and then take a small error value as the final value?

[alvinzhaowei]

I think you can do it in your way. For me I calculated the errors multiple times. For example, if value at t14 is 1, we have predictions for t14 as 2 and 3. The sum of the errors will be (2-1)+(3-1)=3. And the reported error is a mean error of all the data points.

…

On Fri, 25 Nov 2022, 7:14 pm packer-c, ***@***.***> wrote: Do you mean to calculate the two errors, and then take a small error value as the final value? — Reply to this email directly, view it on GitHub <#3 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AEI7WJ2NLZMJFRKQXEFZUZLWKBYNZANCNFSM6AAAAAASKDZ3WY> . You are receiving this because you commented.Message ID: ***@***.***>

[packer-c]

Clearly! Thank you very much for your answer.

[alvinzhaowei]

No problem!

…

On Fri, 25 Nov 2022, 8:10 pm packer-c, ***@***.***> wrote: Clearly! Thank you very much for your answer. — Reply to this email directly, view it on GitHub <#3 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AEI7WJYZHAMBD27FN5PS44DWKB7A7ANCNFSM6AAAAAASKDZ3WY> . You are receiving this because you commented.Message ID: ***@***.***>