Timeline for LSTM performs poorly with monotonically increasing test set values never seen in training. Why?
Current License: CC BY-SA 4.0
5 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Jul 1, 2020 at 16:23 | vote | accept | user134132523 | ||
| Jul 1, 2020 at 16:23 | comment | added | user134132523 | @SimonAlford Thanks for this confirmation. I was thinking of doing this exact thing. Yes, that is what I meant by the poorly worded "standardize the time windows". I was in need of confirmation that this is a neural network issue. | |
| Jul 1, 2020 at 16:05 | comment | added | Simon Alford | They do. Here's another paper on this type of issue, applied to recurrent networks (including LSTM) and translation task: arxiv.org/pdf/1711.00350.pdf | |
| Jul 1, 2020 at 16:01 | comment | added | Sycorax♦ | This is an interesting suggestion, but I wonder if the results from the paper, which pertain to MLPs, would be true to OP's network, which is an lstm network. | |
| Jul 1, 2020 at 15:53 | history | answered | Simon Alford | CC BY-SA 4.0 |