LUluanton·Sep 3, 2018Submission
Predicting Time Series with Neural Networks (Keras)
4 points · Open discussionLukeTonin
keras-seq-2-seq-signal-prediction
Jupyter Notebook2018–2018
LukeTonin / keras-seq-2-seq-signal-prediction
keras-seq-2-seq-signal-prediction
An implementation of a sequence to sequence neural network using an encoder-decoder
2Distinct people
2Mentions
0Explicit recommenders
212GitHub stars
What to know before exploring
What it is
An implementation of a sequence to sequence neural network using an encoder-decoder
GitHub
Where it is used
No documented product use has been verified yet.
What people are saying
No recommendation phrase was found in the indexed comments.
No classified warning is attached to this page. Read the discussions before treating silence as approval.
Also in these discussions
No other repository is repeatedly linked in the same discussions.
This week on Hacker News
Nov 7–Nov 14 UTC · Direct repository links and separately labeled related project stories. Indexed through 11/14/2026, 5:04:57 PM UTC.
No direct repository links were found in the indexed HN discussions over the past seven days. Older discussions are available below.
Mentions over time
HN mentions by hour2 dated mentions
Observed mentions: 2018-09-02 21:57:29 UTC to 2018-09-03 15:13:29 UTC.
Counts come from the repository’s dated source evidence. Index cutoff: 2026-11-02 08:16:59 UTC.
View mention counts and date ranges
| Period | Start (UTC) | End (exclusive, UTC) | Mentions | Submissions | Comments |
|---|---|---|---|---|---|
| Sep 2, 21:00 | 2018-09-02 21:00:00 UTC | 2018-09-02 22:00:00 UTC | 1 | 0 | 1 |
| Sep 2, 22:00 | 2018-09-02 22:00:00 UTC | 2018-09-02 23:00:00 UTC | 0 | 0 | 0 |
| Sep 2, 23:00 | 2018-09-02 23:00:00 UTC | 2018-09-03 00:00:00 UTC | 0 | 0 | 0 |
| Sep 3, 00:00 | 2018-09-03 00:00:00 UTC | 2018-09-03 01:00:00 UTC | 0 | 0 | 0 |
| Sep 3, 01:00 | 2018-09-03 01:00:00 UTC | 2018-09-03 02:00:00 UTC | 0 | 0 | 0 |
| Sep 3, 02:00 | 2018-09-03 02:00:00 UTC | 2018-09-03 03:00:00 UTC | 0 | 0 | 0 |
| Sep 3, 03:00 | 2018-09-03 03:00:00 UTC | 2018-09-03 04:00:00 UTC | 0 | 0 | 0 |
| Sep 3, 04:00 | 2018-09-03 04:00:00 UTC | 2018-09-03 05:00:00 UTC | 0 | 0 | 0 |
| Sep 3, 05:00 | 2018-09-03 05:00:00 UTC | 2018-09-03 06:00:00 UTC | 0 | 0 | 0 |
| Sep 3, 06:00 | 2018-09-03 06:00:00 UTC | 2018-09-03 07:00:00 UTC | 0 | 0 | 0 |
| Sep 3, 07:00 | 2018-09-03 07:00:00 UTC | 2018-09-03 08:00:00 UTC | 0 | 0 | 0 |
| Sep 3, 08:00 | 2018-09-03 08:00:00 UTC | 2018-09-03 09:00:00 UTC | 0 | 0 | 0 |
| Sep 3, 09:00 | 2018-09-03 09:00:00 UTC | 2018-09-03 10:00:00 UTC | 0 | 0 | 0 |
| Sep 3, 10:00 | 2018-09-03 10:00:00 UTC | 2018-09-03 11:00:00 UTC | 0 | 0 | 0 |
| Sep 3, 11:00 | 2018-09-03 11:00:00 UTC | 2018-09-03 12:00:00 UTC | 0 | 0 | 0 |
| Sep 3, 12:00 | 2018-09-03 12:00:00 UTC | 2018-09-03 13:00:00 UTC | 0 | 0 | 0 |
| Sep 3, 13:00 | 2018-09-03 13:00:00 UTC | 2018-09-03 14:00:00 UTC | 0 | 0 | 0 |
| Sep 3, 14:00 | 2018-09-03 14:00:00 UTC | 2018-09-03 15:00:00 UTC | 0 | 0 | 0 |
| Sep 3, 15:00 | 2018-09-03 15:00:00 UTC | 2018-09-03 16:00:00 UTC | 1 | 1 | 0 |
What Hacker News says
2 sourcesI see here that original poster (OP) of the post tried to use many-to-one LSTMs instead of many-to-many LSTMs. I tell that first by looking at the charts. Then I saw the method named "predict_point_by_point" with the comment "Predict each timestep given the last sequence of true data, in effect only predicting 1 step ahead each time" in his code here: https://github.com/jaungiers/LSTM-Neural-Network-for-Time-Se...
I strongly think the system would be better to perform many predictions at once instead, using seq2seq neural networks. The problem is properly explained here at the beginning of this other post: https://github.com/LukeTonin/keras-seq-2-seq-signal-predicti...
This other post is, in turn, derived from my original project here doing seq2seq predictions with TensorFlow: https://github.com/guillaume-chevalier/seq2seq-signal-predic..
Read the rest
.
OP also forgot to cite the image I made: https://en.wikipedia.org/wiki/Long_short-term_memory#/media/...
Well, glad to see that some similar work as mine can get this much traction on HN. I would have loved to get this much traction when I did my post, too. Anyway, I would suggest OP to take a look at seq2seq, as it objectively performs better (and without the "laggy drift" visual effect observed as in OP's figure named "S&P500 multi-sequence prediction").
In other words, using many-to-one neural architectures creates some kind of feedback which doesn't happen with seq2seq which doesn't build on its own accumulated error. It has a decoder with different weights than the encoder, and can be deep (stacked).