Forums  > Pricing & Modelling  > A neural net calibration question  
     
Page 2 of 2Goto to page: 1, [2] Prev
Display using:  

nikol


Total Posts: 853
Joined: Jun 2005
 
Posted: 2019-04-30 13:53
@Baltazar

>>but because you udpate your detector continusly, such information could be of interest, therefor I propose (after you've gotten familiar with svm) to check "http://www.irccyn.ec-nantes.fr/hebergement/Publications/2003/1668.pdf".

Can we have this paper?

At least title

Baltazar


Total Posts: 1773
Joined: Jul 2004
 
Posted: 2019-05-01 00:31
I don't really recall to be honest.
I think it was related to one class support vector for novelty detection.

Qui fait le malin tombe dans le ravin

tbretagn


Total Posts: 281
Joined: Oct 2004
 
Posted: 2019-05-01 10:15
Blast from the past but i imagine it's irrcyn paper

When I think I was there at the time and wasted it on girls, booze and poetry...

Et meme si ce n'est pas vrai, il faut croire en l'histoire ancienne

nikol


Total Posts: 853
Joined: Jun 2005
 
Posted: 2019-05-01 11:36
> wasted it on girls, booze and poetry...

The life is infinite... vs ["Bla-bla-bla" at ending of "Death becomes her"]
Nothing is wasted.

Thank you!

nikol


Total Posts: 853
Joined: Jun 2005
 
Posted: 2019-05-01 12:14
On NN calibration:

My dataset is time series of snapshots of market parameters (multivariable) taken at fixed intervals (5 minutes).
These are injected into ANN (layers of LSTM) by batches of length N (number of sequential snaps).
Prediction variable is next return.

Training is done in cycles over
for i = 1 to M
     batch[i:N+i]

Idea is to keep ANN up to date. However, after a while I get overfitting, which is well identifiable with slow degradation of 'loss' parameter - see plot.

I stop training, let say at 2500. This net works reasonably well for another ~3000 periods (about 10 days). Hence, I am facing operational problem to continuously train new network for the next period, which is a serious hussle. I planned to reduce research time by implementing ANN, but now I have to 'feed this animal' continuously.

Perhaps, I am doing something wrong. Is it possible to keep ANN updated incrementally without total recalibration?


ronin


Total Posts: 512
Joined: May 2006
 
Posted: 2019-05-01 15:08
> > wasted it on girls, booze and poetry...

“I spent half my money on gambling, alcohol and wild women. The other half I wasted.”
― WC Fields

Sorry, couldn't resist...

"There is a SIX am?" -- Arthur
Previous Thread :: Next Thread 
Page 2 of 2Goto to page: 1, [2] Prev