Gamestudio Links
Zorro Links
Newest Posts
How is this possible ? winning percentage 100%
by chsmac85
Today at 14:15
Asset capacity
by tuuPaul
Today at 13:48
New Zorro version 1.96
by jcl
Today at 13:13
Next contest?
by Michael_Schwarz
Today at 10:33
Help! 1.88.6 Crashes <5 minutes after clicking Trade!
by jcl
Today at 10:28
WFO
by jcl
Today at 10:27
TA-Lib not updated
by jcl
Today at 10:24
AUM Magazine
Latest Screens
Worst Case Z
AckCon'18 - Lotter vs the World 2 - Preview Release
Evonet Beta v1.0 : Online Project Development Template
Lake scene
Who's Online
17 registered (Ayumi, alibaba, kvm, chsmac85, KadigunFx, CocaCola, HellThunder, Iglarion, 3 invisible), 466 Guests and 7 Spiders online.
Key: Admin, Global Mod, Mod
Newest Members
fandi, Deedee, avgtrader, aventador, gokhanuck
18103 Registered Users
Topic Options
Rate This Topic
#474739 - 11/02/18 19:16 Understanding Problem - Financial Hacker Article
turtle_trader Offline
Guest

Registered: 11/02/18
Posts: 2
Hello people,

iam working through this article:
http://www.financial-hacker.com/build-better-strategies-part-5-developing-a-machine-learning-system/
and i lack some understanding.

In step 6 when running the function TestOOS() the neural net is pretrained with SAE and afterwards trained, with the feature file that was created before.

Later in Step7 and Step8 we dont call the TestOOS() function directly but run Zorro in training mode.
So my questions are is it neccesary to train the neural net like it was done in step 6 or is the Training by Zorro doing exactly the same?(Creating a .csv feature set, pre train with SAE and train the nn)

I dont get the bridge between those steps.

For example when i want to train the neural net for a different asset is it enough to pick the DeepLearn Strategy in Zorro or do i need to do Step6 beforehand and train afterwards in Zorro again or can i skip the step 6 and work directly with Zorro?


Thank you!

Top
#474759 - 11/04/18 16:48 Re: Understanding Problem - Financial Hacker Article [Re: turtle_trader]
Seymour Offline
Newbie

Registered: 10/02/17
Posts: 18
My understanding is that step 6 is for calibrating the neural network model. Once you are satisfied with the precision or accuracy, or any of your criteria, you would go to step 7.
Whether you are testing step 6 depends on if you want to calibrate the model. If you think the neural network parameters you have chosen are pretty robust and can be generalized for different inputs or different assets, I guess you could just skip step 6, and train and test your data directly.

Top
#474899 - 11/12/18 10:25 Re: Understanding Problem - Financial Hacker Article [Re: Seymour]
turtle_trader Offline
Guest

Registered: 11/02/18
Posts: 2
Thank you Seymour.
What iam also confused about is the WFO tests and on what time exactly is trained and tested on and in which period the out of sample tests happen.

Referring again to the code in the article.
StartDate=20140601 (1st June 2014)
WFOPeriod=252*24 (~ 1 year without weekends)
DataSplit = 90; (90% Training Data / 10% Testing Data)

And the Equity Curve goes from May 2015 to August 2016 which is a the out of sample test or not?
But if thats the case why we need the 90/10 split for the period before that out of sample test what are these 10 % testing for?

Top
#474915 - 11/13/18 01:58 Re: Understanding Problem - Financial Hacker Article [Re: turtle_trader]
Seymour Offline
Newbie

Registered: 10/02/17
Posts: 18
http://zorro-trader.com/manual/en/numwfocycles.htm
The manual explains it better than I do.

Top



Gamestudio download | chip programmers | Zorro platform | shop | Data Protection Policy

oP group Germany GmbH | Birkenstr. 25-27 | 63549 Ronneburg / Germany | info (at) opgroup.de