It seems that the DL indicator values change not only depending on the strategy's start date, as I wrote here https://www.wealth-lab.com/Discussion/DLPredictor-indicator-changes-with-strategy-start-date-12671 , but also depending on the strategy's end date.
If I run a strategy with the DL indicator from 2000 to 2025, the annual returns are as follows:
But if I run a strategy from 2000 to 2010, the annual returns are different:
This means that the DL indicator values also change.
If I run a strategy with the DL indicator from 2000 to 2025, the annual returns are as follows:
But if I run a strategy from 2000 to 2010, the annual returns are different:
This means that the DL indicator values also change.
Rename
If you do not change the end date but run the strategy twice in a row do the returns change? This would indicate something like NSF positions, i.e. not enough capital to take all of the trades.
No, it doesn't change. Single-symbol strategy, non-futures mode, margin factor 9
I have an idea it's having to do with normalization; can you email me the Deep Learning network file?
I sent
Thank you, I have it resolved for next release cycle! It was indeed normalization, certain activation functions require standard deviation normalization and DL was using the std dev of the input data instead of the training data.
Thank you for the promptness. I look forward to the next release.
Your Response
Post
Edit Post
Login is required