Precisely but because PNNs are shallow classifiers the training set must be thoroughly representative of the actual population even more so than other types of NN; PNNs memorize well but don’t infer new behaviors unless fed with all possible variations.
Hopefully Arryex's results will be OOS
I mostly use the Adaptive Net Indicators but also predictions. I keep all CSSA parameters constant for all timeframes so my training rarely takes more than a few minutes. I am very careful at not letting the optimization go all the way through because of overfitting.
You should get similar results live but make sure that your test is done on out-of-sample data. In my trading, I confirm entries too but from several cycles. It’s all based on the cycles shifting to a larger time frame after a news release. Before a news release, I would be on a 5-minute cycle...
CSSA does not look at the future but if you set the training range beyond today, it increases by one bar each time a new bar comes in thus triggering it to recalculate. You have to be careful at setting the training range so that TrainStart + TrainBars < number of bars in your chart.
To me back testing is flawed. Learning from nonstationary market data streams finds an average that is not representing future distributions very well. It took me several years to realize it.
I don’t think it is CSSA you are comparing Ergodic with. I use CSSA from Noxa in my trading and I can tell it does not repaint which makes it so attractive. It is a causal. To me you are looking at SSA which we know repaints the past.
Anyway, if this is true that FullSSA_normalise.mq4 is SSA...
My most successful systems use multiple time-frame cycles. They are based on the observation that cycles shift from one time frame to another. For example a cycle can shift to a larger one due to institutional traders placing huge bets on the daily crossover.
In my trading, I build indicators...