- ago
Hello Fellow Traders.

When creating an NN some indicators are already normalized, like RSI ranging from 0 to 100, but is there an internal normalization process for other data types such as Open, Close, Volume or SMA?
1
457
Solved
7 Replies

Reply

Bookmark

Sort
Glitch8
 ( 11.64% )
- ago
#1
Yes everything is normalized to a 0 to 1 range before being fed to the NN.
2
Cone8
 ( 28.45% )
- ago
#2
QUOTE:
Open, Close, Volume or SMA?
Generally, it's best to choose oscillators as inputs. Just imagine normalizing AAPL's price from 2000 to 2023 - there's almost no value in that.

If you want to feed price movement into the NN, I'd suggest using the difference of 2 moving averages, for example, SMA(20) - SMA(50).
3
Best Answer
- ago
#3
QUOTE:
If you want to feed price movement into the NN, I'd suggest using the difference of 2 moving averages, for example, SMA(20) - SMA(50).


Wright, maybe is the reason why when I use these I have spikes in the learning curves or no convergence to minimum error at all.

Thanks for the advice Cone!
1
ww58
- ago
#4
QUOTE:
Generally, it's best to choose oscillators as inputs. Just imagine normalizing AAPL's price from 2000 to 2023 - there's almost no value in that.

If you want to feed price movement into the NN, I'd suggest using the difference of 2 moving averages, for example, SMA(20) - SMA(50).

As far as I understand the crossover functions are available only as building blocks, but not as an indicator, so I can't set for example the moment of crossover between a fast sma and a slow, so I have to make my own indicator and then feed it to neurolab, right? The same is true for other non oscillators AFAIU. If I need to feed for example zigzag, then I also need to make my own indicator, which will produce a value in the neighborhood of 0-100?
0
Glitch8
 ( 11.64% )
- ago
#5
You do need to create your own indicator for a custom NeuroLab input, but NeuroLab will normalize it internally so you needn’t worry about that step.
1
- ago
#6
QUOTE:
If you want to feed price movement into the NN,... use the difference of 2 moving averages, for example, SMA(20) - SMA(50).

Numerical analysis note: The ratio of the two, SMA(20)/SMA(50), will be more stable and more reliable than the difference, and that's what Fidelity Investments' stock screener uses.

In contrast, the difference will be more spiky, but if your strategy is trading against transient opportunities (not trends), that might be an advantage in that specific case. It just depends on what you're looking for.
1
Cone8
 ( 28.45% )
- ago
#7
fwiw, for the difference of 2 moving averages, for example, SMA(20) - SMA(50) or SMA(20) / SMA(50) you can use the MathIndOpInd Transformer Indicator.
1

Reply

Bookmark

Sort