Dear NN-Experts, Len,

Are generally in NN the results "better" if the output values are binary ( 1 or 0) ?
What is your estimation about that?

Regards
Christian

Not binary, rather a continuum. The reason for a continuum is that I always use the NN indicator value as buy priority. I also use % Equity as the PosSizer. These two are because, on down market days, NN scores rise across the board. I want to buy those instruments where the NN score is highest, ie greatest probability of future gain.

What do I most often use as output variable? Simply change percent of 5-bar Close, often range checked.

Hi Len,

thanks for your answer.
What do you mean by "continuum" ?
Maybe i could not translate it in german correctly.

Continuum == Values go from low to high or from high to low. Not up and down. y=m*x + b.

Hm, i think it would be better for my understanding to have an example.
Do you have an example (input/ouput) where you are using a continuum?
just if you want .
Of course It need not to be your best trained NN:-) , but just to get an estimation or a feeling, how you have it set up in NN.

My usual Output script, limits output to continuous range, -16 to +16...

CODE:
Please log in to see this code.

QUOTE:
Continuum == Values go from low to high or from high to low. Not up and down. y=m*x + b.

I am by no means an expert in NN's, but the slope formula is a good example of how to think about the inputs for your NN. A NN just assigns a positive or negative weight to each input (and a weight to each input in each hidden layer). Therefore, the most important consideration is that your inputs make sense when they are weighted.

A example of bad inputs would be the day of the week. The problem is that the days go from 1 to 7 and then back to 1. Therefore, if it turns out that Monday is the worst day to buy and Wednesday is the best day to buy, your NN is going to tell you that Friday is even better than Wednesday.

I'm not sure I agree with Len that binary values are necessarily bad. In the above example, I think you would get a better result with 5 binary inputs, ie day is Monday, day is Tuesday, day is Wednesday, day is Thursday and day is Friday. That allows your NN to assign a different weight to each day, which may not be continuous.

QUOTE:
day is Monday, day is Tuesday...

That looks like an excellent way to handle inputs. A binary output was the original question, which I would avoid.

The main reason is that I use NN value as buy Priority, and I use percent equity as PosSizer. Usually, on down market days, NN scores rise across the board because potential for gain has increased. I let insufficient equity restrict actual trades to those with the best potential (at least per NN score).

QUOTE:
A binary output was the original question

That's what I get for reading too quickly.

I haven't worked extensively with Neuro-Lab, so this may not be true of what is possible in Wealth-Lab. In terms of outputs, it seems to me there are two things you can ask the NN to do:

1. Output an expected price or percentage increase for the security.
2. Output the probability of a security going up (which is a binary decision and the associated probability).

As I understand it, a NN can either give you an expected price or a probability, but not both. Therefore, if you use #1 as a priority, you can buy securities likely to make large moves, but you don't know the probability of that happening. If you use #2 as a priority, you buy the securities that are most likely to go up, but you don't know how much.

QUOTE:
I use NN value as buy Priority

Is your NN value the probability of the security going up?

QUOTE:
Is your NN value the probability of the security going up?

No, expected gain. See output script in post #6. I wouldn't know how to code for probability. , , but your C# script can return anything.

From the User Guide...

QUOTE:
Using the Trained Neural Network
...
*Neuro-Lab® indicators are normalized in the range from 0 to 100. When the value of the NL indicator is high, the network is indicating that the probability is greater that the condition for which the NN was trained will occur.

I don't believe a word of it!