There are other networks that are sometimes helpful when processing
time series. One is a recurrent Laguerre net. Such a net does not
require a tapped delay line of input vectors, since the Laguerre axon
has memory. Perhaps one could simulate this as an input axon using
L0[i] = alpha*input + (1 - alpha)*L0[i-1];
L1[i] = -(1 - alpha)*L0[i] + L0[i+1] + (1 - alpha)*L1[i+1];
L2[i] = -(1 - alpha)*L1[i] + L1[i+1] + (1 - alpha)*L2[i+1];
L3[i] = -(1 - alpha)*L2[i] + L2[i+1] + (1 - alpha)*L3[i+1];
Then the 4 outputs would depend upon the history of the single
input. If each Ln[i] became an input to the ann, the number of inputs
to the ann would be reduced, and training might be speedier. The value
of alpha is usally a network variable to be learned, but that would
require this axon to be integrated into the FANN, which is not likely.
Instead, one might either guess at alpha, or try to optimize it, but
the latter might be a very time-consuming process. If you have a
suggestion of another method to incorporate such an axon into FANN, I
would greatly appreciate hearing it.
This is interesting. I never thought about memory axons but there definitely is a potential in this approach that is worth further exploration :)
Frankly I have no better idea for incorporation of Laguerrenet net in FANN. Probably you might ask on the FANN forum.
1. What is the difference between FANN and other NN? (in terms of financial time series)
FANN is not a type of a network. It's a software that implements simple multilayer networks on C.
2. Silly question: but in your code init(), you have ArrayResize.
This function only resizes 1 (the first, I think?) dimension of the
array, but if we have a 2 dimension array, how do we resize the 2nd
AFAIK it's not possible in MQL4. It's one of two the biggest limitations of MQL4 (the other is lack of structure types) IMHO.
3. In your explanation of the start() function, you mentioned you
have a "genetic algorithm for optimization process". What do you mean
by this - is this an optimizer that can search for robust parameters
(that is, search for minima/maxima without sharp rises/falls). If so,
could you kindly explain this as well?
What I meant was that i optimized certain parameters using MT4 built in genetic optimization.
And yes: you can have any number of net outputs in FANN but you should really know what you are doing with the network!
Also, can you give us any guidelines in choosing the hidden layer#, hidden neuron#, and training algorithm (besides experimenting) if possible?
Yeah, I read his Thesis a couple of time. It's an excellent source of information about RL and NN.
You're right that there are algorithms like cascade2 that can procude a network with "optimum" number of connections. However those are not much more than educated guess and cannot be used in all circumstances. For example you need a large set of input data and a known output values. In some applications neither is true.
regarding this data issue, here you have a lot of data... but as you said before for those who believe in AI over the capital markets, are a few things that they should know: what you want to do, what kind of inputs you want to use, getting crystal clear data and so on.
regarding cascade2 imho is a solution because we do have a dynamic market with dynamic rules. most of the white papers I have read(and believe me I did read hundreds of it) most of the researchers model the problem(forecasting forex) like a bi dimensional problem which is not at all and it is a multi dimensional problem. where my opinion is: "Understanding the PAST and watch the PRESENT, we can look into the FUTURE" :) so what I try to say is that first using data mining and KDD(Knowladge Discovery in Database) techniques, we should minner the past and from there to extract the fatures(inputs) and just after this you might want to use the power of ANN. otherwise is like having a F1 engine put it into a Trabant car :)
First of all thankyou for your excellent tutorial and work. I really liked your library. In the past I have also created similar libraires based on JNNS and NeuroSolutions. One of the library still exists on the Babypips website.
One reason i liked your library is that its possible for online training and testing from MT4. I would like to contribute to your project with additional functionality (Batch learning & Plotting MSE from MT4). If you need help or looking for contributors I am ready to join the team.
Currently I am using self developed library based on NeuroSolutions, but sooner or later I would like to use FAAN and your library.
Sure! I'd really appreciate your contributions, but please try to keep the coding style so the code remains consistent.
I'm aware of the fact that Fann2MQL lacks many features of FANN so any suggestions or contributions are more than welcome!
How to modify this EA in order to work it toghether with MACD and another indicator, in example MACD+MA ?
In example, modify it to work not only with MACD but also with MA, something in example like this :
When MA 's value of last bar is more than MA last bar -1 (shift back 1 bar) it is a BUY signal,
in example BAR 0 : MA(40) 1.4000; BAR -1 : MA(40) 1.3980 --> Buy signal
when MA's value of the last bar is less than MA last bar -1 (shift back 1 bar) it is a SELL signal,
in example BAR 0 : MA(40) 1.3727; BAR -1 : MA(40) 1.3735 --> Sell signal
MA in example 40 period, close price, exponential.
What to add/modify in the original code of this EA in order to have
both MACD + MA signals before launch the order?
In example, launching a BUY order only when MACD get buy signal AND when MA get also BUY signal.
I hope to be clear.
Thanks for insights in advance.
I tried to test the attached EAs. I tested the EA for the period 1.1.2007 to 1.1.2009. First I ran the EA with SaveAnn=true and NeuroFilter=false and then I ran the EA with SaveAnn=false and NeuroFilter=true. I got the same result. Am I doing anything wrong?