Welcome to Forex-TSD!, one of the largest Forex forums worldwide, where you will be able to find the most complete and reliable Forex information imaginable.
From the list below, select the forum that you want to visit and register to post, as many times you want. It’s absolutely free. Click here for registering on Forex-TSD.
Exclusive Forum
The Exclusive Forum is the only paid section. Once you subscribe, you will get free access to real cutting-edge Trading Systems (automated and not), Indicators, Signals, Articles, etc., that will help and guide you, in ways that you could only imagine, with your Forex trading.
Elite Section
Get access to private discussions, specialized support, indicators and trading systems reported every week.
Advanced Elite Section
For professional traders, trading system developers and any other member who may need to use and/or convert, the most cutting-edge exclusive indicators and trading systems for MT4 and MT5.
I am starting to make my experiments here. Unlikely I am from a different background. My degree is Artificial Intelligence so more technical towards AI techniques such as NN and more newbie in forex unlike the rest of the guru's in this forum...
using the historical price to Predict the future price it self is kinda impossible due to many factors..
as a sum up from the previous experiences.. whats are the good inputs and there expected outputs for a NN ..
anyways,, whats your opinion about whats the good values/indicators to be used as inputs in Neural Network and what would be there expected output(indicators, curve movements..etc..)..
I will test it on a series and different types of neural networks with back error probagation. also found interesting publication about genetic mutation for the results. I just have some problem now on where to find reliable data to feed the NN... where do you use your historical data .. any sources?
Regards,
Hidden Ox.
you can use mT4 export data to mysql (there is sample code in mq4 code base ),and then train data in matlab (matlab has database toolbox)
Hey I've been working on building a neural network that can accurately predict the GBPJPY on a 1H chart and I think I might have done it. I fed price and indicator data from GBPJPY and other pairs (USDJPY,GBPEUR etc) into a back propagation network and trained over the last ten years data. Here's a few pictures of the target/output plot. The output is the green line and the target is the blue line. I'm not sure about you but an error of 4.2x10^-26 is enough for me = ) I've zoomed in to the most recent data to show just how accurate it seems. Now just to implement that network into a working EA and ill be set anyone got any suggestions?
- using typical feed forward neural nets for predicting time series like currencies exchange rates is a very poor idea
- you're trying to predict values at H1 time frame - it can't be done for Forex with good results. Use D1 or H4 (for currencies that have low volatility)
- you use to much data as an input - neural net gets "to used" to training data and it will perform very poor in live trading
- you're getting to excited looking at training data
- it is impossible to train a neural net that will work effectively for a long time. Typical neural net trained to predict time series gives about 20-100 good predictions and then it has to be retrained to fit recent changes
If you want to create useful neural nets for time series predictions read about evolutionary neural networks (feed forward neural networks coded as flexible neural trees; their architecture is optimized using PIPE or GEP; flexible activation function parameters are optimized using PSO, EPSO or simulated annealing etc.)
Implement a simple feed forward optimisation algo, stimulate winning trades with some form of asymmetrical MM . Access your real trading results with standardized statistical measurements, so you can know if the whole model or part of it is broken when start under performing. Go back to optimization/drawing board.
All this only if your network captured, some underlying cyclical exploitable behaviour. Hint that you are on path of a success is if your pattern is of fractal nature.
Looking on your previous posts, you probably suffer of over-training/fitting syndrome.
I would go for big number of lower quality trades, so I could wed them out with simpler procedures...
__________________
We do not see things as they are.
We see them as we are.
- using typical feed forward neural nets for predicting time series like currencies exchange rates is a very poor idea
- you're trying to predict values at H1 time frame - it can't be done for Forex with good results. Use D1 or H4 (for currencies that have low volatility)
- you use to much data as an input - neural net gets "to used" to training data and it will perform very poor in live trading
- you're getting to excited looking at training data
- it is impossible to train a neural net that will work effectively for a long time. Typical neural net trained to predict time series gives about 20-100 good predictions and then it has to be retrained to fit recent changes
If you want to create useful neural nets for time series predictions read about evolutionary neural networks (feed forward neural networks coded as flexible neural trees; their architecture is optimized using PIPE or GEP; flexible activation function parameters are optimized using PSO, EPSO or simulated annealing etc.)
Yeah I thought that was a little too good to be true It took less than an hour to build that network. Would these evolutionary neural networks have anything to do with UTHONNs? I've got an article that suggests this type of ANN works much better than traditional FF back prop models. I'm working with MATLAB, would it be possible to do what you said with this? Also what would you suggest for software or is MATLAB ok?
Implement a simple feed forward optimisation algo, stimulate winning trades with some form of asymmetrical MM . Access your real trading results with standardized statistical measurements, so you can know if the whole model or part of it is broken when start under performing. Go back to optimization/drawing board.
All this only if your network captured, some underlying cyclical exploitable behaviour. Hint that you are on path of a success is if your pattern is of fractal nature.
Looking on your previous posts, you probably suffer of over-training/fitting syndrome.
I would go for big number of lower quality trades, so I could wed them out with simpler procedures...
I've read about over-fitting and I'm assuming this is what has happened here. Would I be better splitting that ten year sample into 10 1 year samples and training on one year and testing on the next? It appears to me that the GBPJPY goes through a 5-10 year cycle. I'd chosen a 10 year sample because I'd hoped to train the network to be able to identify that cycle. The hope being that a network trained over that cycle would be able to identify what part of the cycle it was in and respond appropriately.
My previous strategies have involved waiting for big moves and then cashing in but I would hope to use a neural net as part of a scalping strategy. I'd be looking to trade nearly every tick based on whether the forecasted close was above or below the current.
Evolutionary neural networks are feed forward neural networks build and optimized using genetic computation algorithms and optimization methods. It's something else than UHHONN's (btw UHHONN's look very promising and are next on my testing list ).
Linear Mode
