Concerning reproduction of my results using the F4 framework If any of you have access to F4 here are the files you will need to reproduce the above EURUSD results (other results, just change the settings). Take into account that I have set this up ...

Machine Learning with algoTraderJo

As with our previous machine learning exercises, results are the best for the EURUSD. However the EURUSD results do benefit from a longer horizon of 20 vs the 12 value we tested on the EURJPY and USDJPY (although if you remember our USDJPY tests ...

Machine Learning with algoTraderJo

I also wanted to say that the previously posted EURJPY results were not correct . These results were for another method (which we will be talking about later on) and I posted them erroneously (confused the results, very sorry about that). ...

Machine Learning with algoTraderJo

You should definitely try all those things and share your results with us! I am interested to see if you will come to the same conclusions I did when I attempted to solve the problem in the manner you describe. About integrating R and MT4, that is a ...

Machine Learning with algoTraderJo

Still too early to talk about data-mining bias, but I measure it using my own C based custom software hooked up with R. I gather all the data with C and then use R for the hypothesis tests.

Machine Learning with algoTraderJo

We will get into this much later into the thread. It is not very useful to go into this with so much about the basics still not properly addressed. Yes of course, but still too early to say anything about this. We haven't even seen half of what can ...

Machine Learning with algoTraderJo

This is really where things get the most complicated. You cannot possibly do this based on brute forcing combinations, as you point out there are too many possible ways to combine things to figure out how to build an ensemble. Your questions: I do ...

Machine Learning with algoTraderJo

If you have any trouble send me a PM and I'll help you out.

Machine Learning with algoTraderJo

I sent him all the code to reproduce the USDJPY ensemble results yesterday (which he requested) there should be no problem reproducing. If there is it should be easy to spot.

Machine Learning with algoTraderJo

Any observation that you carry out on historical data to build a trading system is some form of data snooping and introduces mining bias. I account for all of this when I measure data mining bias.

Machine Learning with algoTraderJo

Yes, I do an extensive data mining bias evaluation. I use a variation of White's reality check plus a method similar to the one suggested on the paper you posted. Still too early on the thread to deviate to this topic. I'll start talking about ...

Machine Learning with algoTraderJo

I did not understand the question, please rephrase.

Machine Learning with algoTraderJo

I don't see why you cannot use the bar open. If you are doing an event-driven tick simulation you should have access to the current bar open and all ticks since open without any problems. You should also have access to all previous bar OHLC values ...

Machine Learning with algoTraderJo

These are the results for the A=0, B=130, C=9, D=12 configuration on the EURJPY and USDJPY: image Machine learning is not a "one size fits all" problem.

Machine Learning with algoTraderJo

No cross validation. The machine learning method is already cross-validated since I retrain before each signal, the whole backtest is a cross-validation exercise. I see how the optimization of the machine learning variables can introduce curve ...

Machine Learning with algoTraderJo

There is a very important difference. I don't use the close prices for the calculation of the MSE/MLE I use the open prices. This is why you can get a negative MLE/MSE while I never get this result. I always make a comparison from the open of the ...

Machine Learning with algoTraderJo

I hear you! my first framework took me 3 years to develop and it was a complete nightmare in the end. When you're the only one using something it's difficult to find/fix bugs and get things the way you want them. My main problem was when I wanted to ...

Machine Learning with algoTraderJo

All these tests trade to risk 1% on the SL on each trade.

Machine Learning with algoTraderJo

This is always problematic, one of the reasons why I chose F4 was that there is a community behind it (so bugs get found/fixed) plus the whole source is available. It is very difficult to compare things when private frameworks are being used, ...

Machine Learning with algoTraderJo

First ensure that your data is GMT +1/+2, as always this makes a dramatic difference. Second make sure that you are only training with the right examples. As I mentioned in the first post about this method you need to only use examples that happen ...

Machine Learning with algoTraderJo