I would prefer it to keep this thread only about Forex. If you want open up another thread so that we can discuss your stock machine learning results.

Machine Learning with algoTraderJo

Sorry the data I have is proprietary (got it from Asirikuy) but you can get data from Dukascopy or from FXCM for free (Pipmeup posted some data). To answer your question I also do ML for stocks successfully, using similar techniques. In stocks you ...

Machine Learning with algoTraderJo

Here are the steps from an F4.4.14 unstable release on Windows using Visual Studio 2010: Make sure you can compile the Release candidate first without any modifications (make sure the DLL is created inside the bin folder) Close Visual Studio 2010. ...

Machine Learning with algoTraderJo

Don't despair my friend I'll reproduce on a windows machine and will send you the results along with a link to get the exact library source I use. You definitely have a setup problem somewhere.

Machine Learning with algoTraderJo

Which is great Having reproduction on a totally independent setup also gives me further confidence. These experiments also point out feed dependence, which is a complicated topic we are yet to discuss here. In ML this can become quite ...

Machine Learning with algoTraderJo

This compares a K-NN algorithm using MFE/MSE. Similar conclusions as before although in this case there are some significant differences in the 2008-2015 region: image

Machine Learning with algoTraderJo

This is the result of a comparison using data provided by PipMeUp and the Asirikuy 1987 data set (this algo): image It is not surprising that the algo behaves worse on PipMeUp's data during the 2003-2005 period, this is precisely the period where ...

Machine Learning with algoTraderJo

what is the GMT offset and DST of this data?

Machine Learning with algoTraderJo

dropbox link?

Machine Learning with algoTraderJo

dd/mm/yy HH:MM,open,high,low,close,tick volume Here is a sample for a made up 1M candle: 20/02/87 19:00,1.5283,1.53,1.5282,1.5292,1482.0

Machine Learning with algoTraderJo

We can use a KNN algo inside a similar methodology. We use as target 1 if MLE>MSE and 0 if MSE >= MLE. As inputs we use the bullish/bearish character of the returns for each bar (1 bullish, 0 bearish) instead of the actual return value (that we were ...

Machine Learning with algoTraderJo

There are in fact highly linear results for this technique. These are some of them (triple checked this time!): A=13 GMT +1/+2, B=150, C=3, D=8 with a 30% of the ATR(20) as a stoploss. image We can use a similar MSE/MLE technique to build binary ...

Machine Learning with algoTraderJo

Have you checked if your predictions match mine? Also consider that I build my weeks from 4 Monday GMT +1/+2 to 19 Friday GMT +1/+2. I don't consider data outside of this for any type of calculation within the system (not for machine learning or atr ...

Machine Learning with algoTraderJo

I was using Suse but was having some problems reproducing things with Daniel (on the project we are working on) so I decided to move to Ubuntu to be able to have everything setup exactly the same as him. When doing this I ran into some issues . ...

Machine Learning with algoTraderJo

I have seem to have made a mess on my new linux installation with the EURUSD results . Moving fully to a new Linux distro to work with Daniel has proved to be more challenging than I originally thought. Dealing with all sort of stuff I ...

Machine Learning with algoTraderJo

I will save some MLE/MSE predictions on a file and will post it so that you can compare with the predictions you make.

Machine Learning with algoTraderJo

Parameters that generate positive results on the EURUSD are also not limited to a single set. You have several different combinations that yield good returns with good correlation coefficients. The graph below, for the A=10 GMT +1/+2, B=120, C=11, ...

Machine Learning with algoTraderJo

I absolutely encourage you to try this and share your findings with us!

Machine Learning with algoTraderJo

You remind me of when I first finished college (too long ago!). It seems that you have a lot of ideas about how things "should be" in machine learning but you still don't have any practical experience in the field. You are saying "a model cannot ...

Machine Learning with algoTraderJo

The suggestion is too broad. What exact fundamental data? How would you exactly act on it? I would encourage you to carry out this exercise and let me know what you get. After you do this let me know if you still think fundamental data is so ...

Machine Learning with algoTraderJo