- Search Forex Factory
- 32 Results
- KaBo replied Jan 22, 2016
Could you please poste the setup / code to the post 827? Thanks!
- KaBo replied Jan 15, 2016
I don't want to spoil the euphoria about having a new approach, but aren't the net results factor 5 worse than even the simplest ML experiments before? By which metric do you value the success of the experiment? And at what point or value do you ...
- KaBo replied Dec 25, 2015
Hi Nghei, this is a good idea, at least I thought so, too. In my experiments I couldn't get any edge though. But then again, I haven't done it exactly like you described. I tried omitting or packing data e.g. in one bar. Cheers KaBo
- KaBo replied Sep 29, 2015
Hi Joe, great to see you back with more ml-ideas. I have been developing and testing the past months. As you say, there is a lot of effort required. Even when you have a working framework there are still a million ways to adjust the ...
- KaBo replied Aug 19, 2015
At post Post614 TP and SL are the same for the systems in the ensemble as they are important for the same exit... I guess.
- KaBo replied Aug 17, 2015
Hi Ruby, no, I have not done it yet. Right now I'm using some variation of the algorithm idea and run experiments with it. Since I don't have hundreds or thousands of cores, it takes time. Would be great if Jo posts the code. Always got it going ...
- KaBo replied Aug 15, 2015
How about reducing the search space for each system? E.g. System-1 only searches Sample-Size 60 -100 and System-2 searches Sample-Size 110 - 150. It would still end up in a huge number of system combinations, but a little better.
- KaBo replied Aug 15, 2015
Yes, I see that and get the point, but already to try and explore the search space of 2 systems is huge - I don't see how I could possible calculate the mining bias. However, the mining bias is based on the process. So if there is a way to combine ...
- KaBo replied Aug 15, 2015
Why not search for good systems and test for potential ensemble combinations within the best systems? This process would be a straight forward procedure that we can apply on the bootstrapped most successful systems as well to measure bias.
- KaBo replied Aug 8, 2015
Hi Jo, I will try to get the setup going next week, but how do these results compare to your previous posts? Do you prefer the approach with the ATR TP / SL over the MSE/MLE? I tried some variations with the ATR but never having the SL different to ...
- KaBo replied Aug 6, 2015
I tried and it wasn't improving the results (much), but added a lot of computation time as you need to loop through all the bars that don't have a target/output yet. I thought being able to include the past couple of samples would add value, but it ...
- KaBo replied Aug 6, 2015
As you described, having a larger value for horizon also means that we are excluding the past samples. So a value of 1.000 means we are excluding 1000/24 = 42 days. The model would lack recent information. A way to get around this issue is to have ...
- KaBo replied Jun 11, 2015
Thanks babelproofre for the explanation. My question then is, why would we use the bootstrap with replacement? It would not keep the statistical information such as total up and down during the time. Wouldn't it make more sense to use the random ...
- KaBo replied Jun 10, 2015
Yes, and I am getting very exciting results by now (LinReg -> NN). They are so good to the point that it is hard to believe they are for real - I will double and triple check as we have all been down that road before ;-) I did a lot of testing. It ...
- KaBo replied Jun 10, 2015
yes and yes. You always take the trading decision at the open of a candle so the rest of the candle is still ahead of you, including high and low which are included in the projected horizon.
- KaBo replied May 27, 2015
When combining the system, how do you account for the different time of trading decision? e.g. One has its decision at 6 o'clock and one at 2 o'clock. I'm still a unsure about how to account for the mining bias. For example, I find one good system ...
- KaBo replied May 26, 2015
Linear Regression is already very impressive! However, the combinations you came up with is cherry picking of a massive exploration. So in order to believe that there is no or little bias, I would need to do the experiment from the beginning (as ...
- KaBo replied May 26, 2015
Yippee! Thank you Jo for the description, and Daniel for the support, but it was what I was doing all along. There must had been some problem with my setup, operating system, or who knows. With the new Framework version I ...
- KaBo replied May 19, 2015
I don't even know what else to ask?! I even downloaded and re-installed the F4 4.4.14 from scratch, took your exact files for the EURUSD from post 409 and cannot reproduce the past results. Not even close. I am using Strategy Tester 0.39 ...
- KaBo replied May 18, 2015
I will test tomorrow, as I want to double check to have the setup identical. The results look more along the lines to what I got before, though. (About 40% long drawdown and not quite as steady curves) So is your Linux distro now identical to the ...