DislikedThe results you see on this post use the same trailing stop and machine learning mechanism as before only that this time we use 17 bars from the GBPUSD and 17 bars from the EURUSD for each example used in trainingIgnored
Thanks
Ruby
Algorithmic Quant Trading (Machine Learning + Stat-Arb) 25 replies
Forex with Machine Learning Software Project 1 reply
Machine Learning + Retail Forex = Profitable? (Quant) 1 reply
Potential new machine learning style software. 79 replies
My most recent advancements into machine learning 16 replies
DislikedThe results you see on this post use the same trailing stop and machine learning mechanism as before only that this time we use 17 bars from the GBPUSD and 17 bars from the EURUSD for each example used in trainingIgnored
DislikedNow I have search space of 112 parameters - it takes 2h to train and test (15 years, EOD) in case of 100 "random tickers" it would lead to 2 weeks That's it for now - good luck Special thanks to algoTraderJo, PipMeUp, KaBoIgnored
Disliked{quote} ATJ, Is it an ensemble of two models one for EU and the other for GU? If not, how do you use data from different assets using F4? which input / output functions do you use? Thanks RubyIgnored
rateRequirements = 8000 symbolRequirements = D timeframeRequirements = 0
rateRequirements = 8000,8000 symbolRequirements = D,GBPUSD timeframeRequirements = 0,0
Data<RealVector> inputs(period,RealVector(barsUsed));
Data<RealVector> inputs(period,RealVector(barsUsed*2));
for(j=0;j<barsUsed;j++){ inputs.element(m)[j] = (iOpen(0,tl+j+i)-iOpen(0,1+tl+j+i))/iOpen(0,1+tl+j+i); }
for(j=0;j<barsUsed;j++){ inputs.element(m)[j*2] = (iOpen(0,tl+j+i)-iOpen(0,1+tl+j+i))/iOpen(0,1+tl+j+i); inputs.element(m)[j*2+1] = (iOpen(1,tl+j+i)-iOpen(1,1+tl+j+i))/iOpen(1,1+tl+j+i); }
RegressionDataset p_regression_i_simpleReturn(int barsUsed){ Data<RealVector> inputs(1,RealVector(barsUsed*2)); Data<RealVector> labels(1,RealVector(1)); int j = 0, k=0; for (j=0;j<barsUsed;j++){ inputs.element(0)[j*2] = (iOpen(0,j)-iOpen(0,j+1))/iOpen(0,j+1); inputs.element(0)[j*2+1] = (iOpen(1,j)-iOpen(1,j+1))/iOpen(1,j+1); } labels.element(0)[0] = 0; RegressionDataset dataset(inputs,labels); return dataset; }
DislikedIt took some time to read the whole thread but it was time well spent. Over past few years I worked with NN, SVM, fuzzy logic, ensembles etc. I am very glad that I found people that do more or less the same type of algo trading as me. There are few differences however: - I do it for futures and stocks - now I use SVM binary classification - I use Amibroker, my own dll as a plugin(C++), libsvm - I use a lot more inputs and samples to train classifiers - to verify system parameters I did Walk-Forward (knowing about all the limitations) At first I...Ignored
Disliked... In particular I would like to build a Dynamic Neural network based on outputs (feedback) data. The main problem for me will be to avoid the overfitting. ....Ignored
Disliked{quote} Not exactly sure what dynamic nn is - is it same/similar to recursive/Elman/time delayed NN ?Ignored
DislikedHi, After reading this thread I come to one simple conclusion: It's not real time testable. Because of the possible long drawn down time (like 6months or better) and due to daily TF and little number of trades it's very difficult to asses if we really make money or just being fooled by variance of outcome. On the other hand the number of training bars is very low so its guarantee very high variance of prediction. So my question is: Is anybody able to present system on TF like 1H or lower trained on like 5k-20k bars which will generate more trades...Ignored
Disliked{quote} There is a plethora of H1 systems within this thread; in fact, the majority of systems do not trade on the daily timeframe. Look at the recent post (#720) which provides an M30 system. In my opinion, this is about as good as a backtest can be - a constantly evolving (retraining) algorithm tested over 28.7 years on the M30 timeframe, using a professional development framework (that takes into account historical swaps etc) and uses very high quality historical data (if it is the data that is bundled with the framework, then it is from a source...Ignored
DislikedCan you generate the trades of this system and post it together with the currency data ?? I will try to reproduce it. KrzysztofIgnored
DislikedFor the target: if the SL and TP is not reach at the end of horizon for both, what value are you using ?Ignored