If any here are using pKantuML for ML strategy mining we can post the csv files for the systems we find here and share... Anyone up for it?
Thread Tools  Search this Thread 

Algorithmic Quant Trading (Machine Learning + StatArb) 25 replies
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
DislikedHow do you make the renko charts immune to this problem? Any takers?Ignored
QuoteDislikedThere is something in the same idea as renko charts, but much better (and immune to this problem). It's called directional change.
Disliked{quote} There is something in the same idea as renko charts, but much better (and immune to this problem). It's called directional change. https://arxiv.org/abs/0809.1040 I'm using this as input into my convolutional networks.Ignored
Disliked{quote}CRB are immune to the problem algoTraderJo describes because they "resynchronize" at every pivot.Ignored
Disliked{quote} Hmm, are you sure about that? The CRB construction rules that I found do not mention any kind of pivot/extreme point. They only care about range. So for example, if you start computing them in the middle of a constant up trend, when the trend finishes, the CRB would switch direction at different points, depending on exactly when it was started, so on average they will be "desynched" by half the range.Ignored
Disliked{quote} Hmm, are you sure about that? The CRB construction rules that I found do not mention any kind of pivot/extreme point. They only care about range. So for example, if you start computing them in the middle of a constant up trend, when the trend finishes, the CRB would switch direction at different points, depending on exactly when it was started, so on average they will be "desynched" by half the range.Ignored
Disliked{quote} PipMeUp is correct that constant range charts converge from different starting points. From my tests you can start multiple constant range charts at different points and in approximately 35 bars they will synchronize. The number of bars to synchronization may be a factor of the range chosen. Try the bar creation logic with multiple random starting points and you'll see that constant range charts converge.Ignored
Disliked{quote} I couldn't find the CRB creation logic clearly described anywhere, the general description that I found seems susceptible to sync problems. Doesn't matter anyway, they don't interest me. Directional change "bars" converge in exactly 1 bar.Ignored
DislikedHi algoTraderJo, it's really good to see you back on this thread with new machine learning ideas. I have specific question for you that bothers me a lot recently. When measuring DMB of a strategy that uses returns from multiple symbols as inputs, how do you bootstrap the symbol datasets? Is it safe to bootstrap the bar returns of all symbols independently (effectivelly destroying any correlation between symbols)? Doesn't this lead to underestimation of the datamining bias? Alternative approach is to group bar returns that occurred at particular...Ignored
DislikedThis is an algo I like, it has been trading for a bit more than a year now for me: {image} It uses a two algo linear classifier ensemble. I post the csv for this system, if anyone wants to explore it more. {file}Ignored
Thread Tools  Search this Thread 
