Machine Learning

Any updates on machine learning software? I tried tssb but no luck and actually hard to use. Price Action Lab is user-friendly and I use it for scans and swing trading but I'm looking for a way to use its patterns with a genetic engine to configure the best price action strategy. I think this idea has a lot of potential. Has anyone used genetic software to combine a large number of sub-systems? This will be my next major project. I have already found two CS students to work with me part-time.
 
Hi Solas0077,

I was reading through these posts as I am using GA to develop my own strategies. Could you please explain to me what you mean with "sub-systems"?

Has anyone used genetic software to combine a large number of sub-systems? This will be my next major project. I have already found two CS students to work with me part-time.

I am asking this because I am using a platform that uses sub-systems (or code snippets/trading ideas) in combination with GA to create new trading systems. This helps me allot as I always see other GA software as more blackbox and "as a metaphor" using a machine gun that shoots allot of blanks to see if it can hit a moving target. While if I use my own code snippets I know that only MY ideas are used, which is more in line with MY trading style. And it makes the whole process far less, voodooish.

These are purely my experiences.
 
Hi Solas0077,

I was reading through these posts as I am using GA to develop my own strategies. Could you please explain to me what you mean with "sub-systems"?

I am asking this because I am using a platform that uses sub-systems (or code snippets/trading ideas) in combination with GA to create new trading systems. This helps me allot as I always see other GA software as more blackbox and "as a metaphor" using a machine gun that shoots allot of blanks to see if it can hit a moving target. While if I use my own code snippets I know that only MY ideas are used, which is more in line with MY trading style. And it makes the whole process far less, voodooish.

These are purely my experiences.

Interesting. Could you say what software is that?

Sub-systems: every system can be thought of as a sub-system of a larger system. For example different candlesticks can be thought of as sub-systems. price Action Lab generated a file with similar sub-systems and you can use each one of them as a signal generator but it is more interesting to use genetic algorithms to find ways of combining those signals inn more affective ways, In am working with someone now and we try to formulate this as a function with binary weights and the job of the genetic algo will be to determine their values, 1 or 0.
 
Interesting. Could you say what software is that?

Sub-systems: every system can be thought of as a sub-system of a larger system. For example different candlesticks can be thought of as sub-systems. price Action Lab generated a file with similar sub-systems and you can use each one of them as a signal generator but it is more interesting to use genetic algorithms to find ways of combining those signals inn more affective ways, In am working with someone now and we try to formulate this as a function with binary weights and the job of the genetic algo will be to determine their values, 1 or 0.


So, if I am correct, you are using GA to combine trading signals. I might have worked on something similar (maybe). I was part of a group of traders that had allot of trading ideas and not allot of time. So, we build a GA to combine our trading ideas. Where each idea was a sub-system. And it was not the code that was generated, but the strategies were.

For them, optimizing the process of creating an algorithm was more important than the trading ideas. It is now being implemented by a startup company.

You can find more information about that, at this website: Quantler.com

Are you working on something similar?
 
Last edited:
So, if I am correct, you are using GA to combine trading signals. I might have worked on something similar (maybe). I was part of a group of traders that had allot of trading ideas and not allot of time. So, we build a GA to combine our trading ideas. Where each idea was a sub-system. And it was not the code that was generated, but the strategies were.

For them, optimizing the process of creating an algorithm was more important than the trading ideas. It is now being implemented by a startup company.

You can find more information about that, at this website: Quantler.com

Are you working on something similar?

Thanks but I avoid web based systems because I am concerned with security as something can go wrong even if the owners try to protect it. See what happened to Target and many other companies.

Someone that finances my work purchased a special version of Price Action Lab. We are running 32 program instances on 4 servers with an Intel Zeon E5, 8 cores each and mine the data.. We are developing a genetic algorithm to combined the subsystems generated by all instances to final trading system based on some metrics we use. One of the metrics demands that the final system must perform well across many different markets. It's an interesting project with some other aspects I'm not allowed to discuss. I can tell you though that preliminary results show that plenty of non-arbitraged opportunities still exist in the markets.
 
Thanks but I avoid web based systems because I am concerned with security as something can go wrong even if the owners try to protect it. See what happened to Target and many other companies.

Someone that finances my work purchased a special version of Price Action Lab. We are running 32 program instances on 4 servers with an Intel Zeon E5, 8 cores each and mine the data.. We are developing a genetic algorithm to combined the subsystems generated by all instances to final trading system based on some metrics we use. One of the metrics demands that the final system must perform well across many different markets. It's an interesting project with some other aspects I'm not allowed to discuss. I can tell you though that preliminary results show that plenty of non-arbitraged opportunities still exist in the markets.

Why not take advantage of beta testers you can find here ? Probably for free and willing ?
 
We are running 32 program instances on 4 servers with an Intel Zeon E5, 8 cores each and mine the data..

No parallelism in your models?

One of the metrics demands that the final system must perform well across many different markets.

A fitness function that includes cross validation among multiple samples?

It's an interesting project with some other aspects I'm not allowed to discuss.

No Beta testers? (n)

I can tell you though that preliminary results show that plenty of non-arbitraged opportunities still exist in the markets.

I Agree.
 
results ??

Thanks but I avoid web based systems because I am concerned with security as something can go wrong even if the owners try to protect it. See what happened to Target and many other companies.

Someone that finances my work purchased a special version of Price Action Lab. We are running 32 program instances on 4 servers with an Intel Zeon E5, 8 cores each and mine the data.. We are developing a genetic algorithm to combined the subsystems generated by all instances to final trading system based on some metrics we use. One of the metrics demands that the final system must perform well across many different markets. It's an interesting project with some other aspects I'm not allowed to discuss. I can tell you though that preliminary results show that plenty of non-arbitraged opportunities still exist in the markets.

so can you post some out of sample results ??

Krzysztof
 
Why not take advantage of beta testers you can find here ? Probably for free and willing ?

It's a good idea but it's not up to me because I don't finance the project. But in my next one I'll do that and will cost less for sure.
 
No parallelism in your models?

A fitness function that includes cross validation among multiple samples?

No Beta testers? (n)

Each of the 32 instances works with different timeframes. It's also a little more complicated than that. For a simple example check the Price Action Lab Blog article about "taking advantage of multiple CPU core..." that is at the bottom of the most popular list.

Yes, cross validation among multiple securities. We are trying to see if we can discover true inefficiencies in data that work across many markets and timeframes. One of the guys working on this is also planning to use part of the data for his Doctorate thesis.

Beta testers? No needed now as we are not planning to sell anything to the public. It's all proprietary work.
 
so can you post some out of sample results ??

Krzysztof

Krzysztof,

Any trader who gets over excited about in sample results is an algorithmic trading noob. The dude has a mysterious cabal of project investors contributing funds to purchase an off the shelf software (price action lab) and a few computers to run it on.
He has basically articulated verbatim what is stated on the pal website.

Additionally, he states he doesn't favor genetic algos in one post because of the introduction of the data bias only to write 'a few weeks after' that he uses GA. In 3 to 4 months time he will go quiet.

Sorry for the outburst, reading about so-called traders bragging about the discovery of the next big thing wthout the smallest piece of evidence is infuriating. If I could trades in the opposite direction of these braggers I would quickly amass a Zuck' level fortune.
 
Krzysztof,

Any trader who gets over excited about in sample results is an algorithmic trading noob. The dude has a mysterious cabal of project investors contributing funds to purchase an off the shelf software (price action lab) and a few computers to run it on.
He has basically articulated verbatim what is stated on the pal website.

Additionally, he states he doesn't favor genetic algos in one post because of the introduction of the data bias only to write 'a few weeks after' that he uses GA. In 3 to 4 months time he will go quiet.

Sorry for the outburst, reading about so-called traders bragging about the discovery of the next big thing wthout the smallest piece of evidence is infuriating. If I could trades in the opposite direction of these braggers I would quickly amass a Zuck' level fortune.

I know that, I was just polite...

Krzysztof
 
I know that, I was just polite...

Krzysztof

(y)

Anyway, regarding DeepThought: I downloaded the demo and spent a weekend exploring it. I managed to contruct a number of models that were profitable ON PAPER (this means very little). I initially used the model templates provided with the demo and was able to run the first one after a couple of hours. After this I was customising the templates etc.

Reading the manual is a prerequisite to software operation. The manual is written excellently so the contents are very easy to read but this is extremely undermined by the structure of the manual. To get up and running quickly the structure of the manual makes it necessary to move from a page at the beginning to a page at the end to a page in the middle, beginning, end, middle, beginning, back, beginning etc. This is exacerbated by the need to type into a commandline and the vagaries of this anachronistic technology.

I come from the DOS era so being thrown back to an earlier age was quite nostalgic so the commandline interface wasn't over burdensome. However, I can write code in a variety of languages and can still remember alot of the old dos commands and I'm a finance and AI academic so my background is complementary. A user without a complentary background will find the DeepThought experience extremely daunting and who knows if the experience will even be marginally profitable?

The manual's structure needs to be urgently revisited. I recommend that the vendor review the TSSB manual written by Aronson. A comprehensive hands on tutorial should be at the beginning of the book. The tutorial should cross reference the main body for users requiring a more thorough explanation of concepts.

A word of advice: there is no substitute for the skills and market knowledge gleaned by manually trading. These skills and knowledge add 'invaluable' value when constructing algos then LEAVING them to run.
 
(y)

Anyway, regarding DeepThought: I downloaded the demo and spent a weekend exploring it. I managed to contruct a number of models that were profitable ON PAPER (this means very little). I initially used the model templates provided with the demo and was able to run the first one after a couple of hours. After this I was customising the templates etc.

Reading the manual is a prerequisite to software operation. The manual is written excellently so the contents are very easy to read but this is extremely undermined by the structure of the manual. To get up and running quickly the structure of the manual makes it necessary to move from a page at the beginning to a page at the end to a page in the middle, beginning, end, middle, beginning, back, beginning etc. This is exacerbated by the need to type into a commandline and the vagaries of this anachronistic technology.

I come from the DOS era so being thrown back to an earlier age was quite nostalgic so the commandline interface wasn't over burdensome. However, I can write code in a variety of languages and can still remember alot of the old dos commands and I'm a finance and AI academic so my background is complementary. A user without a complentary background will find the DeepThought experience extremely daunting and who knows if the experience will even be marginally profitable?

The manual's structure needs to be urgently revisited. I recommend that the vendor review the TSSB manual written by Aronson. A comprehensive hands on tutorial should be at the beginning of the book. The tutorial should cross reference the main body for users requiring a more thorough explanation of concepts.

A word of advice: there is no substitute for the skills and market knowledge gleaned by manually trading. These skills and knowledge add 'invaluable' value when constructing algos then LEAVING them to run.

Doesn't quite look likes it's going to take over the markets just yet then ?
If and when one does, manual traders like me will be redundant.
 
Doesn't quite look likes it's going to take over the markets just yet then ?
If and when one does, manual traders like me will be redundant.

No amount of technology is better than sound trader judgement which has been honed via years of experience!

However, traders with sound trader judgement aren't psychologically wired the same as the majority of people so even if one of the majority can read a market profile chart like a trading God they'll probably not profit from this gift via manual trading because they can't overcome the trading demons whispering in their ear: "you are wrong, you will lose, take that profit now, that losing position will turn in to loss eventually so delete the stop-loss, you've been correct 20 time in a row so LOAD 50% of your capital on the next trade etc etc etc.

The more robots trading the market, the more opportunity for the manual trader. This is counter-intuitive (just like much of manual trading) but trading robots make the market more efficient for OTHER trading robots but NOT for manual traders.

So when I wrote" A word of advice: there is no substitute for the skills and market knowledge gleaned by manually trading. These skills and knowledge add 'invaluable' value when constructing algos then LEAVING them to run." I am not advising new traders to risk serious money...open a micro account with £20 and trade for pennies just to get manual trading exposure and knowledge.
 
Krzysztof,

Additionally, he states he doesn't favor genetic algos in one post because of the introduction of the data bias only to write 'a few weeks after' that he uses GA. In 3 to 4 months time he will go quiet.

If you're taking about me I ensure you that I will be developing systems and trading long after you will have given up.

If you are taking about PAL software it doesn't use any GA or GP. This is done by simple programs than shuffle indicators. I suggest you get up to date. Also, envy do not help a man's life.
 
Top