Id like to know how many developers here know about Machine Learning so i can post relevant content and ask questions !
still waiting !!!
I am very interested in ML, but I’m nowhere near knowledgeable!! Big noob here! 
I’ve been working on my own trading platform since started using alpaca, I’ve also been getting ready to release a public version fairly soon. had alpaca look over my source code since everything is open source to my surprise they actually complimented me on it. But before I release I want to implement machine learning and a localized mobile app I’m about halfway there but it is making profitable trades. I have some questions for certain people who understand regression and least squares though…
WOW! That sounds great! You’re way beyond me. I would love to look at you’re code and try to understand what it is doing though.
Hopefully sometime soon I’ll have it released I just don’t want people to use it improperly and I have been working on a way to prevent that. There’s going to be a lot of iterations of the software as I see the stock market is constantly changing or should I say the way stocks are trading is constantly changing but that is kind of expected. There’s a lot of things that you have to account for when programming machine learning like obviously the momentum of a stock moving downwards versus upwards. But one thing that’s really a indecisive deciding factor for me with programming is whether or not to base its machine learning from historical price movements or from market information on the stock, like if the stock market as a whole is moving downwards or upwards. I ended up making this all because I didn’t really see anything specifically meant for day traders who have a limited amount of Trades I’m sure there’s probably something with backtrader that’s native for everyone who programs in Python but for the love of PHP I had to make it on my own.
To anyone following this thread, there’s a lot of information in php already out there and if you want to start by looking at what’s the best for others i suggest looking at the following links
https://php-ml.readthedocs.io/en/latest/
i’ve also been making a auto day trading system
(Here’s a link to my alpha was only meant for facebook friends
- so ill need to replace the link at some point )
https://github.com/givemesite/GMS-AT-OpenSource.V1.2#h.svwf56zhexe7
(IF YOU USE THIS LINK MAKE SURE YOU UPDATE - Check every night !)
(This link is outdated and will be replaced with a BETA that will have the following added)
Emulation
Mobile app (with online setup)
By the end of 2020
I have been using Genetic Algorithms for years to tune my momentum parameters, buy parameters and stop losses… My code is all custom C# code and I have about 50 strategies that I tune. I am now ready to try on-line trading and am currently working with the paper trading APIs.
How long have you been in the game?
I have been using the latest GA for the last 5 Years. I start with an idea for a stock picker and it usually has 10-20 parameters. Rather than guessing at values foe for the parameters, I set up a GA to evolve the best parameters. this used to take about 24 hours to get a good GA run (5000 individuals x 30 generations) and with improved GA, it now takes about 1 hour running C# and a lot of multi-threading. I have 84 algorithms running now all tuned with the GA. These algos are run on daily data.
Lately I have been working with 1 minute data and tuning the algos and the buying and stop-loss parameters. these are now showing very good results so my current project is to move to Alpaca paper trading to try and nail down slippage, etc. Once the paper trading proves itself, I will move to a real trading account. the algos should trade several stocks per day where some with stop out quickly but some I will hold up to a month as long as they are moving up.
Thanks for your interest.
so you been in about 3 years more than me, im interested in what ML you use and how you use it. if your ok with sharing. i’ve been working on a open source project i just got done implementing ML and a mobile platform to use it. ive been making it so it fully automated no need to interfere. i made my own simulator to code when the markets are closed. i have other things im interested in knowing if your cool with sharing, like whats your success rate
I coded the Genetic Algorithm (GA) myself. I am old-school and want to know have things actually work and not just assembly parts from other places. The results is that the GA is now about 25 times faster than the original code. The GA algo is also a hybrid routine. With a GA, you make a run with all your individuals for a single generation. Then breed the bet for the next generation. But before I breed, I expand the neighbors of the best individual in a recursive decent algo chasing a better individual. e.g. if the best individual (highest profit) uses a 14 day RSI, then I will recursively try 13 days and 15 days to see if any better. This continues until the results gets worst. I may have to check another 100 individuals from the best one of my 5000 individuals. I usually go out 30 generations. I have tried all the combination of individuals 1000 - 100,000 and generations 10-100 and 5000 x 30 converges within an hour of processing time pretty well.
PS: I use the GA instead of a neural net because the answers are understandable and explainable instead of NNs which are just matrixes of numbers. I think of them as the same thing just NNS run from results to a matrix of parameters and the GA runs from the parameters to the results.
The reason to start paper trading now is the results are showing 40-80% expected returns and I would like to verify with Alpaca trades.
So, are you running ML code now and how is it going.
i’m running ML in both test and production at the moment, my results vary day by day im in a uptrend at the moment
i had talked briefly on my results in this post, how long does one emulation take for you?
A full GA run (5000 individuals x 30 generations) take 60-90 minutes. And in that time I have reached diminishing returns (e.g. most likely the best results).
I’m gonna take a wild guess and say you’re talking about data points when you mentioned individuals and generations would be training sets? if so im about the same. have you been live tradeing yet ?
In GA an individual is a full algorithm for picking stocks. One individual(algo)'s run is thru all the stocks I have, 8000, in my database of end-of-day data to see if a stock qualifies for purchase and any purchased are run until they limit or stop out to get a profit or loss. A generation is all the individuals have been run against all the stocks (8000x5000). Then the best individuals(e.g. algorithm) are breed together (this takes some parameters from one algo and some parameters for another algo to create a child algo). So 5000 new algos are created and the process continues for another generation. So a run is 8000 stocks x 5000 individuals x 30 generations = 1,200,000,000 tests of individual stocks.
See https://simple.wikipedia.org/wiki/Genetic_algorithm for an explanation.
Have you tested it live ?
Not yet. That is what Alpaca is for. I am working on that.
i’ve tested live, just be safe for the first few months and remember that
live is a little different since its not affecting the market the way moving big money into a stock in live testing would.
You are right, live is different. Looking forward to it. I have been buying and holding lately with the market going up so much. Hard to keep ahead of a market with lots of small trades when holding makes so much. My holds have been making about 15% above the market so difficult to do that trading all the time. But am looking forward to trading on Alpaca as my goal is to aim for 5% per month.