Hey, Hey, Hey Im searching for helping hands on a machine learning daytrading project. I want to try to create a daytrading algorithm based on reinforcement learning and dqn(the algorithm used for alphago) So the idea is to put in historical data and the program tests over and over again new strategies while maximizing ev. I want to use the openai framework gym, python and coffee. I created an experiment where the bot can buy, sell, do nothing and request 2 other stocks for intermarket analysis. The bot gets the day chart and time and so has to work on the chart. On the end of the day the positions get closed and the winnings are calculated and summed to total score. I dont have experience on deep nets. Someone with knowledge here would be a real help. First i would try to start with a convolutional net to detect combined patterns for example top + top same level = dopple top So bot is done too but not tested so far. I dont really expect it to work. But if not the project could be used to rank the chart of a stock to help with decision making. My main problem is i am too lazy So im searching for motivating people. If someone is interested i would put this project on git.
Hey , you are like me .I am very lazy too.I developed winning strategy years ago, but seems can't start trading it for whatever reason.So years going by, and I know I can make a lot of money, but first I need to save up 10 grand and open trading account .Maybe in couple years, I 'll do it.
I'm always interested and I have a lot of programming knowledge -- but if its on git, are we talking about a private repo? The last thing you want happening is designing a good system only to have it get stolen from someone who has the money and connections to run it on a network far better than yours. You can PM me if you'd like -- I'm currently working with Tensorflow myself. I'll write code in Python first and if the logic works, we can always rewrite in C / GoLang, etc.
Sounds like a good idea. Well i wouldnt mind to make it public. Better it works as it stays for years undone on my pc. But i could make it private too if we get 2-4 man together.