Currency trading Forex (Ultra deep neural netowrk) Project
Hello everyone!
For past 2 years I've been now working on a project an automated currency trader
what (should) over perform a human. My project is almost finish, whats left is to code
and I need help with it. Soon my internet access and over all computer access will be fairly limited and I'm kind of a hurry finishing it.
This is non paid request but If one helps me finish my vision, one will be able to
use it and will own all of its rights. It can easily replace full time job and
this is not an scam or dream, my work is based on facts and real scientific data.
To public I can tell only a little about it but I can mention few things
what this job will be about.
Whole project uses standard c++/11 libraries and the only extension we'll be using is cuda. I've build a model what uses different types of neural networks
to predict when to buy or sell currency and it does so in 1 minute timeframe/tick. I've developed my own strategy what we'll be training networks to trade with.
You see, neural networks are powerful but you can't simply just try to teach them by just showing them the past and 'future' and make them try to predict some time span.
I don't think it can be done. Most successful projects i've seen around, they predict trend, 'average prices'. Never can be the noise itself be predicted however, it is tradable if you do it right.
Well imagine putting a moving average , period being 60 and the shift will be -30, you can see that it could be quite easy sometimes to predict that moving average line and when you look more into it you'll see that price what is grossing that line, will eventually almost always coming back near
the moving average line. Ofcourse when you do the shift then you'll be having 30 candles in front empty, without guide. See this is where one of the networks of mine comes in game.
Its a special kind of recurrent neural networks. Networks are also using forex calendar and instead of handling price as candles, the price is just a line without high/low because in 1 minute timeframe, distance between high and low strangely is the distance between ask and price. ( 10-30 points )
We'll be aiming more towards 100> points move and high/low will be assumption in the code.
To be able to participate you must be at least a year or 2 experienced c++ programmer and ofcourse cuda and gpgpu is needed since training neural networks in cpu would take well just too long.
I'll be taking 2 people at most because I really want to keep it low.
I've actually made some progress in deep learning field that I wish to keep out of the public for a while. I can't even mention here with what am I working with exactly, what field of deep learning but everything you need to know will be through private communication.
You do not need to know anything about math, you just need to follow my instructions, everything is already set up.
If you're interested then send me a PM and for head up, I might test your knowledge on programming/cuda field.
Uh and most importantly, I've coded my own simple api what is even more simpler to use, for example plotting graphs or displaying pixels and some math sort of math lib as well. The code size of the project is right now about perhaps 90kb.
It is also linux friendly but then some of the features of my API will be disabled because I was working on windows environment mostly.
Thank you for your attention,
I'll be looking forward to your private message
- Free trading apps
- Over 8,000 signals for copying
- Economic news for exploring financial markets
You agree to website policy and terms of use
Hello everyone!
It should be taking 1-3 trades per day and with 85-90% win/lost ratio. My best results so far before all those huge updates I made in model was over 75% but that did not satisfy me.For past 2 years I've been now working on a project an automated currency trader
what (should) over perform a human. My project is almost finish, whats left is to code
and I need help with it. Soon my internet access and over all computer access will be fairly limited and I'm kind of a hurry finishing it.
This is non paid request but If one helps me finish my vision, one will be able to
use it and will own all of its rights. It can easily replace full time job and
this is not an scam or dream, my work is based on facts and real scientific data.
To public I can tell only a little about it but I can mention few things
what this job will be about.
Whole project uses standard c++/11 libraries and the only extension we'll be using is cuda. I've build a model what uses different types of neural networks
to predict when to buy or sell currency and it does so in 1 minute timeframe/tick. I've developed my own strategy what we'll be training networks to trade with.
You see, neural networks are powerful but you can't simply just try to teach them by just showing them the past and 'future' and make them try to predict some time span.
I don't think it can be done. Most successful projects i've seen around, they predict trend, 'average prices'. Never can be the noise itself be predicted however, it is tradable if you do it right.
Well imagine putting a moving average , period being 60 and the shift will be -30, you can see that it could be quite easy sometimes to predict that moving average line and when you look more into it you'll see that price what is grossing that line, will eventually almost always coming back near
the moving average line. Ofcourse when you do the shift then you'll be having 30 candles in front empty, without guide. See this is where one of the networks of mine comes in game.
Its a special kind of recurrent neural networks. Networks are also using forex calendar and instead of handling price as candles, the price is just a line without high/low because in 1 minute timeframe, distance between high and low strangely is the distance between ask and price. ( 10-30 points )
We'll be aiming more towards 100> points move and high/low will be assumption in the code.
To be able to participate you must be at least a year or 2 experienced c++ programmer and ofcourse cuda and gpgpu is needed since training neural networks in cpu would take well just too long.
I'll be taking 2 people at most because I really want to keep it low.
I've actually made some progress in deep learning field that I wish to keep out of the public for a while. I can't even mention here with what am I working with exactly, what field of deep learning but everything you need to know will be through private communication.
You do not need to know anything about math, you just need to follow my instructions, everything is already set up.
If you're interested then send me a PM and for head up, I might test your knowledge on programming/cuda field.
Uh and most importantly, I've coded my own simple api what is even more simpler to use, for example plotting graphs or displaying pixels and some math sort of math lib as well. The code size of the project is right now about perhaps 90kb.
It is also linux friendly but then some of the features of my API will be disabled because I was working on windows environment mostly.
Thank you for your attention,
I'll be looking forward to your private message