Deep Learning Trading Bot

Do you think this is possible? Why or why not? What factors would be the input variables for the bot?

Other urls found in this thread:

discord.gg/rk2RZt
twitter.com/NSFWRedditImage

well it would be A.I. that would examine pass trends, look at the web for sentiment of how things look and then look at books, determine whale manipulation, personally I think It can be done

I've thought about this. I have no idea how to do it though.

I figure that if it worked, I would assume that multi-billion dollar hedge funds would've already made one. I mean they already have HFT bots. Pretty sure it would be common sense to then make an AI HFT bot.

They may be wary of resting millions (billions) of capital at the whim of a program that could get shrekt by the next flash crash or Black Monday

they probably have programs that are like the OP is alluding to though, not really complicated depending on your definition of "deep learning"

anyone have a copypasta of the code behind AlphaGO or something? lol

Pretty sure all you'd have to do is feed it historical data, give it options, set rewards on outcomes.... then wait 6 months while it learns.

HFT is not so much about complexity but speed. If you have code than on average returns positive gains, that's pretty much everything you need, besides a 4 million facility filled with thousands of computers to perform trades at an insane rate.

surely you could program into the deep learning ai to only handle $x

I think the problem is more of what it learns to do with the $. Once you let it stray from your original directions, it's really doing what ever it see's fit. The people that made AlphaGO don't understand the reasoning behind all of its chosen moves. That kind of uncertainty doesn't exist in HFT as it is, but would exist with AI trading.

how does one find whale manipulation. the sentiment is actually really easy sites already do this, it just needs to be timed. Also how would you find whale manipulation?

normies don't care about altcoins that much yet.

they would rather play the stock market and get a bailout from daddy

more hard than that

i'm not looking for diverse coins, just the APIs for bittrex and coinmarketcap

its not that smart yet lulz

fuck man. Here I thought this wasn't a crypto thread.

We have a bit /bot/ discord if any of you want in.

discord.gg/rk2RZt

do u think a simple amazon ec2 instance or google compute engine would be sufficient

also which coin do you think would be easiest to trac and predic

DL researcher here, this is exactly what I would if I had time

It's possible if you have information on basically every single trader on earth and their likely movements (ie Google can and is likely doing this)

Has a tendency to overfit.

The real problem with all crypto is getting clean data. You can't just fit to the straight price vs time, too much noise. You need to have some signals (sentiment, trading volume, deal book state) that drive the price in a tangible way and model off of those. Then layer noise model on top of that

What do you mean by noise?

Price movement that is essentially random and cannot be attributed to any information. May fit statistical distributions. Sussing out the signal from the noise BEFORE putting it in the model is the hard part.

I may build a trading bot with various buzzwords. Might not actually work, but people may pay to use it

>Price movement that is essentially random

Can price movement ever be considered 'random'?

they can just do that then have a real human have the final say

??????? Does anyone in this thread read anything other than Veeky Forums. There's already AI bots that focus on deep learning. They only use and learn TA at the moment, and there results are entirely crushing humans.
FA is a lot harder to implement.

link ?