Trading BOTTTT

i also built a python bot, for bitmex. it's really slow desu

use c++ if you want speed op

What algos/logic are people using with their bots?

t. C++ masterrace

I am interested in talking to you about strategies you have tried and share ones I am looking into.

Is that an arbitrage bot? Looks interesting. How well does it perform factoring in slippage and fees?

>trying to compete in an unforgiving competition
>for higher order frequency limits “get in contact with us“
>has less than 1000BTC trading volume / day
>asks Veeky Forums how to

user, I got bad news for you...
A start would be to use a C language and don't use inefficient wrappers.
Then use an exchange with a low latency and ensure your own internet ping is top notch (I got fiber into the router

Talk away. I'm no expert, though -- still in tje stage of a high amount of learning.

For the time being, basically. It's the underlying framework for a trading engine, but it can be used for arbitrage (which is what I'm testing it against at the moment).

Fees are easy. Slippage is harder, but I mitigate the risk by testing each trade immediately before executing (and my path finding algorithm tries to take the qty available into account).

I've been writing a prediction NN for last few days and it's working really well. Where do you get your datasets from because I'm having trouble finding large historic data with better resolution than 1 day.

add me on skype: bruisedcrow

I'm also working with NNs -- I plan on using one of my mining rigs for training when the time comes.

I don't have skype (don't trust the software). Another way to get in touch?

chanmail (at) protonmail

Also re: slippage, if you look at the screenshot you'll see it calculates how much it can actually buy/sell using limit orders, and runs with that value. Notice the trades at different rates.

Sent you an email.

What NNs do you guys work with? CNN or LSTM variant?

invest in chainlink

I thought that skype name looked familiar -- you're the same person from That thread was also me. I was overly optimistic about gains, but it's still working so far.

I haven't gotten far with NNs in this context and I haven't evaluated any designs yet. CNN is what I'm the most familiar with, but I'm unsure as to whether or not they're appropriate.

I'm using GRU/CNN at mo for different types of input data

I'm thinking about doing it in java because MUH COMPONENTS, but I'm a Haskell Boi by nature so I'm not really sure.

Might go for the Haskell in the end.

>File: 1511974406721.jpg (49 KB, 768x1021)

golang/elixir

twitter.com/ramikawach/status/879405390118633472

But generally you need to collect your own data.
Write a python script that fetches exchange APIs at regular intervals and start archiving it.

Interesting. I have been working on implementing and testing different types of machine learning. May email you in the future

Nice find, thanks user.

>Write a python script that fetches exchange APIs at regular intervals and start archiving it.

This is basically what I've done, just not python.

Noice, do you happen to have binance historical data as well?

memories

Great dataset. Thanks.

>But generally you need to collect your own data.
Yeah I was afraid this was the case.

>Write a python script that fetches exchange APIs at regular intervals and start archiving it.
I've just bought a raspberry pi and have put ccxt (github.com/ccxt/ccxt) on it pulling every hour (not doing high freq. trading).

Coinigy seems to have huge amount of historical data but absurdly expensive to purchase.

Yeah you can spend 100 times more energy in building a highly optimized assembly bot over a python bot, but if the API has latencies of up to 2 seconds you will have wasted a shitload of time.

If any of you guys want to talk about your experience of building bots in a closer group: join this discord I just created discord.gg/nAZPtf.

I personally have experience in python, with:
*multiple ways to collect data which is accurate to about 15-30 seconds intervals
*using machine learning methods, like neural networks
*computing TA indicators

help a newb here, how is historical data of coins relevant for your daytrading bot?

Depending on what kind of calculation are you doing to produce signals.

They are using neural nets to do trading (which is low effort but low return).

They simply feed the network a bunch of previous trading data, then the network knows at each position the relative increase in price action per unit time.

The network then uses some fancy calculus to change its parameters to optimise itself so that in those same situations the bot would act similarly, hopefully making a profit.

Nope.

Training neural networks, backtesting your strategies, or both.

but that would be assuming the coin will behave the same as it did the previous time the price was in that range, and this is obviously not always the case. I guess the key is in those fancy calculus

Complete noob here, i can handle APIs and stuff like that on Python but i don't know how to put trading indicators on it, i heard about TAlib but i did not find anything on how to run it, any advice or source user?

yeah this is some advanced shit, guess i wont be programming bots until i learn some serious skills if i dont wanna lose all my btc

Start learning then.
Install python + tensorflow and start watching the AI pajeet videos:
youtube.com/channel/UCWN3xxRkmTPmbKwht9FuE5A

i thought pajeets were shit coders?

Surely having the bot run on a web server with an extremely fast connection would be so much better than any residential internet connection

Im a PHP web dev so I'm looking into creating a simple PHP based on to get started

>I believe generalities are axiomatic

I don't know if this thread is for you bud

>I'm a PHP web dev

I'm so sorry

i dont think this board is for you

>I'm so sorry
Why? I'm earning more than my peers

PHP just rends my soul in half man

make that money

Are you asking so you can do this yourself at home?

If so, you are dumb. Read FlashBoys. The banks literally drill through mountains in order to reduce latency by milliseconds. You can't compete on your own. If ifs to get a job, C++ and Python. Python for manipulating data for analysis and C++ to write systems.

I made a shitty bot with nodejs which works withbinance but then i found ccxt on github

Most crypto exchanges use cloudflare which ruins latency for everyone anyway.
Doesn't matter if you run it from home or rent a server in the same datacenter as the exchange, cloudflare will fuck everyone all the same.

Please elaborate. No way around this?

does ccxt have any built in algorithms though? I could make the same thing using an API, so is there any benefit to using this code?

Mine is basic php on a framework croned to load and do shit at given times. Not the best solution but I want control and info be available 24/7 for me on a web

>php on a framework
which framework? Laravel or Symphony?

C++ or fail.

I use machine learning for mine, its doing good so far. Look into CUDA

Just how much are you guys making running your bots? per day?

C++ is probably a must for HFT. Im not sure why you would get into that if you are trading crypto though. With more traditional assets, fair enough, but youll probably just end up losing money.

Just figure out normal algo trading first, python is all you need programming wise. For anybody wanting to break into algo trading, the programming is not that important. Make sure you know the math first. You can make good money with linear models and no offense but people talking about neural networks usually come from a software engineering background with next to no knowledge of the underlying math involved.

Which usually leads you towards a path of trying to fit a neural network on fairly random noisy data (happens a lot in academia for example) and draw some conclusions from that. You certainly can make neural networks work, for example CNNs based on googles wavenet infrastructure have consistently outperformed traditional spectral analysis for signal smoothing, etc and are fairly easy to implement in something like tensorflow. But remember, shit in, shit out. So know what you are doing first, neural networks are no magic bullet.

comp sci. student here. considering to get into algo trading.

how much time would it approx. take for me to learn C++?
have experience in Obj. C, Swift, Python and some C#

and also, how much money did you make with your bot?

oh and are you running it on a pi?

Maybe Im misunderstanding something, but why would you want to run something like this on a pi? I run everything on a dedi.

If you stick to modern C++ you should be able to pick it up fairly fast desu. But again, unless you really want to do HFT for some reason, python will be just fine.

I made a fuckton of money pairs trading ETFs when the going was good. When it comes to crypto, I focus on momentum strategies and have several models running at any given time. Averaging something like 30k-50k a month, not my main area of trading though so keep that in mind.

Just use Python, jesus. It's ideal.

My bots are in NodeJS, come at me. Love it.

This is not something you'd want to run on a Pi. You want it running 'physically' (latency) as close/fast to the exchange as possible, and on a machine with sufficient specs.

For NodeJS I ensure that CPU clock speed is the best available.

my dog need surgery :( would be nice if anyone could donate some btc

17PBGKqP69b7NnvPNQL9GKeNQhAkY1hBj3

I wrote a script for that in python also.