Trading bot

It's a pretty nice cyberpunk book, but I like The Stars My Destination more.

This

You faggots are going to get destroyed by a GS quant

PHP is fine, any database is fine. The money is in the algorithm and the algorithm will tell you what part of the stack needs to be optimized. Everything else is just bike shedding.
t.user making 1btc a week with a bot

Attached: Screenshot_20180502-081009.png (1536x2048, 112K)

>thinking all of the trade on an exchange in every kind of market is made by a normal human being but not bots.

Attached: 1524140971422.gif (268x268, 1.41M)

my trading bot sucks ass. plus keep cmd prompt open all the time.

>PHP is fine, any database is fine

you can't make this shit up

Okay clueless. Except I actually make money and some of my stack is literally written in php

php 7 should be fast enough unless you suck at dev, if multithreading is required then you should use go instead of c/c++

Of all the things you ask the most inane question

What you should be really asking is what trading strategy will your bot use

i ltierally bought a script but why do any research?
i hope its worth it.

>bought a script
>hope its worth it
Ask yourself this, if someone had a money printing machine why would they sell it instead of use it themselves? the rule of thumb is if its being sold, it doesn't work

Also randomly handling out API privileges is how you end up with a lot of VIA if you catch y drift

I wanna start off with - I don't know your expertise level, I'm only going off the info in the op. You're making a student/junior level developer mistake. The technologies you choose to do your work doesn't matter at all. Use z80 assembler transpiled to webASM to run on a python interpreter if you want.

Get the data in, munge it as you need, and push the result out. Use tech you like and are comfortable with.

they just sell it per percent. you have 10% a day scripts. and you have 1% a day scripts.

those are ponzis, they don't actually trade anything

a ponzi is different than a scam.

Would love to discuss things with other botters but as soon as you get to the gravy you're discussing things you really don't want to share.
Kind of paradoxical.

I know this feel. I still find it funny to be an annoying bastard, such as suggesting the following:
ARBITRAGE ARBITRAGE ARBITRAGE ARBITRAGE ARBITRAGE ARBITRAGE ARBITRAGE ARBITRAGE ARBITRAGE ARBITRAGE ARBITRAGE ARBITRAGE ARBITRAGE ARBITRAGE ARBITRAGE ARBITRAGE ARBITRAGE ARBITRAGE ARBITRAGE ARBITRAGE ARBITRAGE ARBITRAGE ARBITRAGE ARBITRAGE ARBITRAGE ARBITRAGE ARBITRAGE ARBITRAGE ARBITRAGE ARBITRAGE ARBITRAGE ARBITRAGE ARBITRAGE ARBITRAGE ARBITRAGE ARBITRAGE ARBITRAGE ARBITRAGE ARBITRAGE ARBITRAGE ARBITRAGE ARBITRAGE ARBITRAGE ARBITRAGE ARBITRAGE ARBITRAGE ARBITRAGE ARBITRAGE ARBITRAGE ARBITRAGE ARBITRAGE ARBITRAGE ARBITRAGE ARBITRAGE ARBITRAGE ARBITRAGE ARBITRAGE ARBITRAGE ARBITRAGE

Yeah, it really sucks. I trade very successfully with my bot but aside from discussing generalities like who has the best api, what language you're using, how many exchanges you're running on, shit like that, there's no way I'm going to spill the beans on what actually makes the money. And those generalities don't make a huge difference. The key really is the algorithm. I don't care if you code the shit in gbasic, use flat files and dial-up, the money is made where you buy and where you sell. Furthermore, commercial bots are trash, most TA is trash, and unless somebody is monumentally stupid, anything you read by a bot maker is also trash. Including this comment.

usually anything promising 1% a day pays you out with deposits from others, pure ponzi

sshhh pls don't say the a word

nodoby cares what you can code and how beautiful, well written it is, or in what languages.

the pudding is in the strategy, noone will tell you any part of their strategy.

>cmd prompt
>windows

But why?

I wrote an arbitrator bot, it can handle +10 exchanges, with all coins on them

my arbitrator bot can find cycles from any coin through any exchanges, any lengths, back to profit of same coin.
in less than 20ms.

it does not poll, the arbitrator only receives streams of data from exchanges.

yet, it does not make profit, due to exchanges "timely" timing my buy/sell requests out, sometimes a buy/sell-request is just stalled for +25s only then to complete with a loss. this happens on ALL exchanges ALL of them.

Thank you for all the answers, I think I will just go with Phalcon and MariaDB with normal API calls, instead of sockets. Maybe if algo needs a bit more speed will see into changing things up.

Are you the elusive one with the BASH bot?

Here's an idea (I think a good one too?) that I can share here that wont be bad if lots of people know of it:
>Design bot based off TA that runs on EtherDelta
>Creates a market order once a certain condition is met (e.g. RSI touches 10 or whatever)
>Encourage client to use one coin only
>Sell this bot to others claiming it's profitable. Get biased backtested results for proof
Now for the fun part:
>For same coins, constantly monitor the same indicator(s) that the sold bot monitors. As soon as the right conditions appear, the client's bot will place a limit order
>You know exactly when they will buy and sell, so you set your bot to place an extremely favourable limit order in the opposite direction that no sane person would trade manually.
>The client's bot also has the same price setting on their side and gets rekt

Fuck arbitrage, honestly. I spent 8 weeks scanning 25 (TWENTY FIVE) exchanges with hundreds of trading symbols and the results conclusively showed that the risk is not worth it for the slim profits gained. The high volatility that's good for arbitrage is the same volatility which makes it dodgy to get involved in the first place

Is there no other way to get that data? I was wondering if early stage pumps could be detected somehow, but... do I really need to scrap all this data myself for weeks to come?

No, you need to scan orderbooks, which isn't stored historically. In fact, historical data stores are rubbish usually as they often just show daily OHLC.
Here's what you do:
>Get a cheap server up and running that collects data every minute for a bunch of coins. Use cryptocompare.com/api/#-api-data-histominute- , you can get up to 2000 minutes in one call.
>While that's running, develop your bot
>Use those results to backtest your strategy
Don't know if you'll be able to find that data online.. I'm sure someone out there sells it or something

This isn't for orderbooks by the way, this is for 1min OHLC.

Now this is getting interesting

Thanks. Now that I think about it just selling collected data would make for a nice side income...

I recommend Python. I just finished writing one myself and there's lots of API modules that help.

I'm not reading this whole thread so sorry if someone else has suggested this, but I'll share what I did.

Deep Learning Convolutional Recursive Linear Regression Neural Network built using tensorflow in python. I use the java API for binance.

Make sure you use something like a 3 second wait time between trades or you will get banned. I trained it on the entire history of CMC (making a distinct beak in data when they removed Korea) and used nodes with relational algorithms so that it considers things like Market Cap and RSI and how all the charts work together. I also use some hard coded rules. My bot is rather inconsistent in earnings, but it almost never loses money. Yesterday, it earned about 17%. I'll answer questions.

Attached: higgs512pixel.png (512x512, 400K)

I'm a few months ahead of you on the top 100 so good luck with that

Attached: jungle.png (420x420, 131K)

I would use Python.. Lot's of great statistical/machine learning libs available, and you will need them. Use pandas for data exploration, numpy and scikit-learn to discover correlation.
About the trading strategy in a nutshell: ARIMA and simple "conventional" regression models will not work. I think your best bet would be to parse crypto forums, twitter accounts, ((Jow Forums maybe)) and then apply sentiment analysis/natural language processing algos to formalise the "public opinion" about a coin.
That is a proven strategy. It's hard pull off tho, because you would need to parse literally hundreds of gigabytes of textual data, + it's really fucking hard to backtest a model like that.
t. data engineer/big data dev

>rekking your clients
But why would anyone use your bot if it's bound to lose? they'll catch up eventually. It'll work for a while sure, but you'll have to keep a stream of shilling which will be your biggest expense. (Whether it is your time or funds)

How did you project the data for the convolution layers?
What loss method did you use?
What was your training objective?

What activation function did you use for hidden layers and for the output layer?

This

I wouldn't rekk them, not my thing. Hence I don't give a shit about saying the idea on Jow Forums, I'm sure there are enough scumbags here that would be willing to.

don't listen. the speed your software runs will affect your profit. especially in this market.

There's great bots on github.
I've experimented with a c++ one and nodejs frontend. C++ because sockets r lightning fast and you will get your orders in first the faster your software.

>I wouldn't rekk them, not my thing. Hence I don't give a shit about saying the idea on Jow Forums, I'm sure there are enough scumbags here that would be willing to.
My question wasn't about morality, but the potential profitability, you'll have to expend resources on marketing your bot even if it's shit. For a competent product you could track the success rate and occasionally fuck it up and pocket the revenue, but then you actually need a successful bot, and if you've got it why not just request more money for it.

The only people that would use the bot are people that are naive, don't know how to read source code, and have believed they've finally found a way they can make a quick buck. I wouldn't say they deserve it because people are inherently greedy and the victims will be mostly good folks who aren't as tech-savvy.

To answer your question, if you were to do this you'd need to focus on marketing the product to boomer crypto investors who think they've finally made it (there are lots of them believe me). If people start talking shit about your product online (which they eventually will), then you close shop and sell the exact same product under a different brand, voila problem solved! Bot doesn't have to be successful, people just need to BELIEVE it's successful. Just look at any ponzi scheme today that guarantees users get payments from a "trading bot", people are retarded enough to eat that up

bipolar is self regularizing, selu too.
I've had good results with crelu on other things.
The final layer ideally would not need activation to keep it in bounds

I use a weight averager to compare broad data relationships and normalize values for convolution layers. I have a large number of input nodes btw.

I used an L1 loss position regressor

looks like this:

S:=∑i=0n|yi−h(xi)|

training objective was to predict when a change in direction of a given percentage takes place. I trained it across various percentages using regression. It seems to be most profitable when observing differences. There are only 2 outputs for each coin's prediction. change direction or not. When that happens, it checks to see if it should buy or sell.

I'm on 2 different devices so just realized you tagged me. Yep, that's me. Stack keeps getting higher and higher. For the nonbelievers, bots work.

Why not Haskell?

If your bot depends on speed, prepare to get rekt. The big boys encode their algorithms on FPGA's that can execute an order before the data can even make it through the network stack on a regular PC. You are not ready

Godspeed. I like the simplicity of the objective.
I've been considering how a GAN could work on this type of data. There is a lot of recent work on aligning datasets and I think it could be formulated as an alignment.
I've managed to make alignment work without cycle loss on images but it's spotty at best. Soon.

Why haskell? The whole muh functional programing shit is such a fucking meme.
I use mainly python, because literally every tech supports it. If I need speed I use C or Go.

You guys are all going to get destroyed by me, because I am a NEET with programming skills.

Mind linking the github repo for these bots?

Attached: 3220FB497440476EB603EAED39CA31B3.jpg (1152x2048, 199K)

One thing I would like to discuss/compare findings on is that I'm certain some sites run their own bots or fake volume/order data. This is not whining because my bot is too slow, I get that some people jave faster/better algos, but there are some sites where my strategy simply never even yields 1 opportunity (statistically impossible).
Example would be liqui.io.
But don't want to discuss muh algorithms :)

*run their own bots that see orders before others

Thinking of starting on KuCoin & Binance first.

How good are the APIs?

damn this thread is still being discussed?

the only thing juicer than a retail trader is a retail trader with a bot

people like you make my company tons of money

HitBTC is notorious for this. Fortunately the aren't too aggressive.

heh years ago 2 girls I was hanging out with were drinking and we got to wrestling... long story short I ended up with her locking her legs around my neck and holding my face in her asscheeks and the other chick was laughing her ass off and shoving my face in the one girls cheeks deeper while sitting on my back. spent like 30mins like that.... wonder what those 2 are doing these days

Python obviously

Care to explain how?

A link to a good book or a scientific paper would suffice ;)

Thank your for your input. I felt like you could make giant stacks of money if you used machine learning + NLP (twitter, biz, reddit, wherever faggots talk about low cap crypto) to gauge shilling and then trade new token listing. Your toughts ?

currently have a python script making me .01 btc daily and i'm no programmer.

again, always listen to biz, they sure know what's best not to do.

if you got the logic, code up some api's in python or php. it's not that hard and all the api's out there have support for both.

In fx trading, exchanges are known to take the opposite positions of their clients. Seeing that >90% of traders lose money, they try to get maximum value out of them before they go broke. And most go broke fast. So the game is stacked against you, same in crypto. You have to find the good spots amd strategies.
Winning traders exist. But it's a challenge.

Watch this to see the stuff exchanges are up to
m.youtube.com/watch?v=oL_cLtw6WNM

Yo user, If your firm is really making bank and you are not profiting, hit me up @minminmax on Telegram. We are currently managing a porto of 25M and are willing to compensate you for any knowledge about markets you might have.. you could come work with us and make bank.

fuck typo: @ maxmaxmin