Bot bro here

Any predictions yet for today?

really interested to see your predictions. I have a friend working on making a bot for me. something similar to what you have.

>using a bot to do anything besides pure arbitrage or liquidity trades
You could be retarded lad

yeah isn't that the point of the thread?

also realize it is extremely dubious for us to assume you are running a bot and not just blowing smoke up our arses

What's your accuracy rate? Rate of failure/dips? Do you have any place to follow your bot other than Veeky Forums?

No, nothing found yet. Guess you can't force it.

I'm doing this experimental bot just out of curiosity.

Felt like 3/4 accurate when you account for some coins I manually ruled out after the bot suggested them (like very low volume coins with graphs that don't make any sense).
No other place. I am just running it the 2nd time publicly. Still too innacurate and too early to share or make an official thing out of it.

I only ran it once before publicly (see thread linked in OP).
It suggested SYS (correctly) and XVC (incorrectly). But a quick look at XVCs price chart was enough to see that it's all over the place and probs not true.

yo, OP. interesting...I don't want your predictions but would love to learn more about how you built it and maybe collaborate? I have a few years python experience. anyway to contact you off here?

>a few years python experience

nice meme

Python was released in 1991 ;^)

That's pretty awesome, OP. Thanks for the service. I'll follow your posts.

I don't really feel like sharing my IP or half finished bot. I came quite far and would like to optimise it myself for now; I hope you understand.

See this reply of mine for its basic way how it works. It is quite easy to code. Just call the JSON files from bittrex' API page (they have a good documentation there).
Then use the values of interest and do some maths with them to compare / take averages / filter, etc.
The “difficult“ bit is to adjust and optimise those values to filter the right coins.

so? there's no fucking value to having "a few years python experience" no matter how many years it's been out

what exchange has a 5 second volume indicator in its API?

how is there no value? are you retarded?

Thanks. Although I'm sorry today has been disappointing so far. No notifications so far. Could be because bitcoin is rising and alts are dropping today.
I will continue to make these threads tho over the next few days and weeks

let me put it into perspective for you.

if you take a two or three year "programming" course at a college, you're still fucking useless.

so why should i care if some inbred has been playing with python in his spare time for a couple of years?

Not quite sure of the timeframe the API gives out tbqh.
I just check the bittrex /getmarketsummaries/ one every 5 seconds and they are different every time. Adding them up and taking the average every 15 min has been working quite well so far. Not sure if it is 1s or 1m though what the actual value returns

cool. that'll be handy.

man just thinking back to all the garbage code that high school kids were writing with "a few years Pascal experience"

and then you remember that Python is supposed to be the new Pascal

ok dude, this is the part where i burst your bubble

the volume number the bittrex api gives you is 24 hour volume

that means it is wildly impacted by things that happened before you even started watching

Thanks for the effort opie

I am not a good programmer either. Did it a bit in uni and taught myself a bit, too.
As said above you don't need a lot of knowledge to call a REST API and read out a JSON file. I used a bit of google and managed my way through

yeah that's fine, i'm not saying experience is bad, but it's not a qualification in and of itself either, and the sheer number of "learn how to program in 48 hours" ads on Facebook is super embarrassing

people think you can be useful and casual at the same time and it doesn't work that way

i work in IT and manage multiple production redhat servers and network infrastructure. I'm far from some inbred

No that isn't true. It literally changes the value a lot every time I call it (like x2, x0.5). That wouldn't make sense with a 24h value.
It is also completely different to the 24h value that blockfolio shows.
I will look into it and try to figure it out what exactly the value is

the ENTIRE 24 hour volume across ALL exchanges for FirstBlood is just over 420 BTC

are you telling me that you really believe that 140 BTC is being transacted every 5 seconds on bittrex?

you literally got mindfucked by a pump and dump that happened yesterday, that's all

jesus you're a dick

what in that post is dickish to you? the word "fuck"? do you love jesus or something?

Hmm possibly... it would make the bot more inaccurate but so far the win/fail ratio is still fairly good.
Do you have any source that it is the 24h data?

i looked for it in their documentation to be honest but i never found it, it's just obvious if you look at the numbers yourself, try it and it will add up that way

i wanted that data to be useful but in the end there were no strong correlations that i could find

it's like, you want to know that a pump is coming, so you watch the volume, and you see it's kind of steady and then suddenly drops by a lot

so you conclude there's no pump coming

but then you miss the pump and you realize there was another pump yesterday that just slid out of the window and crashed the 24 hour volume through the floor

It maked probably sense and seems in the right area. Still a bit weird that blockfolio often shows very different values.
Anyways, I got a filter in place that checks the price 24h ago and makes sure there was no major pump or dump going on.
I have multiple filters like this to improve the accuracy of the bot but yeah I guess there is still quite a chance that it can fucl up. (Like my last XVC suggestion).

It is probably used really only as suggestion tool. Look at the last few days or even weeks of a suggested coin and make a decision yourself.
Still much better than trying to find stuff like this yourself

Finally got my bot working with a strategy that is ~70%/month profitable.

If I had got into bot trading a year ago I'd be so fucking rich... The sideways volatility on some coins 12-5 months ago must have made some people very wealthy.

Long shot, but does anyone here have Bittrex data from the last year they'd be willing to share?

pull volume from another data source?

The objections about volume are right. Bittrex returns you sum of all transactions' volume from last 24H. That's why I gave up with my attempt at writing such a bot. However, you may write a fix that goes over an orderbook every second and actualizes volume - that's however a bit more harder and error-prone than you might think - you'll understand difficulties during first tests.

Any suggestions? Thw difficulty would be to find one that has all or most the bittrex coins + a representative and reliable volume going.

At the moment the worst that can happen is that it either misses a pump or it suggests a coin with a unreasonable and random behaviour. But that can easily be discarded by looking at the last few days manually before deciding to buy in

>>database volume data over time
>>compare db volume data to bittrex 24 hour data
>>come up with ratio (divide one over the other)
>>watch ratio as indicator for volume changes

Well my bot kinda does that by comparing it every 15 mins. But how would you rule out the change didn't happen 24 hours?
So far I just make it more unlikely by putting extra filters in place where more values are getting compared

you can do this in like 15 lines of python (excluding api wrapper).

I collect coinmktcap data for all coin volumes from the coinmktcap api every ten minutes.

the ratio solves your problem

a possible problem with this is you will get banned from bittrex if you hit their API too aggressively for a sustained amount of time

Coinmarketcap also shows the last 24h doesn't it?

I'll have a think about your ratio suggestion and try about with it. Maybe I use it as additional filter to improve the accuracy.

make an algorithm inducing randoms that mitigate this.

Source? I am not aware of any restrictions of the public API

pro tip: there are many sources of free proxies that bittrex don't blacklist...

wow i never realised coinmarketcap update tickers with that sort of resolution... useful

Yup, I see that at least three of us in this thread have had tried to write bot for Trex.

then you're adding RNG to your analysis

i didn't mean to say that they will detect you querying at fixed periods, just that they don't like receiving too many requests in a certain timeframe

again, i don't think this is clearly stated in their documentation, but people have asked in their slack or discord or whatever numerous times and they've gotten vague but consistent responses along the lines i laid out

yeah so now you're adding more unknowns to what you'd like to be as consistent and reliable a dataset as you can get, i wouldn't want to have to worry about variable latency or the proxy going down at night while i'm collecting data or whatever

>We are currently restricting orders to 500 open orders and 200,000 orders a day.

>I collect coinmktcap data for all coin volumes from the coinmktcap api every ten minutes.
Smart guy.

essentially you are saying we are using volume changes as an indicator of volatility...

one of the greatest measures to use when daytrading stocks/cryptos is to look at the 52 week high over the 52 week low (certain time period high, over same certain time period low), make a ratio off that, and trade the stocks/cryptos that have the closest ratio to 1.
based on this H over L volatility make valuations -> sell overvalued ones, buy undervalued ones

this trading method, for sure gains, is called mean variance

Currently my bot is requesting the public JSON api for /getmarketsummaries/ every 5 seconds. Running thea bot for various hours I never encountered an issue like getting banned.

That is regarding trades, not API calls

i can't speak authoritatively on this so i won't tell you that's too high or too low, but keep in mind it's a lot of data, like hundreds of megabytes after a few hours, and if they don't like it you could get fucked

Interesting! Thanks for sharing different opinions and possible attempts.

Just from my feeling your way sounds more appropriate for a stable-ish (stock!?)market.

My bot tries to look at the smaller picture. Just trying to catch the coin when the volume increases just before the price follows (what happens very often in this order, looking at historic charts)

i'm sure you can imagine how many terabytes of bandwidth they go through every day from people doing the same thing, it's hard to imagine them manually noticing YOU but then it's also not hard to imagine a script being set up to just automatically clobber you

Also known as True Range?
Most of you guys in this thread seem to be reinventing the wheel.

I've been okay so far. If my request times out I just get a new proxy. 10 min ticks so It's very improbable that I get 30 successive fais.

Lol, I do the exact same thing.

"reinventing the wheel" is the essence of fucking crypto

they have, if you search enough you find old tweets about some limits which are nowhere to be found on the api page

they dont ban you, they just block the requests which are too much

i requested trade data for every coin, threaded, every few minutes, most of the time half of the coins returned time outs because they throttled

Finally something useful on Veeky Forums

Don't know if you saw something an user posted in a /bot/ thread about a month ago, but if your sell conditions are met on some position, try only selling with a probability of 0.5. Don't know why it works so well, but it does...

thot bump

what data did you want that wasn't in the summary json?

actual trades

i pretty much got everything they offer through api, trades, ticks, books

gotta have data to build some model

Has anyone tried botting with the ichimoku indicator?

oh, you mean the order book one i guess. when you said coin i was thinking of that request type which is like a lightweight subset of the summary on a per coin basis.

sorry /public/getmarkethistory i suppose

jesus christ six captchas

I've also written a python-Bittrex bot.
Buys dips, sells highs and looks back and adjusts his strategy

>your pic
I'm also a Physics PhD and have worked in the quantum computation field before. In fact I wrote Python applications and guys there too.

>I don't really feel like sharing my ... half finished bot.
You will not lose face and there's no competition. There's huge whole lot of bot usage going on.

Here you have a nice Python wrapper for Bittrex (years old and well working)
github.com/ericsomdahl/python-bittrex

Here's a crypto bot (over 3000 commits on github) that essentially speaks with all sorts of exchanges
github.com/kroitor/ccxt

In fact I've been approached on LinkedIn by people who want me to do math for them in this field. If you want to exchange ideas, I'm all ears.

Also, if you're into crypto and do some coding, you should join the Western NEO devs (Slack etc.). I think it's the most promising community.

Cheers for all your input anons! Although my bot didn't detect anything tonight (possibly caused by the btc surge) at least we got a good discussion together!
I will further look into improvements regarding the volume readings and also continue to run my bot semi-live (posting my results as soon as it notifies me) the next few days

Bedtime now, cya anons!

"Anonymous (ID: zROxZdtJ) 08/29/17(Tue)05:03:40 No.3252220▶

If you're doing short-term/day trading, I find entry prices are less of a big deal as opposed to long term holding/flipping. If there's fairly high volume at the time too, you generally don't even need to make entry point decisions based on longer period charts as well. If something is on a fairly strong uptrend or has great support, and you've been doing your research on the coin or have been monitoring recent historical activity, I find it almost almost impossible to not be able to make 1-2% gains after a few trades by entry on even 1 to 5 minute charts. I frequently manage to do this even with as low as one buy/sell.

Generally having some kind of real-time display from websocket data using API/bots can make it easier to see whats going on in the books. If used properly, trading bots can take a lot of the stress and allow you to capitalize on a lot more opportunities as well"

from another thread

yeah cheers have fun

can i assume from these statements that you'll be sharing the github of your bot too? Pepe

how do you detect said dips/highs?

i mean actually detecting tops and bottoms in a
series of values is fairly easy but is suspect it isnt that easy

Yeah I might give a trading bot a try at some point. This is kinda an early stage bot to experiment with “buy“ indicators.

Thanks for the bot posted I'll have a look tomorrow.

All in all I'm not taking this serious enough to join proper communities about it. Doing it on my own pace as a hobby.

Care to elaborate? Would like to try it as well.

I sometimes wonder if bot trading is a self fulfilling prophecy; that there are so many bots employing a similar strategy that they all buy at the same time, temporarily pumping the price, then exit.

Backtest with various different indicators... I think everyone is too much a jew to give you their exact strategy... but as said, you don't need to reinvent the wheel.

you will notice a pattern if you watch raw charts

the peak of a pump doesn't drop straight down the way it rose straight up, there's a single step on it every time

guess why

The methods are here

github.com/ericsomdahl/python-bittrex/blob/master/bittrex/bittrex.py

for n in range(10**8):
asks.append()
buys.append()
if buycondition(asks, buys):
buymethod
if sellcondition(asks, buys)
sellmethod on the last item in buys for 4% higher price
time.sleep(15 minutes)

The conditions are naive:
pastebin.com/iKTrEhp2

but it works quite well. Pic related is BTC-PAY. I've set it to sell at a plus of 4% but those simple "don't sell while in the midst of the moon" seem to have convinced it at 12%.

What I want to improve my approach to move to a rolling average (exponential weights) and, as you say, maybe think how to do the dips better. Sells aren't such an issue, but for dips maybe once could detect by interpolating. If the bot has an issue, then it's that you buy in too early during a fall. Still, you trade on 10 markets 24/7, so that's an advance

There's no pace to be had. But looking at the Python client for NEO is a good low key way for a natural science pleb like me to get into comp sci of internet networks

The psychology of fomo on a pump, and fear of losing out on dip?

>sharing your full name, bittrex usage
I feel like that's asking to be hacked.

hes not the author of python-bittrex dude

...

how do you handle limit buy/sell?
bittrex seems to have removed market buy/sells from the api due to misuse, so do you just take the highest bit/ask or use more/less with a timeout

You can go and check out hundreds of bot code committing github peoples email addresses.
I don't think it's all that dangerous, and some input is worth it.
Besides, I think stochastic and category theory is more important that muh money :^)

Here's my contacts for anyone who want to discuss how to improve upon buy and sell-conditions: www.axiomsofchoice.org

I buy the ask "A" and sell when the bid goes over "(1+p/100)·A", i.e. p percent gain. If I were to set a limit in advance, I might buy during a moon when I could have sold higher. In the best case, I'd like to teach him to sell when he's hovering on a plateau for 30 minutes. Again, I'm open for ideas.
My naive approach would be to do more tickers and fit a bump function and catch the right spot with respect to that function, but I feel this wouldn't work and so I didn't even start.
Sadly, something strongly typed (i.e. more compiletime bug catches) would be preferable to test stuff where you have to wait for realtime input to test it properly, but I'm mostly a script kiddy.

i've been writing in kotlin, it's relentlessly modern which is not always a good thing but there are some big benefits

i would say it was a positive experience if it weren't for the fact that it keeps crashing the jvm somehow, only started happening after the last update

one fucking embarrassing thing though is that it doesn't have chained comparators so you can't do

A > B > C

i should make a request for it on their issue tracker, it's kind of amazing they didn't bother with it

(and you can't even implement it yourself because you're forbidden from overriding operators, you can only override the methods they call, fucking modern horseshit)

You're in CV? You applied any ML stuff to optimize your bot or train a model to predict price moves? I've done some stuff in CV (cs231n and andrew Ng's course), I want to try train some rnn soon...

Elaborate please? Is your bot available via GitHub or anything?

read the thread, there are quite a few golden nuggets of info in here...