/biz/ - Business & Finance

Nobody reads. Try sports. For cheerleaders.

>>22919924
What?

The fact that no one is replying honestly makes me happy. It means you retards haven't tried this yet.

i'd be more concerned with the logic than the language, anon. what makes you think your logic is superior to existing binance bots?

>>22920337
Lol we just can't wait to see you make it or die in fire. Keep us updated, anon.

>>22920430
I don't. I just want to experiment a bit since it's not exactly something your average shitcoin trader is doing. I'll just start with something simple.

>>22920530
then i'd suggest looking at exiting projects on github before anything else.

>>22920738
I will. Gonna play with binance's API for a bit first though.

>>22919622
I've attempted to create models for a trading bot, never got consistent enough results to put it into production though.
What method are you going to use to decide on trades for your bot?

>>22921599
this, to tell a bot what to do, you need to know how to trade in the first place, and trading is gambling, why even bother automate the random process of gambling?

This is a great idea. I have a CS background and have started playing with some CEX apis and using LSTMs to analyze coins. It would also be interesting to make a bot that looks at all prices across CEXs and DEXs and arbitrages.

Tg group?

your bot will malfunction and you'll end up with $0

i have a way better bot idea

If your biggest concern is what programming language you're going to use to write your trading bot, you are literally ngmi.

t. profitable bot

>>22921921

May I plzZzzzzzz intern for you for free

>>22921921
I'm not asking what language to use faggot. I'm asking for general tips.

I've tried LSTM approach and it just learned to be a good ema-like indicator.

>>22921766
>LSTMs to analyze coins
I've used models like this before, they don't generalize well to future data based purely on price action.
It's quite easy to accidentally create a model that merely remembers prior actions, and does not do any sort of deduction on existing data.
Make sure you're giving it independent rounds of training data, and that you randomize it and make sure they don't overlap. You will overfit it otherwise.
I haven't worked on any models recently, but the idea I have for my next bot is to use a a sort of hybrid Evolutionary neural network(probably with recurrent layers and possibly with convolutional layers) along with a markov decision process. Bots will decide on actions via the network model, and act on a predefined decision space according to a set of parameters generated through the evolutionary process. Might also do some evolutionary data selection
I don't have high hopes in general for market predicting bots though.

Tbh a better question is what’re your data sources, and how do they give you an edge

Read advances in machine learning by Marcos Prado

Python if you’re trying to make a trading algo

>>22921970

Thx for the tips wise anon. I’m trying to integrate sentiment analysis combined with price action for better results.

Could you explain this idea more? Are you using a genetic algorithm to select hyper parameters?

When it comes to a project like this the scrapy python library might come in very handy.
I personally like python because of how flexible the language is, and this is a really good idea thats come into my head once before, I'm just trying to quit being a lazy stoner.

>>22922023
>Are you using a genetic algorithm to select hyper parameters?
well I haven't done it yet lol, but yeah that's the idea. Had a project in the last year of undergrad where I wrote a program for an evolutionary neural network, I've loved the idea ever since, seems like a lot of untapped potential.
In addition to that, things like deciding on a threshold for selling/buying/holding/bet sizing will be decided by an MDP, the actions of which will vary based on a genetic algorithm.
The intuition for the idea is that each instance of the network/decision process in the GA corresponds to a sort of 'personality' of a trader, with some traders being more risk tolerant and having stronger hands. The GA will create a bunch of instances of 'traders', and have their individual networks decide on the next decision in the MDP, the progression of MDP states and their hyperparameters will be learned through the GA. Reinforcement reward for the MDP is simply the amount of money that they gain.
A bunch of simulations will then be done on previous data, and some subset of the best performers will enter the next round, just like a typical GA algo.
I'm still in the stage of conceptualizing exactly how it will work, but that's roughly what I'm thinking

>Are you using a genetic algorithm to select hyper parameters?
ngmi

Most people don't appreciate just how hard it is to write a profitable bot.

>>22921958
What are you trying to build exactly? What is the objective

>>22922352
Just something that swing trades. I'm just experimenting as I said before. I'm not even a good trader.

>>22922352
Honestly as a follow-up there is more money to be made learning solidity. Bug hunting is extremely profitable even as a white hat hacker, but if you're a cunt you can try to find errors in all these defi projects that keep popping up and steal the money locked in the smart contracts. Way more profitable to focus on Ethereum than some Binance bot which is probably going to be competing with hundreds of others already in production.

>>22919622
Look up Hummingbird bot

>>22919622
bots on binance will soon be automated