/biz/ - Business & Finance

This thread is inspiring me to go back and hammer away at my genetic algo I was working on. Written for bittrex from a long time but I'd like to rejigger it to work on Binance.

Biggest issue with it is it would evolve signals that frequently would trigger on all alts at once (ie when BTC is shitting the bed and everything is dipping). I want to somehow parse my altcoin backtest logs based on the top 100 altcoin average, ie, only buy shitcoin X if it's dipping relative to how altcoins in general are doing. Or if it's [whatevering] relative to how altcoins in general are doing, based on the signals the algo evolves.

It still made 2x vs holding BTC in October/November which was pretty decent for a shitty bot written by a mouthbreathing retard in PHP.

pic attached is the output format for the signals it would generate, I can hardly even read what it's saying it's been so long since I tinkered with this shit.

>>4694258
I don't use gekko. I don't use anything out of the box. I give the bot the loosest constraints possible then let it evolve buy signals over time. The buy signals fight, fuck, make buy signal babies and then I repeat the process. Then I run them through a few different batches of other backtest data to make sure they're not just overfits.

Attached is an example of one the buy signals the bot has evolved and actively uses.

>>4307298
>Real "scripts" that work rely on up to 200 independent indicators

That's not true in the least. I've been netting around 1% daily gains with my bot the last month; it uses multiple buy indicator files to identify good picks and each one usually only has around 3-4 data checks in it. So if a coin passes all 3 or 4 checks it'll buy in. Here's one of my better-performing ones for instance. Stuff at the bottom are the coins it chose during backtesting. There are a few losses at the bottom of the list but it's like less than 1/10 of the cases.