Neural Nets Trading System - Artificial Intelligence Strategy Robot Test

This strategy robot I wrote in C# is using genetically evolved neural nets. It learns trading rules by itself. It's trained on 40% of data sample and validated the rest 60% - shown by the red line division. The neural nets was able to generalize and exploit certain property of the Bitcoin market. Strategy holds well on the unseen data. Avoids sharp drops in price and takes advantage of rallies.

If you're a pro trader or trading company in NYC and want to colab contact me at RobTheQuant at gmail

[vask]

This is awesome. How many inputs are you using? Does each gray block represent a single input? And for how many generations you had to train the NN? Sick.

[alonzoc]

I worked on something like this with arbitrage a long time ago. how are your inputs encoded that is usually the key. my current project is processing news feeds to determine sentiment.

[Stisse]

You running it on the uptrend? so what happen when you get spikes and trendchanges?? Big losses?

[mattenricochanneling]

Hi. I watched this video over and over again and again.. First of all nice that you develpoed this, but what i see the more date the soft has learrned the less deals it starting to make. WHat happens affter you give im 1000 days of trading?

[EemanAlipour]

what is that and how does it work? would you explain ir teach it? is there any tutorial video?

[Corpsecreate]

Does this accommodate transaction costs/slippage?

[RobotronSage]

Looks like it's great at avoiding dumps but doesn't seem to be good at holding the pumps
If i'm reading this correctly the bot barely had an advantage vs holding strategy (apart from during the dumps)

[boogaloobomber]

Useless bc commitions are too high. But interesting concept.