Training Trading AI

[Wictorian]

DISCLAIMER : I know almost nothing about ML.

 

Where can I get data to train the AI?

 

Also should I train a different AI for each market ( with the markets data) or train one with data from multiple markets?

[Prodigy_Smit]

If you plan on using this with your own money then don't. You are guaranteed to loose money, even the best of the best human stock traders can only get around 50% of their trade predictions accurate an AI can definitely not. YET.

[Wictorian]

Just now, Prodigy_Smit said:

If you plan on using this with your own money then don't. You are guaranteed to loose money, even the best of the best human stock traders can only get around 50% of their trade predictions accurate an AI can definitely not. YET.

I would guess an AI would do a lot better than humans.

[WereCatf]

9 minutes ago, Wictorian said:

I would guess an AI would do a lot better than humans.

Human-traders can follow the news, they understand motivations, they can see correlations between events, they can, to an extent, predict government-imposed regulations -- just try training basic NLP to an AI and you'll quickly see that that is a humongous task.

[Wictorian]

3 minutes ago, WereCatf said:

Human-traders can follow the news, they understand motivations, they can see correlations between events, they can, to an extent, predict government-imposed regulations -- just try training basic NLP to an AI and you'll quickly see that that is a humongous task.

Yeah, thats the biggest disadvantage, but you can follow them yourseşf and interfere in case of emergency. Besides, manhy traders just analyze the charts. They are two different categories.

[Sauron]

Just now, Wictorian said:

I would guess an AI would do a lot better than humans.

An AI is only as good as the humans training it, and as you said yourself...

Just now, Wictorian said:

I know almost nothing about ML.

 

[Wictorian]

2 hours ago, Sauron said:

An AI is only as good as the humans training it, and as you said yourself...

 

LOL. 
How many different ways are there to train an AI? I only know anout deep learning and I suppose you give the information (price and time) and it will get the job done?

[WereCatf]

31 minutes ago, Wictorian said:

How many different ways are there to train an AI?

Practically an unlimited number of ways. See e.g. https://en.wikipedia.org/wiki/Machine_learning#Approaches -- you are completely out of your depth here. An AI isn't a magic box that you can just put numbers in and get a desired result out; you need to choose a training method that suits your needs, then choose a model that suits your needs, then figure out all the data that you need while filtering out any data that might introduce unwanted bias and so on.

[Sauron]

1 hour ago, Wictorian said:

I suppose you give the information (price and time) and it will get the job done?

It's far from an exact science right now, typically successful models are developed by researchers over years of work where they try different techniques to improve their results.

[shadow_ray]

Don't get discouraged, create a framework which can test your algorithm.  Give it actual market data and a small amount of "virtual" money and let it trade on real wold prices (you might want to include transaction and other fees as well).

This could give you a feel for how well would it perform in the real world.

 

It does not involve risks and It doesn't require any investment other then your time.

[Wictorian]

23 hours ago, shadow_ray said:

Don't get discouraged, create a framework which can test your algorithm.  Give it actual market data and a small amount of "virtual" money and let it trade on real wold prices (you might want to include transaction and other fees as well).

This could give you a feel for how well would it perform in the real world.

 

It does not involve risks and It doesn't require any investment other then your time.

Actually I was asking about how I could get the data to feed it.