I'm trying to optimize a combination of 8 cards with 64 card characters. No repeats and order doesn't matter.

n!/(n!(n-r)!) = 4,426,165,368 combinations

I have everything set up, including the data scrapper. But I don't know how many games my machine needs to learn from to start seeing patterns. For example, in 14,000 games analyzed, only 834 decks had a certain character.

Analyzing the next 7 cards from 834 decks is 621,216,192

So I guess I need more data before reliable patterns emerge... but how much data? Thank you and god bless

  • 2
    $\begingroup$ The general rule: as much as you can get. $\endgroup$
    – Manngo
    Aug 9, 2017 at 16:50

1 Answer 1


Sorry but no one can give you a number.

As stated already in the comment by Manngo, the general rule: is as much as you can. I have also seen the 10x number being thrown around, see here. According to this "rule", you need roughly 10 times as many examples as there are degrees of freedom in your model.

How much data you will need is something you can only answers to through empirical investigation. The amount of data you need to collect is influenced by the complexity of the data, the dimensionality of the data and the algorithm you intend to train(although this can be controlled somewhat through regulirazation, see this good blog post by Jake Vanderplas for more information).

This question might be of interest to you.


You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .