Live data from Hacker News

Generating Magic cards using deep, recursive neural networks

mtgsalvation.com

11–20 of 69 posts

Re: Generating Magic cards using deep, recursive neural networks

#11
post #7
post #6

If I'm interested in Neural Networks, and I've mainly been working in web development (I know Java, C, and Python). Where would be a good place for me to start so that I can eventually make something like this project?

Andrej Karpathy's blog is excellent as a first introduction. Start with http://karpathy.github.io/neuralnets/ He also has a fantastic article specifically about recurrent neural networks here: http://karpathy.github.io/2015/05/21/rnn-effectiveness/

Thanks! I'm really enjoying seeing the huge variety of domains people have applied char-rnn to. I haven't anticipated response at this scale, to both the original blog post and the code release. (I try to keep a somewhat up to date list of responses on the bottom of the blog - cooking recipes, music, obama speeches, magic cards...)

In terms of learning, I would also encourage people to try the materials from our CS231n class - this is a class I taught at Stanford last quarter with my adviser. It's technically about Convolutional Networks, but most of the class material is building up generic Neural Networks, backprop, and so on. You can also try our IPython Notebook assignments.

Course notes: http://cs231n.github.io/ Syllabus with slides too: http://cs231n.stanford.edu/syllabus.html

Another good pointer is Andrew Ng's Coursera class - that's a thorough introduction as well.

Re: Generating Magic cards using deep, recursive neural networks

#12
post #6

If I'm interested in Neural Networks, and I've mainly been working in web development (I know Java, C, and Python). Where would be a good place for me to start so that I can eventually make something like this project?

I've found http://neuralnetworksanddeeplearning.com by Michael Neilson to be an excellent resource.

Re: Generating Magic cards using deep, recursive neural networks

#13
post #10

Earlier quoted context omitted.

Matlab has a neural network toolbox that you can use to train small recurrent networks. There is a book that I can link if you are interested.

I am interested :)

This is from one of the people who helped write the toolbox in Matlab. There is a page limit on the Amazon publishing service, so that is why it's posted here[1]. Chapter 14 describes notation for the recurrent networks and explains their training.

[1]:http://hagan.okstate.edu/NNDesign.pdf

Re: Generating Magic cards using deep, recursive neural networks

#14
post #8

How about IDE that suggest you next line of code? Or next variable, or method to call? Add static analysis and it can actually be fairly useful. E.g. val name = object.[maybe you wwant to use getFirstName]?

http://pliny.rice.edu/

http://www.darpa.mil/program/probabilistic-programming-for-a...

Re: Generating Magic cards using deep, recursive neural networks

#15
post #6

If I'm interested in Neural Networks, and I've mainly been working in web development (I know Java, C, and Python). Where would be a good place for me to start so that I can eventually make something like this project?

Keras http://keras.io/ is a neural networks library for Python that is rather easy to get started with compared to the alternatives.

I also recommend the tutorials on http://deeplearning.net/tutorial/ (Python / Theano)

And Andrej Karpathy's blog is also a great resource for explanations of deep learning concepts in simple terms: https://karpathy.github.io/

Re: Generating Magic cards using deep, recursive neural networks

#16
post #8

How about IDE that suggest you next line of code? Or next variable, or method to call? Add static analysis and it can actually be fairly useful. E.g. val name = object.[maybe you wwant to use getFirstName]?

Believe me, it's been thought of (it's been in my head ever since my colleague Frank Seide introduced me to RNNs 4 years ago, I'm sure I wasn't the first!). It is one thing to generate code that looks like code, quite another to generate code that is useful. Character-by-character probably isn't good enough for this task.

Anyways, I talk about how to use probability (but not compute it!) in http://research.microsoft.com/pubs/179363/mcdirmid12.pdf

Re: Generating Magic cards using deep, recursive neural networks

#17
Magic is an interesting game just from an text analytics perspective. Using a source like mtgjson you get a reasonably sized corpus of text that looks natural enough for a human to comfortably grok but sufficiently templated to apply even naive methods and acquire awesome results.

I built a tool that uses n-grams to find cards similar to each other. The result is comparable to manual suggestions done by users but takes around a minute rather than months to cleanly update after a new set is released!

Re: Generating Magic cards using deep, recursive neural networks

#18
post #12
post #6

If I'm interested in Neural Networks, and I've mainly been working in web development (I know Java, C, and Python). Where would be a good place for me to start so that I can eventually make something like this project?

I've found http://neuralnetworksanddeeplearning.com by Michael Neilson to be an excellent resource.

This is the one I've always recommended as well. It's a little longer, but very in-depth.

Re: Generating Magic cards using deep, recursive neural networks

#19
> Shring the Artist

> 2BB

> Legendary Creature - Cat

> Flying

> Whenever you cast a spell, you may return target creature card from your graveyard to your hand.

> 2/2

That's actually a fairly plausible card, in terms of mechanics. It's thematically appropriate for black, it's about the right cost, it's powerful but specialized (returns creature cards only, and only to hand rather than to play), and it's legendary to avoid multiplying its effect. This would work well in a deck with many creatures with "enters the battlefield" effects (common in black, as well as blue and red which are often used together with black), as well as many ways to sacrifice cards. Would even work as the general in an EDH/Commander deck.

The only thing slightly odd is the flavor: it's a flying cat. And with the right art (and perhaps a "black cat crossing your path" joke in the flavor text), that could work quite well.

Post reply on HN