А кто-то работал с нейронными сетями? Коллега спрашивает: "Anyone familiar with neural nets? I'm wondering how something like https://twitter.com/deepdrumpf would work. Since those neural nets usually need a vector as an input, I'm having a hard time understanding how you would even feed text into it. First there's words, which I guess you could map to a vector that has a 1 somewhere at the index of the word in a dictionary. But then per word you'd already have the entire size of the english dictionary as an input. And it wouldn't even encode the sequence of words that way but just one. Could of course also map the entire dictionary to the range [0,1], then you could represent multiple words. But somehow I also don't see how that makes any sense."