Re: Graph rearrange based on nodes distance

Hello, watchmaker, you wrote: W> There is other idea - clustering. The choice is great: from remarkable hierarchical algorithms, to banal k-averages. In a network the heap of examples like such is, where from ready components (k-means and word2vec) collect . Perfectly, that is necessary. Downloaded their trained 300 measured basis, it is sensitive understood as to consider proximity of words,  here such criterion (it simultaneously stimulates the close words to be a row and far (with negative dot product) to part far away): SUM (dot (normalized_word_vector1, normalized_word_vector2) * (number of words / 2 - word distance)) and , Vovan (I it named Vovan) was recreated AI. He can be asked now and it, thinking 10 seconds, produces curious answers. As a matter of fact, it is terms sorted by coherence to each other - the values are closer, the words will stand more close. I will result a couple of answers here: //uh-oh [' c ++ ', ' unsafe ', ' safe '] [' java ', ' safe ', ' unsafe ']//sausages are bad for health, salads are better [' good ', ' bad ', ' sausage '] [' salad ', ' good ', ' bad ']//sex and money always come together [' algorithm ', ' interface ', ' software ', ' sex ', ' money ', ' violin ', ' jazz ', ' blues', ' whale ']//epic ones. [' capitalization ', ' vaccination ', ' prostitution ', ' emancipation ', ' domination ', ' annihilation ', ' anathema ', ' extrovert '] [' atheism ', ' science ', ' belief ', ' church ', ' lie '] If there will be still questions to Vovan, write, I with pleasure will ask Still it is necessary  a sorting part, for small  full enough search to make, for 800 words it is necessary to write something of iterated, avoiding local minima. On an input there will be a proximity matrix 800x800, on an output it is necessary to collect from them in the list, similar .