René's URL Explorer Experiment

{"@context":"https://schema.org","@type":"DiscussionForumPosting","headline":"Bug in AdaGrad implementation","articleBody":"AdaGrad does not increase the size of the weight vector while learning. Weight Vector dimensions might increase if there are new features seen while feature extraction from unseen training examples. \r\n\r\nCause:\r\nhttps://github.com/IllinoisCogComp/lbjava/blob/master/lbjava/src/main/java/edu/illinois/cs/cogcomp/lbjava/learn/AdaGrad.java#L189\r\n\r\nExample feature:\r\n\r\n```\r\ndiscrete MyTestFeature(MyData d) \u003c- {\r\n    return d.isCapitalized() ? \"YES\" : \"NO\"\r\n}\r\n```\r\nFor this example, weight vector should have size 3 - YES, NO, Bias Term. But `exampleFeatures.length` is only 1 here.\r\n\r\nCompare with implementation of StochasticGradientDescent.\r\n","author":{"url":"https://github.com/bhargav","@type":"Person","name":"bhargav"},"datePublished":"2017-02-05T19:07:05.000Z","interactionStatistic":{"@type":"InteractionCounter","interactionType":"https://schema.org/CommentAction","userInteractionCount":2},"url":"https://github.com/99/lbjava/issues/99"}