656 karma · joined December 5, 2015
Semantic accuracy: 63.84 % Syntactic accuracy: 67.00 %
Here performance is not great (great would be >80% on semantic and >70% on syntactic).
As this task requires nearest neighbor lookups, performance is impacted by vocabulary size. Since models trained using Wikipedia alone usually limit vocabulary to something ~300k words, we can try that to get scores which are comparable to those posted by the GloVe [1] and LexVec [2] papers by only using the first 300k words in the pre-trained vectors, giving the following results:
Semantic accuracy: 77.75 % Syntactic accuracy: 72.55 %
Impressive stuff!
[1] http://nlp.stanford.edu/pubs/glove.pdf - https://github.com/stanfordnlp/GloVe
[2] https://arxiv.org/pdf/1606.01283v1 - https://github.com/alexandres/lexvec
WS-353 Similarity: 0.781
WS-353 Relatedness: 0.682
MEN: 0.765
MTurk: 0.679
RW: 0.487
SimLex: 0.380
MC: 0.812
RG: 0.800
SCWS: 0.667
Impressive for a model trained on Wikipedia alone!
I will post analogy scores for this model as soon as they are done computing.
From a non-practical point of view, you're missing out on the pure joy of writing Ruby. Everyone should try it at least a couple of times to see just how expressive and fun programming can be. As someone who programs regularly in Python, Go, JavaScript, and Java, Ruby blows them away in FUN points.
As a side note, it would be awesome if someone wrote a Ruby -> Python compiler. I would love to be able to call NumPy from Ruby code. I would even be happy with a (generous) subset of the language. I think gluing together Python libs with Ruby would be a pleasure, and people could start using DSL magic to implement better APIs for existing libraries.
If an uncaught exeception does happen, it moves from unlikely to likely, and so gets plugged up in the code. Rinse and repeat. Again, the obvious advantage being errors that are easy to track.
In go this approach would result in a possibly nightmarish debugging session.
Though initially the diligent programmer will check each and every call with if err != nil {}, there comes a time when laziness kicks in. When that time comes, you'll get hit with unexpected behavior (aka: tracking down nil references).
You might argue the same would happen with exceptions. Yes, they will blow up in your face as well if you fail to catch them. But when they do, you'll hopefully have a well specified restriction that enables to you pinpoint the issue almost immediately.
Basically, with exceptions you know the guy who shot you in the gut because you were being lazy, whereas with Go's error system he's hiding behind the bushes with a dozen or so of his conspirators.
Seems like a PR stunt to attract talent.
My interest is in machine learning, so I would like to see decent Linux drivers and more work on the OpenCL environment.
1) host this myself. Something like apt-get install hyperdev.
2) have an integrated SQL database in the runtime. Concept of migrations should be known by the runtime.
3) have a one click deploy option that synced everything to a remote runtime, be it on my own servers or hosted by a 3rd party. Syncing to a runtime with older schema should automatically run migrations.
UNTIL I discovered that if you disable "Use a prediction service to help complete searches and URLs" option for the Omnibar it works PERFECTLY, just like Firefox. Type in any part of the URL or title of the page and bingo.