snoble

Oh neat, this is scala via scalajs.

This looks extremely cool.

It confuses me why they wouldn't just verify the images since they have the signature in the manifest. Is this because they don't want to wait for a complete image before the start streaming through the pipeline? Is this actually a significant time saver?

I'm the king of using ambiguous terminology (I wish I were clever enough to be ambiguous right now).

fork the code and post the benchmark results

how much emphasis to give to which messages is always the hardest part for me to writing these posts. I'll keep in mind that at times I can be too subtle.

I'd greatly appreciate it if you left this as a comment on the original post.

Sure, but it's not a post about the bestest solution. It's a post about how using more advanced math can ("can" not "will") lead to a worse solution for your context. Fibonacci numbers in a long int is certainly a toy example; but it's a toy example that had an actual reference.

I do distinguish between math and Math. Decide for yourself what that distinction means.

best and scariest quote of the post

  Twitter has a PIN code feature that requires every message to be prepended with a four-digit alphanumeric code. This feature mitigates the issue, but is not available to users inside the United States.

So they fixed the problem... but are withholding the fix from tons of users?

try a `gem update`. Memory performance should be much better now but I'm still curious about speed

the code isn't special. but it's easy for anyone to install and use

sigh, memory leak. It's fixed in github. When camilo is around I'll get him to update the gem

yup, the order is biased. definitely worth a couple of minutes to fix that

but yes. it has to wait until the stream is done before producing any output.

the magic of reservoir sampling is the memory footprint is the size of the output while being completely random. In this particular implementation the order of the output is slightly biased but each row has an equal chance of being in the output.

So the faults of AirBnB would be much clearer if the author would make clear that he understands he made obvious mistakes. This is one of those cases a little bit of mea culpa would have made his point so much clearer. As it is the reader skips over judgement for AirBnB and only thinks about how can this author think what he did to his family was ok.

Oh, I'm a big fan of logistic regressions. And this model actually has some logistic regression in the mix. But I just used this break down of 5 groups to easily report the performance of the model. I can also report a gini but it is way more visceral to see "this is who I predicted to be my top performing group, this is how big that group was, and this is how they actually performed."

In general your work may be complicated/sophisticated. But your results need to be simple.

Um... "You can write RESTful SOAP, I don't think the two ideas are perpendicular." (emphasis mine)

He's saying they aren't perpendicular and also saying they are independent.

I think you may be confused on the meaning of 'perpendicular'

Ah ha! you guys weren't arguing about whitespace in code at all. This thread is just a proxy war for the vim/emacs holy war. I should have guessed.

Thanks wtvanhest

I think we basically agree; you may just be taking issue with my very explicit representation of the formula. All that's really happening here is just a weighted average of daily churn rates. I really think you need to average over a period of time to deal with normal volatility.

While the number '30' is in the formulation is not intended to mean that this metric can only be measured for a month. Rather it's just there to normalize the metric to always be comparable to the monthly rate. It would be very reasonable to take this metric for a month and for every week in the month and see if any of the weeks are substantially higher or lower.

I would be cautious with using this, or anything based on daily churn, to be used in a formula for predictions. If you want to predict what customers will do over x days it is far better to measure what customers have done over x days. What you lose in currency you more than make up for in having taken a direct measurement. Though I would happily use this metric to play in a model with computed weights.

Thanks again, Steven

That's very interesting context and all the more reason to spend time thinking about this stuff (or face the litigious consequences). I hadn't heard of this case before writing this article but I don't find the similar definitions all that surprising. We tried to stay pretty close to the sorts of definitions one might learn in an intro financial analysis course.

I first got R when I was reading Statistics: an introduction using R (http://www.amazon.com/Statistics-Introduction-Michael-J-Craw...).

That said, I ultimately never fell in love with R so perhaps my experience is a bad example (though I would argue there are lots of legitimate reasons not to fall in love with R)