mikeskim

8 ms·

1.
Show HN: Endless Library – Why I should keep my day job(play.google.com)
2 points by mikeskim 10y ago|0 comments
2.
Ask HN: Career advice for data science / ML / AI?
3 points by mikeskim 10y ago|2 comments
3.
by mikeskim 10y ago
I went from barely using data.table to only using data.table for basically everything in less than a few years. I think this is the trend given it's faster than basically everything: https://github.com/Rdatatable/d
4.
by mikeskim 10y ago
The way I use Python in machine learning is quite different from how many others in competitive ML use Python. I use Python purely for Python 2.7 with Pypy and try not to touch or use numpy,scipy,pandas,etc. R's data.table is possibly
5.
by mikeskim 10y ago
It is a fact that students at top tech schools (think Caltech, MIT, Harvey Mudd, etc.) study longer compared to students at comparable liberal arts schools. I believe it takes time and practice to get good at anything. Students at tech scho
6.
by mikeskim 10y ago
I wish academics would publish pure python implementations of their "new" algorithms. Standard python with Pypy is enough for speed of development and runtime. The biggest thing about t-SNE is that it's been used in competiti
7.
by mikeskim 10y ago
I would not be surprised if most academic CS research is not reproducible. This is true for many other fields outside of CS, and I've seen it first hand in machine learning. It's a problem but it's also just how things are.
8.
by mikeskim 10y ago
I know people who work at those research divisions (they work with neural networks) without a PhD. So I just falsified your claim.
9.
by mikeskim 10y ago
This is incorrect in almost every way. When you have 2^m independent observations that you can use to cross validate (where m is very large), overfitting is exceptionally difficult almost regardless of the number of features you have. Overf
10.
by mikeskim 10y ago
This is supposed to be faster than XGBoost? I'm skeptical, but I'd like to know the specifics of the benchmarks and maybe an outline of the code / reasons why. It was not benchmarked by the same person who did https:/&#
11.
by mikeskim 11y ago
If people would just learn to resample (cross validate, use subsampling or the bootstrap), we wouldn't be having this pointless discussion at all.
12.
by mikeskim 11y ago
I'd like feedback on if this algorithm is new or has been published before. I think it's new, but I've been wrong before on these types of issues. Thanks: Mike.
13.
Show HN: DBoost stochastic gradient boosting via Ridge with random dummies(github.com)
8 points by mikeskim 11y ago|1 comments
14.
by mikeskim 11y ago
compete at www.kaggle.com defeat is the best teacher.
15.
by mikeskim 11y ago
You wouldn't find it intuitive if your first language was Scheme.
16.
by mikeskim 11y ago
This could change your life: http://adv-r.had.co.nz/
17.
by mikeskim 11y ago
Pypy is actually much faster than both standard Python 2.7 and R in basically everything requiring the standard library.
18.
by mikeskim 11y ago
This might even be useful when applied to learning say data.table or other parts of the Hadleyverse. These days data.table is not an option given the size of data and how slow R data frames are.
19.
by mikeskim 11y ago
You could just use data, machine learning and optimal control to make better estimates. It has aleady worked for fighter jets, self driving cars, and automated factories. With enough data it will work for anything.
20.
by mikeskim 11y ago
Tommy Kono would disagree with you. Albeit he was a weightlifter.
21.
by mikeskim 11y ago
it is probably easier to just wrap R functions as a subprocess in python. you would have to write some R code to do this which might involve writing to a flat file. i do this all the time in R wrapping up something like rgf and calling from
22.
by mikeskim 11y ago
As long as you make the public leaderboard set small, and the private one shot leaderboard set very large, the number of submissions matters very little in the final rankings. The only real issue is hand labeling the public leaderboard set
23.
by mikeskim 11y ago
so you mean there are 5 entries you can select to be scored in the end, and there is one private leaderboard no one can see but the admins? do you know the size of the private leaderboard test set?
24.
by mikeskim 11y ago
Is there a private one shot hold out in this particular competition?
25.
by mikeskim 11y ago
You can do all of those things in R except maybe Simulink with a GUI (at least not yet, maybe soon?), and you have the choice of working with multiple implementations. https://cran.r-project.org/web/packages/FKF&#x
26.
by mikeskim 11y ago
i wrote a paper about leveraging public random streams like this one. it can be downloaded here https://drive.google.com/file/d/0B9IkyvYlZZe7TldTRGlSMnpQX0U...
27.
by mikeskim 11y ago
A highly biased sample of less than 300 is supposed to really say anything of any significance for the general population? I think not.
28.
by mikeskim 11y ago
can you fully automate data driven investment on these platforms?
29.
by mikeskim 11y ago
has anyone tried asking technical questions back? e.g. a list of putnam questions.
30.
by mikeskim 11y ago
this game feels nothing like street fighter alpha to anyone who has played sf competitively. it might look like it though.

More ›