Show HN: cloudlab - numerical processing with javascript & json | Hacker News Reader

Show HN: cloudlab - numerical processing with javascript & json

cloudlab.io

12 pointshuragok4 comments

First of all, I think this is a great idea. It's something I've thought a lot about myself for several years now -- with the name 'cloudlab', even! -- but ultimately decided to work on a different set of problems in science. Anyway, the key questions you'll need to answer are, in approximate order of importance:

* What do I do with my mountains of MATLAB code? Will cloudlab run my existing mcode and access my existing mat files as-is? Will you provide a conversion tool? How perfect will it be? If I have 5 years of work in mcode, and huge amounts of organizational knowledge around that process, why would I ever redo that?

* How do you handle security? My data is my entire career. I can't take a risk with that. Running MATLAB at all, much less locally on underpowered hardware blows, but you still need to convince me that my data is more safe with you than it is with me. Which is hard, considering my data ranges from VPN'd to airgapped.

* How much does it cost? MATLAB may not be free, but it's paid for via a convoluted pre-approved purchasing system once every ten years by someone who isn't me. I can't decide to go spend money on a MATLAB-that-isn't-MATLAB on a whim, much less a lot of it. Being free now does nothing to inspire confidence in either longevity or affordability into the future. See: Google App Engine.

* What kinds of data formats do you support, and how big can the data get? I have something like 5 terabytes of data lying around waiting to be analyzed. JSON doesn't really sound compelling when I'm talking about huge matrices or gigs of timeseries data. Do you support HDF5? How does plotting work when I want to plot ten million points, or draw a fine 3D mesh? It would almost certainly need to be rendered server-side and presented as an image... it looks like you're using basic JS charting libraries, though, from the screenshots. How well does that scale to real datasets?

If your scripting environment makes lots of assumptions about your platform in a way that makes it nonportable code, and you don't open-source it, it'll be hard to get people to make the investment and move projects onto your platform now knowing it could disappear at any time. If I spend 6 months collecting data, then another 3 months analyzing it, and then you announce you're shutting down because you failed to raise VC while I'm in the process of preparing a manuscript, that would set me back months and generally be a disaster.

You don't expose too many docs without requiring a registration, and the screenshots don't really give much away. What's up with that?

Edit: Also, the fixed password policy isn't really amusing. I guess you guys will track it and see what happens, but I bet that you'll notice a way depressed return user rate, because every time I'll want to log back in I'll need to go search through my email first, and I'll never really "activate" as a user after the first time or two playing around. If you're serious about account security, use 2-factor (https://moocode.com/posts/3-using-the-google-authenticator-a...) auth or something similar.

I've just pushed a little fix that allows anonymous users to view documentation at http://cloudlab.io/documentation

At the moment, the software is a proof of concept. There is much, much more planned vis-a-vis data formats, APIs and built-in libraries.

As for your Matlab code, I've given thought to a transcompiler. It's not high on the list, but it has been given thought.

If you're serious about converting any significant percentage of current MATLAB users (which would be the vast majority of the scientific community), either a JS mcode interpreter or a code translator should be near the top of your list. With that you'll turn heads; without it, everyone will ask why they'd ever throw out their lab's huge library of accumulated, well-understood, thoroughly-debugged (hard!) code.

Also, just out of curiosity, what are your guys' backgrounds? Why did you decide to build this instead of something else? Is this something you'll use (or, would have before you stopped whatever you were doing to start this), or did you just identify a market need?

its large but its not the entire scientific community. I used to be a matlab user, my entire phd thesis work was all in matlab, we were in medical imaging. Then I left to work in industry in finance and we use python numpy scipy for all our work and its much better than matlab. R, Fortran, C/C++/Java are also highly represented in scientific work.

Reply on news.ycombinator.com