Monday, June 27, 2011

Two Roads Diverge - Machine Learning or IOS Dev?

Now that my latest project (for those interested in such things - 30 k lines of Haskell, 200 k lines of (mostly legacy) C/C++ code, a few thousand lines of Lua, signal processing, some NLP ish things) is "done done"[1] I have to choose [2] a new project.

Choosing a new project is always an exciting time, but also a mildly stressful one. For every choice made, a half dozen equally worthy alternatives have to be rejected. And I do a lot of agonizing over what is the 'right' project to take up.


One way to maintain a sense of continuity among projects is to examine what could have been done better on the finished project and see if you can build a project around fixing those deficiencies.

Working on the last project exposed some flaws in my dev chops - I know nothing about Network Programming and this caused me to take longer than usual to fix a few nasty bugs that cropped up. So I'd like to take a couple of months off and work through Stevens's books and close this gap and then build some customized network monitoring tools which I could have used when my hair was on fire. Our visualization and rendering subsystem used an Open Source renderer that fell down on large datasets. NLP algorithms in Haskell had to be painfully built one by one. The Computer Vision library we used (Open CV) is a friggin mess that needs serious surgery. And so on. Doing all that would multiply the existing codebase's power by a factor of 10. And also make good building blocks for new projects.

And many ML projects have the strange property that completing them successfully opens up even more ambitious projects. The folks who sponsored the last project want me to do more stuff for them.


Another way to choose a new project to work on is to find great people you'd like to work with and build a project around what they are doing or are interested in. My stubborn refusal to move to the USA somewhat limits my choice in this regard - Not many people or companies in Bangalore are doing anything interesting in Machine Learning. But otoh a few people have bounced really (really really) interesting IOS projects (and start up plans) to me. On the one hand, this means I have to go over to the Dark Side and sell my soul to the evil but competent folks at Apple and learn Objective C and overpay for a MacBook and the annually renewed right to put software I write on hardware I already paid for and so on. Being a storm trooper for Darth Steve is a proposition that requires some thought. But on the other hand, I would be working with ultra competent devs again (Working alone, or as the only dev on a team is the only negative - and it is a small one - in my 'lifestyle'. Fixing that would rock).

A third choice - I actually thought of sitting down and writing a book, just for a change of pace. I have a few ideas for some tech books I think are missing from thes shelves and every dev I pitched reacted with a variant of "I'd buy that RIGHT now - please please write it". What stops me is that people who have written successful tech books say that it is a pretty thankless task, and with some exceptions, financially unrewarding (though your "prestige" goes up- something I don't care a rat's ass about). If I had to choose between spending a thousand hours writing a book and a thousand hours writing code, it is somewhat hard to choose the former.

Hence the "two roads diverge" tone that permeates my thoughts. I could dive deeper into Machine Learning(and allied areas) or go do mobile app stuff. Choosing promises to be interesting.

Two Roads Diverge and all that jazz[3]

But first, before I have to make a choice, clear the backlog of people to meet (I thank you all for your patience and suffering my erratic schedules), places to visit, things to do. (Metaphorically) lie on a beach somewhere with no computers in sight. Relax, refresh. Then decide.


[1] Most projects have an official "done " date and then a later "done done" date. In this case the project was 'done' some time ago and then a rookie dev wiped out the source control repo while simultaneously trying to alter the Haskell code (vs writing a minor script in Lua, which is what the situation called for), bringing the whole cluster down, causing the (non dev) owners of the project to send an SOS to me to get on a plane pronto and put out the fire.
Some fences have been built to avoid this kind of FUBAR situation from happening again so now I am "done done"

[2] One significant milestone in one's evolution as a developer is when you realize that you have more ideas than you can implement in your lifetime. You are even luckier when people pay you to implement them (vs being assigned to some Godawful Leasing System dev in some enterprise dev body shop say)

[3] - From Frost's poem, of course

Two roads diverged in a yellow wood,
And sorry I could not travel both
And be one traveler, long I stood
And looked down one as far as I could
To where it bent in the undergrowth.

...................................

I shall be telling this with a sigh
Somewhere ages and ages hence:
Two roads diverged in a wood, and I
I took the one less traveled by,

And that has made all the difference.

9 comments:

  1. Hi Ravi
    Best of luck on the road you are going to choose.I am sure you will feel happy about it in future.

    What are the book ideas you have in mind? Please share them.Experienced developers writing books will be of immense help to others who genuinely want to improve their skills.

    Vignesh

    ReplyDelete
  2. Since you've already done many projects in ML/NLP, perhaps it's time to try out iOS dev as something different? :)

    ReplyDelete
  3. Hi Ravi,

    Please have a look at MonoTouch - http://monotouch.net/ using which you can write C# code for iPhone. This obviates the need for learning Obj. C and you get memory management for free.

    Good luck,
    Manoj.

    ReplyDelete
  4. learning Objective C is trivial. Why should I learn C# (yuck!) instead? When developing for a platform, best to develop in the language it is written in.

    And, if I wanted another language, I'd write a lisp/ml to Objective C compiler. And I don't. Learning Obj C is 2 days work at most.

    ReplyDelete
  5. P.S. Check out this 2-day-old Economic Times article on exactly this topic - http://swaroopch.net/jXVRjO

    ReplyDelete
  6. @swaroop yes the "change of scenery" argument is a compelling one. OTOH the machine learning projects are very "magic" ey and "deep tech" satisfying and have real impacts on the real world.

    Thanks for the ET links!

    ReplyDelete
  7. I hope you stick to ML and the road you take :)

    ReplyDelete
  8. I'm really curious about what the completed project was about... and the book ideas too!

    ReplyDelete
  9. @KR can't talk about the project in public. Likewise for the book ideas, because I may just convert one of them into an actual book some day. But over a beer is a different story ;). One of these days ...

    ReplyDelete