214 karma · joined March 25, 2012
https://sg.linkedin.com/in/mailshanx
Do you postpone comp-talk until an offer is on the table? Or is it safe to assume that large companies generally have an extremely elastic range?
Do you think it's a good idea say something like "If you can offer $OBVIOUSLY_EXTREMELY_HIGH, i will immediately accept. For lesser offers, i will evaluate them on a case by case basis."
How do you respond when they insist on knowing your comp details?
Also, do you think any of these places might be open to hire a hybrid programmer-quant i.e. someone who isn't as good at math / programming as a typical quant/dev, but is good at connecting the two and working in the middle?
You can use https://www.csauthors.net/distance/mandar-chitre/paul-erdos to find yours.
If you land a position at a Big 4/5, you can expect to make 300k+ (total comp), and still retain flexible work hours, PTO, etc. Competition for dev talent is fierce, and with good interview / negotiation skills, it is possible to make integer multiples of your current comp without sacrificing perks such as a good team and flexible hours.
Also, read this: http://haseebq.com/how-not-to-bomb-your-offer-negotiation/
I have had a reverse path - I spent my early career working for a research lab, where the the work was immensely intellectually satisfying and fun for the most part. However, the pay was so ridiculously low that my industry salary was a few integer multiples of my researcher pay.
I feel like becoming a great engineer is almost orthogonal to getting paid the big bucks. It feels like job hopping, salary negotiation, and most importantly, the opportunity to do great work, all require some deliberate and strategic planning - and that these are the things that actually impact salary levels much, much more than one's raw engineering skills. I could of course be wrong, but would love to hear your thoughts on this.
I studied EE in school, and focussed on statistical signal processing for my Masters. My bachelor was from an okay school in India, and my master was from the top-ranked Asian Uni (National University of Singapore). Most of my professional experience consists of the 6 years that i spent as a research engineer in a research lab, working on various robotics / signal processing / ML type projects. I now work as a Data Scientist for a finance firm in Singapore. I have had no formal CS education, and am a completely self taught programmer. I do enjoy the academic side of CS, and have taught myself many topics in algorithms & data structures, design patterns, computer architecture, etc.
Also, if you have any observations or advise for developers who are NOT based in the US, i'd love to hear that.
Do you have any thoughts about what the general environment will be like, and is there any chance the place will be mired in bureaucracy? I'm relatively early in my career as well, and have never worked for a Big Co before.
I'd love to get in touch with you, but you don't have contact info in your profile. I'll be grateful if you can drop me a line - mailshanx at yahoo dot co dot in
It's not quite clear to me: care to explain?
The key is to get them to explain things until you understand the essence at a foundational level - hopefully you understand what combination of the notions of statistical independence / power / dimensionality / algorithmic complexity has birthed novel system or algorithm they built.
The process selects for people who have a) already built something substantial, and b) have the communication skills to explain it.
You will be surprised: you can make significant gains by including the zip code - i've seen that happen in a competitive setting. Where you live probably contains some signal about your credit worthiness.
Having said that, of course it doesn't make sense to simply feed the raw zip code to the tree. An appropriate encoding (most people would use a one-hot encoding, though there exist better ones) of the zip code will be key to extracting signal in a robust way.
>> ... the tree will have serious overfitting issues
Isn't it an almost standard practise now to use an ensemble of trees, such as a random forest? Decision trees have long been known to be prone to overfitting.