5 interesting things (24/01/2015)

Top 30 books ranked by total number of links to Amazon in Hacker News comments – I must say this analysis surprises me. First because of the low numbers – most common book was mentioned only 53 times. Second, I expected “The Lean Startup” to be one of the 30 top books. I guess one of the caveats as described in the post itself are the links themselves that may differ a bit or change over the years and therefore hard to match.


LDA2VEC – getting the best from both worlds LDA + word2vec. Stitch fix definitely brand themselves as one of the leading companies  technology and research wise doing some very interesting things.


Automatic colorizing – coloring gray scale images using deep learning. There are many works now trying and checking tensor flow particularly and deep learning on general. I really liked this case, maybe also because it uses a set which was already trained for other purposes.
Gestures Typing – First of all I liked the question – answer format of the post. The post looks for an optimal keyboard. But what is optimal keyboard? depends who you ask. On the other hand I think qwerty (or qwertz in Germany) is such a strong standard that I wonder what will make it change.
Roomba + voice recognition – problems of the very rich part of the world. Yesterday night we spoke with friends which complained their Roomba sometimes get stuck in all kind of corners, cable and so on in their house. We talked about several ideas to solve it, one was voice command and this morning I found a guy which already tried it –

5 interesting things (10/01/2016)

Everything.me open source their inheritance – Everything.me was an Israeli startup closed few weeks ago. They now open source major parts of their code and tools including their prediction algorithm, Re:dash and others.


Python 3 module of the week – over the years Python Module of the week by Doug Hellman was one of the most reliable documentation resources for Python’s standard library. The documentation is always accompanied with very good examples to almost any functionality. It is also available as a command line tool and was translated to Chinese, German, Italian, Spanish and Japanese. And.. it is now updated to Python version 3.5. Kudos.


The Star wars social network – although I’m not a star wars fan I think this provides a very accessible introduction to graph algorithms and measurements.


D3 in Jupyter – an intersection between 2 tools which I use quite a lot and find very important for data scientist. Not those tools specifically but tools which make the data science magic more approachable to others so one can share its’ findings and get feedback.


Bonus – Building interactive dashboards with Jupyter – http://blog.dominodatalab.com/interactive-dashboards-in-jupyter/

Tl;dr man page – given a bash command creates a tldr of the man page. More of a gimmick but a nice one.


Surviving Black Friday & Turning Behavioural Signals Into User Profiles

I was on a visit in Israel and went to “Surviving Black Friday & Turning Behavioural Signals Into User Profiles” meetup. It is a very long name which actually implying on 2 talks. The meetup took palce in Sears Israel offices, which is the department behind Shop Your Way.

The first talk – “Surviving Black Friday” was given by Omri Fima who is a resilience tech lead in Sears. Omri talked about resilience and scalability lessons learned based on Black Friday. Every internet shopping site knows that the traffic is much higher on Black Friday, so how can you prepare and test if your system can deal with such a load? What happens if one service fails? What is a graceful failure and what is less graceful? He presented 7 steps to make your service more stable and mentioned few tools both for testing and development.
Pablo Rosenman, VP Development @ Adience gave the second talk – “Turning Behavioural Signals Into User Profiles” (slides, video). Pablo presented 2 of Adience products – Adience SDK and Events SDK and showed how they use AWS services in their pipeline. He talked about Adience pipeline and what were their main concerns and focus when designing and implementing it – scability, decoupling  and cost effective. In the end he also presented what they would do differently if they would design it today.
All in all, interesting talks and a very good atmosphere. Looking forward that Omri’s slide will also be available online.