Database Reference
In-Depth Information
If you are using R and you actually want to deploy the code into some live
web service, then quite often you are better off doing it in some other lan-
guage, depending on how big the load is going to be. You have to really think
about how many transactions are going to go through your model or system.
In some cases, you just have to choose C++ and its libraries because you
have to optimize at a very low level. This is what we've done at Planet OS
on a few occasions where there are so many computations involved that it's
necessary to optimize the machine learning. In other cases, you just want a
quick exploratory answer. Then you can use tools like Vowpal Wabbit, Weka,
or some other library.
Gutierrez: What have you learned from building successful data teams?
Karpištšenko: You need different roles and different personalities on the
team. You need statisticians, software engineers, machine learning experts,
system engineers, visualization experts, interaction experts, product manag-
ers, and business development experts. Oftentimes, not all of these roles are
specifically within your team. Some of these roles could be your stakeholders
instead. But if you look at your extended team, you have to have all of these
different perspectives involved. And if you don't, then you usually end up not
seeing your result go into the live production environment or finding success
that fast.
The best examples I've heard about where this was true have been when
operations have been disregarded. As people deployed a new service, they
found out that the network configurations of the system that they used would
not fit into the existing infrastructure. As they examined this issue, they found
that the rabbit hole went deeper and their results getting deployed so late
that the models were already outdated and they had to start over to build
new ones.
Not only do you have to have all the roles, you also have to have the right per-
sonalities. You don't want to have everyone in exploratory mode. You want
some people to be very focused on development. You want some people to
be very focused on quality testing. You want some people to be the cheerful
team members who get you through the highs and lows. Every personality in
this mix of personalities is relevant because it enables frequent regular com-
munication patterns and allows for team building and reflection.
To build successful teams and projects, I strongly believe in the Kaizen approach.
Kaizen was made famous in part by Japanese car manufacturers involved in
continuous improvement. I believe you should always be looking for ways to
improve things, just small things. Just try it out. You'll fail sometimes, but over
time things will become better. If that involves people aspects and communica-
tion aspects instead of the normal project aspects, then that's all great. Usually,
things fall apart because of not improving the people aspect.
Gutierrez: How do you think about hiring?
 
Search WWH ::




Custom Search