Fork me on GitHub
#data-science
<
2016-03-04
>
actsasgeek01:03:58

I love Clojure and your YMMV but I’ve found that Python is going to be the better choice for now. There are simply a lot resources in Python: books, tutorials, examples, libraries. Moving from Octave (Andrew Ng’s course) to Python/Numpy is trivial. If you really do have a year or two, spike a prototype in Python, learn Clojure on a background thread and in 6 months see how you feel.

kaiyes01:03:21

@actsasgeek: thanks for the advise. I got exactly this same answer from a local clojure dev

actsasgeek01:03:30

you might also look at Programming Collective Intelligence (book). I find it’s a good intro for people. It covers some things not in Ng’s course.

kaiyes01:03:18

one thing that is a bit confusing to me is what tool does what. but I guess only time and studies will fix that.

kaiyes01:03:01

for example...storm is written in clojure. So choosing that (if needed) should work well with clojure.

actsasgeek01:03:56

You have to start with the problem and work backwards. It can’t start with “I need machine learning” or “I need a data infrastructure”. It starts with “I have a need to do X.” Does X involve text? Does X involve supervised or unsupervised learning? If supervised, then is it classification or regression? At scale, will I be moving truly big data or can I just use Postgres.

actsasgeek01:03:14

You don’t start a project by saying, “I like drills. I need a drill."

kaiyes01:03:06

was about to write that...project description....

actsasgeek01:03:04

And it’s kind of hard. If you don’t know yet what supervised or unsupervised are, or if you should start with logistic regression or deep learning, you might need to do some research. Andrew Ng (Machine Learning), Jeff Hammerbacher (Data Science)...they’ll say the same thing. Start simple, get yourself a baseline. Iterate and improve. So what if your hot new text clustering app uses TF-IDF and k-means. Start simple. As the old XPers used to say, build one to throw away.

actsasgeek01:03:01

they also used to say, it’s easier to build a 4in mirror then a 6in mirror than just build a 6in mirror (apparently it comes from hobby telescope making?)

kaiyes01:03:58

sorry was in a phone...moved to a pc now

kaiyes01:03:19

ya, you are right.. I am just a beginner

kaiyes01:03:27

and wanted to get a high level overview of whats possible in clojure in comparison to python...so I could focus on whats right infront of me

kaiyes01:03:15

but thanks....I think I should stop asking questions now

actsasgeek01:03:24

haha, no worries. Ng’s course is a great start. I think the PCI book will be helpful too.

yogidevbear20:03:58

Hi everyone. So glad I found this channel simple_smile I've been looking at data science and machine learning for a few weeks and am very keep to steer my career in this direction. Is there anyone in this channel to works with data science or machine learning on a daily basis?

base69820:03:46

if anyone knows a way to distr, but with incanter

base69820:03:02

I'm calculating the log normal mean by hand currently

base69821:03:18

This is the function I have for it:

base69821:03:45

(defn log-normal-mean [mean var]
  (Math/log (/ mean
               (Math/sqrt (Math/pow (+ 1
                                       (/ var mean))
                                    2)))))

jonahbenton21:03:01

hey @yogidevbear: from time to time, not on a daily basis

yogidevbear21:03:54

Hi @jonahbenton. I saw your links above. Thanks for sharing. You answered my question before I asked it :thumbsup:

jonahbenton21:03:38

great, happy to help simple_smile