R or Python?

Anyone prefer R to Python? Asking for a friend's roadmap. 馃槈

30replies Oldest first
  • Oldest first
  • Newest first
  • Active threads
  • Popular
  • Obviously depends on context, but strong bias towards Python

    Reply Like 2
  • We're largely a Python shop, so we would prefer R. As Claus states, there are contexts where we use R as well, but the vast majority of our institutional knowledge revolves around Python.

    Reply Like 1
  • We have work being done in both. Anything machine learning is Python, as well as some modeling. Viz and statistical analysis usually lives in R.

     

    Long term I think Python overtakes R on viz and statistics, because Pandas, numpy, etc. function very closely to R while also being more efficient and flexible.

    Reply Like 1
  • My preference would be to support Python because often Periscope is used for quick prototyping of insights which are then productionalized into pipelines and Python pipelines are much easier to maintain than R while R is mostly interchangeable with Python when you include stats libraries.

    For context, I am Director of Data Engineering who actually does this productionalizing :-)

    Reply Like 2
  • Personally, I prefer Python, for the totally biased reason that I am much more familiar with it than R. 

    I found this fantastic, detailed, (and well-cited!) infographic comparing the two from the perspective of data analysis, take a look!

    https://www.datacamp.com/community/tutorials/r-or-python-for-data-analysis

    Reply Like
  • Awesome post  Alok  , I think it highlights both the use cases that Brett Farmer described. The biggest takeaway I liked was:

     

    But I guess another one that I liked (as an R aficionado myself) was:

     

    Reply Like
  • I think we have a consensus here #PythonForever.

    馃槑

    Reply Like
  • I think Python tends to be a favorite for computer scientists who tend not to do much with statistics. Indeed, many computer scientists don't really think statistics and probability is worthwhile at all. R, however, is much more practical and better documented than Python for statistics and statistical visuals. One thing to think about is that Microsoft, when it did Azure ML, did the R integration FIRST and then did Python in SQL Server 2017. I think this speaks volumes. The uptake on R to the wider world will probably tend to be swifter with business analysts as well - who will be forced to migrate upstream as their work is automated. Most of those folks will also tend to prefer to trade flexibility for capability. So, while some of us like to dual wield the two languages (and SQL of course!), I think many have very prematurely proclaimed the death of R and statistics. Many, many problems are not development problems - they're analytical ones that merit well known solutions where flexibility just isn't at a premium. I've seen this debate so many times, in so many places, but almost always between computer scientists looking for "the answer". Python is one of many "supposed to's" in data science. Such rigid thinking creates opportunities for less dogmatic thinkers, of course. 

    Reply Like 5
  • Python has overtaken R for us, as well.  My vote is for Python..

    Reply Like 1
  • We have a strong bias towards Python as well. Not to say that we wouldn't use R, but in my view having Python support would have a larger and more immediate impact.

    Reply Like 1
  • As a statistician, I'd vote for R!

    Reply Like 4
    • Asena Uyar I agree. Many statisticians and business analysts are not going to be interested in learning to code the way you have to in Python. I personally like both languages, but what Periscope needs to think about is that the market for business analysts is much larger than the market for data science geeks like us. Moreover, data scientists are more inclined towards build instead of buy anyway. So if you really think about it, doing SQL based BI + statistics is probably the largest market. I think Microsoft realized this already. Just look at MRAN, as well as SQL Server 2016 (which preceded 2017's Python friendly implementation).

      Reply Like 3
    • Adam Brzozowski Adam, that is exactly what I am thinking. I am already achieving my data transformation/manipulation goals with SQL. Also, a huge fan of SQL views and snippets (macros) in Periscope.
      What I need is to be able to conduct more statistical analysis at this point and I think R will be the best answer to that. 
      Again, I think the answer to this question (R or Python?) depends on what you are trying to solve. 
      I like this post a lot which compares pros/cons for both R and Python. 
      https://www.kdnuggets.com/2015/05/r-vs-python-data-science.html

      Reply Like
    • Asena Uyar I'm actually wondering when we'll see stored procedure like looping and variable declaration in Periscope. That would get us much more than even this language integration. 

      Reply Like
  • Academia seems to support R but a lot of tech companies have research teams that will focus on developing tools with Python since it's a more universal language. Tensorflow by Google is a good example of this. 

    Reply Like
    • mshmei yes that's true, but given that Microsoft deliberately developed MRAN, as well as chose to incorporate R as a database service first in 2016 over Python, I think there's more to this story than the conventional wisdom would seem to indicate.

      Reply Like
      • mshmei
      • mshmei
      • 1 yr ago
      • Reported - view

      Adam Brzozowski that's a fair point but I think a lot of smaller companies will still have a hard time integrating R with other parts of their technology stack even with the increase in performance offered by MRAN. Marketing analytics using the Facebook business manager API is a good example of this. For me, MRAN looks like a nice-to-have. I'm sure it increases the performance of your models but I still don't really see R as something people use for writing production code. I kind of see it as a language for prototyping where it's really fast to write and tune models. For instance, you can tune a randomForest pretty easily by playing around with mtry, ntree etc.

       

      To be fair though, I'll take mathematica over both anyday ;)

      Reply Like
    • mshmei  I tend to agree on desktop MRAN itself, but just take a look at how powerful the in database utilities are for Azure. I think a lot of scoring type activities will end up happening there in the future anyway for both scalability and security reasons. As such, there's an entire category of data mining that can be done quite well with a read replica and R. Many types of companies also lend themselves to this kind of structure as well. 

      Of course, that still leaves a huge amount of activities that don't lend themselves to that kind of production model and that's where Python will be very successful for a long time because of its licensing restrictions more than anything. 

      This is why I like to be able to wield multiple tools. If I had to think about this question as Periscope, I'd probably base it on my user base. Which is the largest group right now? Are they data scientists or business analysts? 

      Reply Like
      • mshmei
      • mshmei
      • 1 yr ago
      • Reported - view

      Adam Brzozowski hmmm, I'll have to give you a benefit of a doubt.

      Reply Like
  • One more question I'd have is which version of Python? 3X or 2X would be supported?

    Reply Like
      • Sean Cook
      • Product Analyst @ Facebook
      • Sean_Cook
      • 1 yr ago
      • Reported - view

      Adam Brzozowski We see more 2X, but ideally we wouldn't force people to make that choice  in the longer run.

      Reply Like
  • Definitely Python.

    Reply Like
  • I guess I'm in the minority here, but as a BI analyst with a stats background, I strongly prefer R.  I've also found that the younger applicants we've been interviewing lately are more experienced and comfortable with R compared to Python, so at least for us as a company I see that being the direction we head from the BI side.  I know on the engineering side they use Python much more frequently, but they also don't use Periscope as much as we do. 

    Reply Like 1
    • Jessica Tepper  Hi Jessica - this is my sense of things too. Engineers tend to be focused on scaling applications and algorithms vs trying to get answers to business questions. There are far more business analysts out there in the world too than data scientists - much to the chagrin of many. 

      Reply Like
  • Have you seen our new website? 馃槈

     

    Reply Like 1
    • Sean Cook yes! This is fantastic Sean. As always, Periscope is just over the top awesome by not fighting against customer preferences. Which license bands will get this new feature!

      Reply Like
      • Sean Cook
      • Product Analyst @ Facebook
      • Sean_Cook
      • 1 yr ago
      • Reported - view

      Adam Brzozowski Thanks! If you reach out to your account manager, Julie, she will have details as they become available.

      Reply Like
Like8 Follow
  • 8 Likes
  • 1 yr agoLast active
  • 30Replies
  • 1495Views
  • 17 Following