More Notes on SAS
SAS employee Scott Mongeau writes:
Journeying through the vast graveyard of open source vanity projects I come in on as a mop-up-agent on in any given month, one espies a dystopian wasteland of ruined towers and battered factories strung together with bubble-gum and tape.
If you want to see a vanity project, check out the SAS rock collection.
This comment comes from a lead architect at a major European financial services organization:
We’ve just completed a large SAS upgrade project. While we had no interest in SAS Viya, SAS legacy remains an important part of our landscape…
Would we invest in SAS if we were starting today from scratch? No.
But none of SAS’s biggest customers are starting from scratch, and things like continuity, risk, and embedded institutional expertise are incredibly important to these customers…
In the endgame though, I can’t see how SAS will be able to resist the cost pressure coming from the Python/R/Hadoop/FOSS-in-general ecosystem, as those tools continue to mature.
Insightful comment. Large enterprises have complex systems and can’t simply throw everything out and start from scratch. Call it the weight of history.
But in the long run, open data science will rule.
A commenter who calls himself “Mr. Toad” writes:
SAS have shown some belated movement towards a more modern architecture. You can install SAS servers using a package management system, and automate the process using Ansible…
But their licensing model is straight out of the 90’s. there’s no point using automated load balancing if your license costs are a fixed multi-million dollar per year cost.
Amen, Brother Toad. Amen.
A person who posts under the name of a town in The League of Gentlemen writes:
A few years back, I was an out-and-out SAS guy. This was mainly because the company I worked for had everything in SAS…
Then, our company acquired a competitor to SAS; SAS was booted out and literally, overnight, we were asked to stop using SAS and switch over. Thanks to this development, my team and I scrambled to port it over to the new tool. Of course, the transition wasn’t smooth and a lot of our programs couldn’t be migrated. This is when I began exploring Python and fell in love with it. I rediscovered programming, powerful ML libraries and the awesomeness of the open-source paradigm…
I end my random anecdote to say that I’m grateful for this (unexpected) development that helped me accelerate my career in data science, which would have definitely not happened if the abrupt removal of SAS hadn’t taken place. I would’ve most likely still be churning regular dashboards built on legacy SAS code for some bank.
Or drawing unemployment after the latest round of “early retirements.”
About a year ago, a sales guy from SAS came to give a talk about Viya at our university…
The presentation included showing off a Jupyter notebook and showing how you can easily use pandas to load in your data set, build a model (with Viya), and plot the results with matplotlib. Standard data science stuff, but distributed automagically thanks to Viya…
During the QA, I asked the following questions:
— So you are basically attempting the same as what Spark has done, and are just using R/Python/…as a client? Yes, but we’re way more powerful, manageable, etc…
— Can I see the source code of your models, or build new ones myself? No… well maybe, you can write them in SAS base and call them.
— Can I at least export the model to something like PMML? Oh yes, you can export (shows this off)… to SAS base code…
I like SAS, they have some very smart people working there and some great consultants (PhD’s, often), but there’s too much sales going on. This is what killed IBM as well.
Heh. Too much sales going on. Carl Farrell begs to differ.
Lest you think I’m cherry-picking the negative comments, consider this post by someone who calls xerself “bigger_cheese”:
We use SAS Enterprise Guide as a Business Intelligence/analytics tool at my work (Industrial Plant) – most of our production reporting is SAS based. We also use it for adhoc analysis/troubleshooting. I.e to spot long term declines in yield and that sort of thing…
Where SAS succeeds is they make it very easy for non programmers to do things like query databases extract, transform, merge data, create computed columns and output graphs and reports…
SAS to me is a bit like MATLAB (and hey we use both in Engineering world). Everyone hates this software and wants something better but nothing else approaches the usability of these tools so models continue to be built using them.
Agreed. Enterprise Guide is a nice product. I’ve used it myself. It’s mature, stable, and full of features. It’s also bundled for free in the most basic SAS license. Which is why SAS never shows it to analysts: SAS’ business model depends on pushing new products.
Change accelerates. Companies that were solidly SAS-centric a few years ago now embrace open data science.
Recently, I met with a large SAS customer. On the surface, the buyers seemed very committed to SAS. LinkedIn profiles for key people in the room touted SAS backgrounds. On initial discovery, the client revealed almost exclusive reliance on Legacy SAS — Base and STAT, for the most part.
The client’s most senior executive — we’ll call him Mr. Big — kicked off the discussion. “We want more scalability, more speed, and more throughput. I need my team to be able to build models faster and to deploy them faster. Our clients demand more speed and more innovation.”
“Okay,” I said. “There are two different paths you can take. Option #1 is to build on your investment in SAS. Viya provides scalability. Factory Miner can accelerate your model-building process. Model Manager and Scoring Accelerator will speed your model deployment process.
“Option #2 will be harder. You can deploy open data science tools, such as Apache Spark, Python, and R. Retrain your people to use these tools. Consider using a commercial tool such as Dataiku, DataRobot, Kogentix, or RapidMiner to broaden access to a wider range of users.”
Mr. Big pondered the options for a moment, then spoke: “Tell us more about Option #2.”