News for hackers. Stuff that matters.
Re:Ask HN: Data scientists, what does your workflow look like? by uptownfunk on Monday November 05, 2018 @01:24AM (#18380285)
One of the following:

RStudio > write.table(x, “clipboard”, ...) > paste to excel > email data to BA who makes slide

Python/Anaconda + jupyterlab nbs + sklearn

Excel + Solver + PowerPoint

Obviously track everything on Git etc.

Many more things you can do here:

Use R notebooks, Jupyter notebooks, even have a build server and make each one of your projects an R package

Re:Ask HN: Data scientists, what does your workflow look like? by tixocloud on Monday November 05, 2018 @06:59AM (#18381615)
Interesting. Do you code in both Python and R? What would be your rational for picking one over the other?

Also, would you have a need for a build server?

Re:Ask HN: Data scientists, what does your workflow look like? by uptownfunk on Monday November 05, 2018 @09:17AM (#18382431)
Yes code in both. Depends on who I am working with really, prefer R but python much easier to integrate in a production environment.

I don’t have a need personally but some companies that are doing industrial scale modeling (on the order of building and maintaining thousands of models) do use a build server to Basically check that code is formatted properly and can have a model run in a somewhat automated fashion.

Re:Ask HN: Data scientists, what does your workflow look like? by tixocloud on Tuesday November 06, 2018 @01:05PM (#18393361)
What makes productionizing R difficult?

I'm curious about whether the rise of Python in data science is really just because of the lack of flexibility to integrate with other systems. I've read that R seems to do better in data science/analytics work but when it comes to integration, it's more challenging.

And any insights on which companies are already doing industrial scale modeling?

Re:Ask HN: Data scientists, what does your workflow look like? by RayDonnelly on Monday November 05, 2018 @09:36AM (#18382565)
Do you use Anaconda's R. If so I'd like to know of any rough edges (package count being a known one we're looking to address).

Thanks.