Jason Github

Post date: Sep 28, 2017 8:7:10 PM

Scott,

I have something more to say in response to your email, but I just stumbled upon this paper, which you might already know about. I did not read it yet, but the graphical abstract is a pbox.

Scott,

Very cool. I had been trying to follow the literature on this topic for a while. Alessandro Antonucci from the IPG seemed to be a leading voice on this topic at the time, but it looks like people working on (Bayesian) deep learning have also become interested. See here too.

I think that the easiest way to get the pbox.r project on to GitHub is to download the Windows GUI and fiddle around with it until you feel comfortable enough to rely on it for development. Once the code is on GitHub, it will be searchable. Guaranteed open-source fame.

I think that it will be a relatively straight-forward process to write interval/pbox libraries for Python, using pbox.r as a guide. The library would (should) act to extend numpy/scipy to allow for more sophisticated interval and pbox calculations. There does exist a Python library for interval arithmetic, so it might be worth investigating.

pbox Python, and pbox.r on Github/CRAN

Scott,

I have been wanting to ask if there is interest in putting pbox.r on GitHub. GitHub is super easy to use these days, and there now exists a really nice GUI interface (a desktop application) for Windows.

I also wanted gauge your interest in developing interval statistics and pbox libraries for Python.

Both require work, of course, but I have this fantasy that sending this email accounts for half of it.

Jason:

I think all of those are great ideas.

I've been talking to a few people about these ideas, particularly the interval stats stuff, including you, Luke Green, Shibo Yao now at Rutgers, and Marco de Angelis here in Liverpool. Lev may have proprietary interests, we might want to consider.

It looks like I will finally have a funded PhD student whose project will be to introduce epistemic uncertainty into machine learning algorithms. The problem will be to find some competent young person who is interested in this problem.

Maybe we should have a group discussion about this, perhaps asynchronously if necessary. On the other hand, talking just slows us down.

Scott