Re: Not to incite a language war but apparently the Corona lockdown was based on 13 year old undocumented C-Code

[Olivier Henley <olivier.henley@gmail.com>]

On Saturday, May 16, 2020 at 1:01:05 AM UTC-4, Anatoly Chernyshev wrote:
> "Clowns will dominate the pandemic-forecasting domain..."
> 
> I like that one...
> 
> Oh dear, you've lured me out. As a person who just published a Covid-19 forecasting model programmed in Ada, I can't stay away from this heated conversation.
> 
> There are so many concerns raised about scientists doing their rookie programming without resorting to professionals, so I don't even know where to start.
> 
> First of all, we are curious bunch, and learning something new (like programming) is always fun. Second, in scientific programming the chance that someone in power will lock down a country based on the output of your program is negligible (certainly it wasn't on the radars 13 years ago). Third, fourth... 
> 
> To make a long story short, for the experiment's sake, is there anyone willing to review my code? It's not a marvel of Ada programming, it just works.
> 
> Yet it's quite short - I went from the model's formulation, programming, testing, data acquisition, manuscript writeup in a matter of a fortnight.
> 
> If the review is useful, in terms that it helps to produce more accurate forecasts or identifies a drastic flaw in the code, I would be more than happy to do a revised version of the manuscript in a coauthorship with the reviewer(s). Or even go for a peer-reviewed publication when dust settles.
> 
> The model forecasts are here:
> April 10th: https://xph.co.nz/temp/album/
> May 12th: https://xph.co.nz/temp/album2005/
> Some technical details: https://xph.co.nz/index.php/covid-19-progression-model/
> The model description: https://www.medrxiv.org/content/10.1101/2020.04.03.20052985v1
> 
> This is not an epidemiology-based model (so you don't have to be one). Rather, it's a phenomenological one. It can predict what will happen, but can't (yet) say why this will happen. A contribution on the model is welcomed too.
> 
> The code is not published (since I didn't expect anyone would care), but I'll do it once there's interest.
> 
> If keen, please write to me directly: a~at~xph~dot~co~dot~nz
> 
> Anatoly

Hi Anatoly,

Maybe, if you agree, are interested, and your code is not a nightmare to work with (refactor to expose things needed to drive it in UI), I could find time to code a graphical tool for it. Please consider what I did for Gautier's work (the screenshot does not show but everything is dynamic and the export to .csv works like a charm). It took me half a day learning the qt5ada binding on the fly:

https://github.com/ohenley/covidsim

My plan would be quite straightforward; in loose Model-View-Controller terminology:

1. Take your 'engine/simulation' code, make a git repo, publish it as a fork on GitHub. No fork this time if your code is not on GitHub already. Would be best for you to own it (create a Github user, upload your code, etc. But if you are not up to that, I will take care of it. You can collaborate on my repo with proper credentials explicitly set anyways.)

2. I refactor your code to push at the frontier, only parameters and simulation outputs to be driven/drive the UI layer. I do not touch the simulation calculations, this is not my expertise, and if the engine internals break I want *you* to fix it. I only/mostly touch on the inputs and outputs of those computes for them to become parametrizable externally. (This layer is the Model, your know-how.)

3. I create the visual layer, expose the parameters and behavior we want for the tool. (This layer is the View and Controller). I make another git repo out of this. This layer owns your layer to 'talk' with it.

We end up we 2 codebase/repos, like that, you and other scientists can fix the 'engine' without me knowing/caring because you touch on the 'model' layer only.

Anytime you come up with new features/computations etc working, you tell me, I repeat steps 2. and 3. to have the feature exposed in the visual tool.

Let me know what you think if you are interested. I will forward this to your mail too.

Olivier

p.s: If your codebase is huge/complex/lot of features maybe we can start to expose only a subset in a tool. From there, because it's available through github, people in this thread could collaborate (clone, edit, push)