Python Interface

[harrisonzhu508]

To make it easier for further development within the community, I suggest we convert some of the pre-stan code in base.r into Python. Should be mostly a pandas-numpy-datetime job. After which we can use the Pystan[https://pystan.readthedocs.io/en/latest/] interface to compile the Stan code and ingest data via Python.

[am5113]

Hi Harrison,

I have been working on this for last couple of days in a private repo. It's basically done. Just need to do some testing.

What do you think is the best way of making this open source? Should the Python version have it's own repo?

[s-mishra]

Hi @am5113 the best option would be to add that as a pull request to this repo with python as a new folder. However, anything that gets added needs to be updated regularly with changes to the model.

[dentarthur]

Hi @am5113

I've put some extensive notes out of the way, here:
thecapitalistcycle#2

Sorry, no time to make short and clear.

About subsequent steps unleashed by yours.

[harrisonzhu508]

Added the main data-wrangling class to the python branch up to line 166 in base.r. People can then inherit this base class to play around with their own models. https://github.com/ImperialCollegeLondon/covid19model/blob/python_integration/notebooks/Test.ipynb Here's an example to my attempt.

The rest should follow up quite nicely by using pystan and whatever wrangling is needed. The plotting code can remain in R for our official results, whereas the community is free to create any visualisation they would like.

[am5113]

Hi @am5113 the best option would be to add that as a pull request to this repo with python as a new folder. However, anything that gets added needs to be updated regularly with changes to the model.

@s-mishra, I agree. I'd be happy to stay on top of this. I'm sure there's an army of Python coders who will make sure of it too.

[harrisonzhu508]

@am5113 thanks for your enthusiasm to contribute to our project! I will soon submit a PR as well with an implementation that I have done as well - I think we should merge our work to avoid replication :)

[defjaf]

Both, for what it's worth I've also got a Jupyter notebook that also replicates the data-wrangling features of base.r. Ignore if it just lowers S/N but let me know if I can help.

[harrisonzhu508]

@defjaf oh that's cool stuff - do you have a link? I can show you the class I made for data wrangling to compare:
https://github.com/ImperialCollegeLondon/covid19model/blob/python_integration/notebooks/Test.ipynb

https://github.com/ImperialCollegeLondon/covid19model/blob/python_integration/Python/dataset.py

[am5113]

You can also check mine out at: https://github.com/am5113/covid19model/tree/master/Python

I put in a crude test using rpy2 to check that the consistency between R and Python.

[harrisonzhu508]

A pull request has been made to develop an initial prototype #13

[defjaf]

There's a working notebook at https://github.com/defjaf/covid19model/blob/master/python_ahj/COVID19-base.ipynb

It's basically a direct translation of base.r into python; not quite as pythonic as the HierarchicalDataset class...

It uses numpy, pandas and pystan, as well as some utility code (adapted from Michael Betancourt <@betanalpha>) included in the repo.