Notes of the month, London summer edition

I just realised that I collected a bit of a backlog recently - therefore, without much further ado, the (by-now-not-really-)monthly random link collection.

Spatiotemporal modeling with integer-based autoregressive processes
The first time I encountered this type of model, but fitting integer-based autoregressive processes on low-count event time series just makes sense. Now separate the spatial and temporal dynamics and it seems like you get a pretty decent model for crime events.
https://arxiv.org/pdf/1304.5642v1.pdf

Backtesting distributional forecasts for counterparty credit risk
This paper has an interesting look at the problem of backtesting long-term forecasts based on a shorter set of historic data. The fundamental issue with "past performance does not indicate future returns" remains, though...
https://papers.ssrn.com/sol3/papers.cfm?abstract_id=2930047

An introduction to quantile regression
When addressing pretty much any type of prediction problem, you want to get an understanding of the associated uncertainty. On an aggregate level, this can be done by evaluating the loss function on a test dataset - on the level of individual datapoints, this can be done by predicting the distribution of possible values. Here is a soft introduction to this topic.  
https://towardsdatascience.com/quantile-regression-from-linear-models-to-trees-to-deep-learning-af3738b527c3

TabNine - the autocompleter you didn't know you are missing
Autocomplete in Gmail, where you can tab through half of your more generic emails, is already pretty cool. But you know what's really great? Smart autocomplete for code.
https://tabnine.com/

And some more links focused on useful python development tools:

Code formatters - Flake8, Black and  autopep8
A code formatter in a pre-commit hook really helps to improve consistency and readability. Out of these three options, Black is the most opinionated one, while autopep8 (as the name states) only ensures PEP 8  consistency. I'd personally go with Black and forget about the topic.  
https://flake8.pycqa.org/en/latest/
https://pypi.org/project/black/
https://pypi.org/project/autopep8/

Doctest - since not many people really enjoy writing separate tests
Probably most of you already know about this one - you put a line of python code in your docstring and the expected result, that's it. It's a fast way to prevent the most obvious regression bugs, but gets a bit out of hand as tests become more complex.
https://docs.python.org/2/library/doctest.html

Also, in completely unrelated news:

Pizza arbitrage
That's what you get when an online delivery platform onboards restaurants without their consent and misprices the items...
https://themargins.substack.com/p/doordash-and-pizza-arbitrage

Our future - what it is like to live in the hottest city in the world
While I very much enjoy the current weather here in London, this article gives an idea what the future will look like for more than one city on this world. The second link are the IPCC projections for long-term climate change.
https://time.com/longform/jacobabad-extreme-heat/
https://www.ipcc.ch/site/assets/uploads/2018/02/WG1AR5_Chapter12_FINAL.pdf

Show Comments