Notes of the month, Late October (some people might call it 1. November) edition

This time just a brief update - life is good, but a bit busy. Machine Learning is cool, but not a silver bullet. And working with smart people on hard problems is fun!

Also, we'll be in London around New Year's Eve - let me know if you're up to something.

Events
29.-30.10.2019: Neudata - Alternative Data Conference
https://www.neudata.co/events/london-conference-2019
21.11.2019: Some applications of ML in Finance (Frontiers), Citi Stirling Square
https://www.maths.ox.ac.uk/node/34031
13.-14.11.: Bid Data London
https://bigdataldn.com/

Bank of England: Machine Learning in Finance
BoE aggregated the perspectives of asset managers, banks and insurances on the current state of Machine Learning in UK financial services. It turns out that the majority use it already somewhere / in some form, but the general expectation is that the usage will double over the next three years. The most prominent use case seems to be AML / anti-fraud, with applications in other areas picking up.  
https://www.bankofengland.co.uk/report/2019/machine-learning-in-uk-financial-services

150 successful ML models
Booking.com rolled out already quite a few ML models in production - and they were kind enough to sum up their main insights in a paper. To save you some time, here are my main take-aways (it's definitely worth reading the paper, though):
- Offline performance metrics are just a health check
- It matters how you set up your machine learning experiment
- People really don't like waiting for prediction results
- Monitor your models, for example with response distribution chart
- Set up experimentation through randomized controlled trials
https://blog.acolyer.org/2019/10/07/150-successful-machine-learning-models/

Apache Superset
I just came across a rather useful and open-source visualisation tool for those among us that don't have a Tableau license available at work. To use is, you just point it at a database, pick a table you want to look at and choose the format - table, graph, geo-plots, whatever you prefer. Take a set of graphs/tables, put them together on a screen and you got your dashboard.
Of course it's not quite as smooth as Tableau or as flexible as a hand-crafted plotly dashboard, but still very useful for quick visualisations.  
Here is the link: https://superset.incubator.apache.org/

Now reading (again): The 24-hour Wine Expert
A surprisingly casual book on the basics of wine, the grapes, the regions and some of the more obnoxious traditions around it. An entertaining read and it features some good lists of wines to explore - that's why I went back to it.
https://www.amazon.co.uk/24-Hour-Wine-Expert-Jancis-Robinson/dp/0141981814

Show Comments