Anomaly Detection and Conference Round Up
Happy February everyone,
Last month was an unusually social one for me with two virtual events.
First was a long term favourite of mine; MeasureCamp. And then a new one (for me); Bhav Patel’s CRAP Talks.
I’ve outlined my part in both of these events below, but before that I’d like to point out a new feature in Forecast Forge; anomaly detection. You can read more about this here. I hope to add more machine learning algorithms that work on timeseries data over the coming months. I’m thinking “breakout detection” will be next up but if you have something else you’re desperate to see then please reply to this email and let me know.
MeasureCamp
This was my first time attending a virtual MeasureCamp although I knew a bit of what to expect because I had been involved in a test event in lockdown number one.
I presented these slides to introduce a discussion on “MLUX”; the user experience of working with machine learning.
In summary:
- I am not talking about the consumer UX of products such as newsfeeds, Spotify/Netflix recommendations etc.
- Instead, I am interested in analysts and practitioners as users of machine learning.
- At the moment this splits into two very different groups of users. There are people who can program custom loss functions in Tensorflow and train new algorithms on their own data, and there are those who try to use the results of these algorithms without being able to change them.
- I don’t like this. I think it alienates people in the latter group from their work (often for no good reason) and more and more people are ending up on the wrong side of the divide.
- With new developments like mega-large neural networks it is possible that only a small handful of people will actually understand what is going on with these models and they will all work for three or four companies in Silicon Valley.
- As far as I know there isn’t a perfect solution that fixes all of this at once. In some cases there will be a business imperative to use a third party algorithm that you don’t understand simply because it performs better than any of the alternatives.
- But I hope that there is a gap where more specific domain knowledge of the problem combined with a bit of algorithmic help can lead to a better solution. This is where Forecast Forge sits and, based on what I’ve seen so far I am optimistic!
The discussion that followed didn’t really talk about forecasting at all (which is fine - this is a bigger problem than one area). People expressed frustration with things like product recommendations that lag six months behind the current fashions and the difficulty of comparing the uplift from different ways of doing things.
From one angle it is easy to say that the problem lies in a lack of skills/training for the relevant people. But this will never be a complete solution because there are so few places where one can learn about the development and training of world class models.
In my (biased - look where you are reading this!) opinion improvements in the tooling is more likely to be the best way forward. Libraries like Keras or Huggingface’s transformers allow someone with my level of skill to get started in this area - I would not be able to code or train stuff like this from scratch. And I think similar tooling can and should exist for other people too.
CRAP Talks
In this case CRAP stands for Conversion Rate, Analytics and Product. I have been aware of it as a offline meetup for a few years now but because it is mainly in London and I don’t have a focus on conversion rate optimisation or product analytics I have never made the effort to attend. This is my loss as I thought the virtual event was excellent.
The main part of the event was a presentation from Facebook’s marketing science team about Robyn which is a new open source project they have launched to help with Marketing Mix Modelling.
I have done a small amount of MMM in the past and I wish Robyn had been available then. I was learning as I went along without the support of someone with more experience so I spent a lot of time thinking and worrying about the details of what I was doing and whether or not it was the right approach. A page like this would have saved me so much stress even if didn’t also come with code that does all the calculations for you.
Following this presentation there was a discussion on forecasting which Bhav was kind enough to invite me to be a part of. I think he was hoping for some juicy arguments but (boringly!) I mostly agreed with everyone else. A quick summary of some of the points covered:
- Pick an error metric that matches your business needs and then work to find a forecasting model that minimises it. [I also believe that you should look for a model that is well calibrated in that close to 80% of the data lies within the 80% interval, 90% within the 90% interval etc. But I don’t have strong opinions on how to balance this with the error metric. What to do if one forecast has worse errors but better calibrations?]
- Once you have a model you should stick with it for a bit - how long exactly will depend on your business tempo. You can update the forecasts as frequently as you like but the underlying model should stay the same.
I think I went a bit off piste talking about how your forecast is your “model of the world” but people seemed very tolerant of my ramblings ;-)
I look forward to seeing what else comes out of CRAP Talks in the months and years to come.