Hi ,
This post was meant to go out last Thursday, but something went wrong with the email scheduler and somehow it didn't. As a result, I'm sending it now. I will return to the regular schedule from tomorrow onwards.
What are your views on AutoML?
That is, tools, such as H2O.ai, that simplify the training of ML models by automating the parameter tuning and algorithm selection processes.
Although I can see the advantages of such tools - particularly with regard to time
savings and the democratisation of ML - my preference is still to perform these tasks the "old fashioned" way.
Performing parameter tuning and algorithm selection manually makes it easier for me to understand the hidden intricacies of my models and know what to do when things go wrong - which they inevitably will.
ML models are inherently complicated and obscuring those complexities has the potential to create more harm than good.
It's for this reason that start-up finance expert Lauren Pearl's unconventional approach to teaching financial modelling to start-up founders resonates with me so strongly.
Rather than starting with Excel - the standard tool for building financial models - Lauren has her students start with pen and paper.
By writing the core equations underpinning the model in words, rather than obscuring them behind Excel formulae, model users are
better able to understand the levers that underpin their business - enabling them to better understand their models when they do eventually encode them in Excel.
Statistician George Box once wrote that "all models are wrong (but) some are useful". For a model to be useful, the model developer and end users must be able to understand what is going into the model and what is coming
out.
This is true, whether you're dealing with a deep neural network or a simple Excel financial model of a newly launched start-up.
More sophistocated tools, from Excel to AutoML, may save you time - but you can't short-cut gaining understanding.
I had the opportunity to speak to Lauren Pearl about financial model building in a recent episode of Value Driven Data Science.
To listen to our full conversation, click the link below: Episode 34: Financial Modelling for Start-Up Founders
Talk again soon,
Dr Genevieve Hayes.
