Machine Learning Spends A Year In Burgundy

Note: this post in its present state is a placeholder for a later transcription of my talk at the 2018 ML4All machine learning conference in Portland, OR. It contains references to material, code and image credits related to my talk. Ultimately, it represents one technologist’s (my) fumblings towards an understanding of how to apply machine learning, and a collection of breadcrumbs aimed at helping other technologists make a similar exploration.

my framework for applying machine learning to a problem

General Reference links

https://www.npr.org/sections/ed/2017/12/12/568378251/does-preschool-pay-off-tulsa-says-yes Amazing study on the long-term effects of attending preschool… <soapbox>why is federal funding for preschool so low??? </soapbox>

https://en.wikipedia.org/wiki/Linear_regression

https://en.wikipedia.org/wiki/Combinatorial_explosion

Wine-specific links

For the annual vintage scores for Cote de Nuits Burgundy, I used an average of scores from Robert Parker and Wine Spectator.

https://www.robertparker.com/resources/vintage-chart — source of annual regional scores

http://www.winespectator.com/vintagecharts/search/id/30 — source of annual regional scores

Viticultural links

https://learn.winecoolerdirect.com/life-cycle-of-a-wine-grape/ — annual cycle of winegrape maturity

https://www.evineyardapp.com/blog/2017/03/01/why-the-need-to-calculate-growing-degree-days-in-vineyard/

https://en.wikipedia.org/wiki/Growing_degree-day — what the heck is GDD

https://en.wikipedia.org/wiki/Winkler_index — how the heck do you apply GDD to grape growing and ultimately winemaking

http://www.oregonwinepress.com/article?articleId=783 — specific growing degree day information for pinot noir grapes

Machine Learning-specific Links

https://www.quora.com/Why-is-Machine-Learning-difficult-to-understand — provided the beginning of my approaching ML framework

http://dataconomy.com/2016/03/the-perfect-pairing-machine-learning-and-wine/ — prior work that pointed me in the right direction

https://github.com/ZuzooVn/machine-learning-for-software-engineers — awesome collection of links

https://hackernoon.com/machine-learning-with-javascript-part-1-9b97f3ed4fe5 — ultimately failed, but gave me confidence that I could use JS

https://medium.com/@MaximilianLloyd/multiple-linear-regression-in-machine-learning-4711209604b7 — source recipe for the ‘done by hand’ solution… this is really a Keys To The Castle post in terms of applying machine learning using JavaScript and it deserves WAY more love than it has

Code Written

This is my attempt at implementing a multiple linear regression supervised learning algorithm on my own. And by on my own, I mean, I followed a recipe almost to the letter.

Images used

(THANK YOU to my fellow flickr users for making your images available through the Create Commons license)

https://www.flickr.com/photos/mwichary/2283389280/

https://www.flickr.com/photos/johnnysilvercloud/14891866586/

https://www.flickr.com/photos/ab-streetart/5905015911/

https://xkcd.com/1425/

Summaries / Conclusions

coding by hand vs. using a library vs. ML-as-a-service

Software engineer and team leader by experience, startup and business nerd by nature. PDX. https://jonoropeza.com

Software engineer and team leader by experience, startup and business nerd by nature. PDX. https://jonoropeza.com