A better understanding about machine learning in less than 15 minutes

analytic Coffee!,

3blue1brown Here is another podcast that can be helpful in about 1 hour to learn how computer-based neural networks work. 

playlist?list=PLZHQObOWTQDNU6R167000DxZCJB 3pi 

Thanks, Tom! I'd also recommend Data Smart - walks you through machine learning ideas using Excel with a brief intro to R. I just got it and started working through it.

Kyle Wilbert Data Smart looks like the next level up in terms of being able to take steps to actually build an ML Model.  Have you made your way all the way through this content?  If so what do you think about what you learned?  Have you been able to build any models that have been useful?

Thank you both for these links/resources!

I've not yet had a chance to work through the book. I read the first couple chapters and also read through the chapter on forecasting. I built a similar forecasting model based on a course in Lynda, but the one in the book has more explanation about types of smoothing, when to use them, and how to write the formulas. I've used it to create a rough model to help us know when and how often we need to repair our interactive pen devices. I've also used it to get an idea of what our ticket sales may look like 3-4 quarters into the future. It's similar to Facebook Prophet. The K-Means Clustering chapter had a nice example of segmenting your visitors based on purchase history. Definitely applicable for most of us.

Kyle Wilbert

Have you run into Survival Analysis/failure analysis?

Sounds like you are trying to do some failure analysis for your pens. 

Here is an interesting article I just ran into today discussing the use of Survival Analysis for things like Churn Modeling.  I've become interested in Kaplan-Meier estimator and Cox regression model and others as a potential way to understand membership renewals/churn.  Individual performance attendance forecasting.  Is there anyone out there interested in these subjects that wants to dive in to see if this type of modeling could be of help in better understanding and communicating.  I could use a partner or two in working through these ideas.

--Tom

Kyle Wilbert

I've been working on "walk-up" Attendance Forecasting based on Weather, school calendars, holiday calendar, a little economic data, marketing spend & exhibits.  I've discovered the simple decomposition time series forecasting methods like ARIMA / Facebook Prophet are not producing useful walk-in attendance models for me.   I believe Samuel Tran and an intern build a model that is primarily based on weather.  

If any of you are interested I would love to learn what you are working on regarding Museum "Walk-In" Attendance Forecasting and see how we might be able to work together to refine museum attendance modeling.   Let me know if you are interested. 

--Tom

It would certainly be nice to have a model like that. I believe the Art Institute of Chicago has a model that does something like this. Here's an interesting paper that explores the exhibit angle. I get the gist of it, but the math is beyond my statistical knowledge at the moment.

Kyle Wilbert

I don't have access to informs Publications online. :-( So have, as of yet not been able to read the article.  However the abstract is very interesting.  I've reached out to my wife with this citation (she is an academic reference librarian.) Hopefully, I'll be able to take a look.

The interesting finding described in the abstract for me is that "in addition to seasonality, exhibition type, display location, and life cycle affect attendance significantly." In the work that I've done with my museum, I can see an impact that the exhibitions seem to have on overall attendance.  However, the magnitude I'm finding in my work shows the impact of the exhibition as relatively small. 

I'm super interested in the fact they say "In this paper, we build a model to study and manage the impact of exhibitions on the number of museum visitors." I want to know more about this model that they are constructing.  What is their methodology?  Can I apply the methodology to the models that I have been building?  If so what impact does that have on my models?

Thanks for sharing.  Once I have had an opportunity to read the research, It would be great to have a conversation about the implications and impact of this study.