, Kyle Wilbert, Maery Simmons (Past Member), Heath Wilder, , Brian Parker (Past Member), (et. al.)
What have you been experimenting with when it comes to Machine Learning (ML)?
Right now I'm working on building a prototype production environment for our first ML models. (This is in contrast to the development environment on my laptop.) The first model for me will be making our attendance model production. Looking at short-term (0-10 day), medium-term (1 month) and long-term(18 months) forecast of "walk-in" attendance.
The prototype production environment will be on a small Linux based system running Dataiku's DSS. This is providing me a production infrastructure for the usual data science tools, Python, R notebooks, mostly scikit learn based models, some nice visual ETL tools and additional visualization capabilities.
My plans are then to work with members of the INFORM The Institute for Operations Research and the Management Sciences. (https://connect.informs.org/probonoanalytics/home) to fine-tune and validate modeling approach. So, we can know how confident we can be in such a model.
I'm curious what you may be doing? And are their anythings that analytic Coffee! members can do to help one another to move ML forward withing the Cultural Non-Profit Community?
--Tom
I haven't been doing much yet except starting to build my own understanding of Machine Learning! I'm eager to hear more about what everyone else is doing, as I think ML is going to be a really great tool for the non-profit community.
Hi Tom,
That sounds exciting!
I'm in more of a learning phase - still working through some Data Science courses on DataQuest. Next, will likely be the text Introduction to Statistical Learning. Work-wise I will continue to do one-off ML projects as I see a need for them. Definitely not looking to put anything into production. Mostly small-scale experiments.
Thanks for asking!
Kyle
I've got a pretty simplistic sales forecasting model running in Microsoft Power BI at the moment (via the REST API).
It basically creates a model of our average ticket sales trend over time for a single concert, and uses it to
This then informs my digital marketing strategy. Definitely paints and over-simplified picture, but a helpful starting point.
-O
I'm becoming a big fan of these posts; keep it up, Tom!
I haven't had the luxury of getting to run any sort of analysis since I began in April, as we've been in implementation and our current database is very, very messy. Once I get it cleaned up by Q2 2020 (fingers crossed) I plan to run a lot of analyses based around our demographics in the context of outreach and program participation. Clustering algorithms will probably be at the forefront of my ML endeavors.
As far as how this community can help one another, I think the area that we can be most helpful with aiding in understanding what tools work best for what types of analysis.
Looking forward to seeing you all pioneering ML methods for your organizations!
Currently, not planning anything but I've been exploring ways it might be useful to us. I'm excited to hear what other folks have planned.
I pln on eventually having enough time to get into some Power BI. With the new shift in my role I should be able to get that happening by EOY. Sales forecasting is my main aim.
Heath Wilder
When it comes to Sales Forecasting how are you hoping that Power BI will help?
They are separate thoughts actually.
I have patterns I want to challenge including Timing of onsale relative to internal (current show), external (local arts co's subs launch), and static (Xmas, school hols, EOFY) events.
pretty standard stuff really
Heath Wilder as always you are pushing into new ground.
I thought that you might be trying to put both of those things together, and was wondering if you had found something new and awesome that Microsoft had added to Power BI to help with the challenge. I've really liked Power BI since it came out as Power Query and Power Pivot in MS Excel. However, It would not have been the tool that I would have personally grabbed for when trying to do something predictive. I know that one can add R (in particular Microsoft's variant of R in Power BI.) And R can definitely be used for a variety of forecasting. And Microsoft is clearly trying to keep Power BI relevant in the AI / ML space. Over the last year though I've not been able to keep up on all of those MS developments. Here is a recent video I found on the subject. The Key influencer part of the video might be relevant to your sales timing question.
On the forecasting side, it sounds like you have an intuition that sale timing will make a difference to sales volume. Do you have a specific hypothesis you want to test? Have you thought of a method to test or evaluate the hypothesis? I'd love to hear more about what you are trying to do when you are ready to share. It's my hope that analytic Coffee becomes the place for folks in the Tessitura Community to test and share such ideas! I'm really looking forward to hearing more.
Thank you sir. You are a prince.
I'll have a bit of think and get back to you