Dave Wentzel
linkedin.com/in/dwentzel
davew@microsoft.com
- Meeting ID: 973 8428 9784
- 11:00-12:15
- Why do data projects fail?
- Design Thinking/Rapid Prototyping: hopefully this link works.
After discussing with Professor Pursell, we are going to change topics/agenda. If you want to see the original content of this workshop just switch to the original branch in git. Just ask if you want the full workshop with solutions too.
Here are some topic ideas...feel free to vote or let Prof Pursell know if any of this is interesting. Common feedback I receive is that I have a LOT of ideas, but very few are good. That's a symptom of a good Design Thinker. I'm interested in your feedback.
| Topic Idea | Description | Vote | Order of Session |
|---|---|---|---|
| Why Data Projects Fail? | Data projects in Corporate America have a high fail rate. Why? And how can YOU change the culture. | Y | 1 |
| Design Thinking | This is an approach to understanding problems that is becoming popular in CorpAmerica. This works VERY well for data science projects. Here is a podcast with an overview | Y | 2 |
| Avoiding Analytical Mistakes | Data is hard. Numbers are hard. Make sure you don't make mistakes interpreting your data. I'll show you some methods that data scientists use to avoid making analytical mistakes. | Y | 3 |
| Make Your Data Tell A Story | Data projects are difficult. But if your data tells a story it's much easier to convey meaning. | Y | 4 |
| Data science for Anyone | This is probably too simplistic for your class. But commonly I get asked what is data science and how can we upskill employees to be Citizen Data Scientists. It's not very technical, it's just the thought processes that we use in business. | N | |
| MLOps in AzMLService | MLOps is very hot right now. Data scientists don't like to spend time doing repeatable things that can be automated. My intent was to show how to do this Week 4 with our CNN. Perhaps this is still valuable? I have a workshop where we take a VERY simplistic ML use case and develop it from "laptop" to "endpoint". This is surprisingly very hard to do without automation. I can likely give each student access to my AMLS workspace where I have everything working already for me but I've never tested if I can give guests access. If you don't want to do hands-on we could just demo it and students could do it on their own in their Azure subscription (it's in a gh repo) | Y | 5 |
| Prescriptive Analytics | "What do we do next" is THE thing that executives want in 2021. And very few folks understand how to do it. I've been asked by my employer to put together a demo of this based on work I've done for customers but haven't had time. I have some very simplistic demos. Frankly, this is so new (the concept has been around for 20 years but the implementation and tech has only recently made it feasible) that this might be a really good topic for a dissertation, a gh repo where you demo this to future employers, or even a book. | Y | 6 |
| Data Lakes | We can cover part of this in other sessions, or have a dedicated session to this. In CorpAmer everyone has a data lake but the consensus is it's not always showing value. Understanding DLs from a business perspective is critical | Y | 7 |
| Data Engineering/Databricks | Knowledge of Spark is in demand. I can show patterns that we use and I also have a multi-day workshop, but just understanding the basic patterns is likely enough. This wouldn't be "training"...it's more like "here's how we do it in the real world." | Y | 8 |
| Creating a Corporate AI strategy, culture, ethics | Occasionally I do these for executives. I'm not sure college students would appreciate this, but it is top-of-mind with execs who have tried and failed at AI. | N | |
| End-to-End Big Data with Databricks | This is really more of a hands-on workshop. About 16 hours to complete. We take a raw dataset and learn about it, we do regression on it. It's a few years old, totally databricks-focused, might not work in your azure subscriptions (we limit databricks in some subscription types). | N | |
| Feedback Loops and Rapid Prototyping | Two huge topics that most data scientists don't do well. What I see is data scientists that jump right to algorithm development. That skips a lot of interesting conversations about how we are going to present our data to the user. I like to walk through a RP that I built for a customer. When I show this to other customers their jaws drop, it is a very compelling demo, created in 3 days. But it's just that...a prototype. Much of it doesn't really work...but you wouldn't know that. That's a prototype. We then iterate over it and build it out over time, using DT as the guardrails to ensure we are building what our customers want. It deeply incorporates the feedback concepts. You don't need to be a data scientist to appreciate FLs and RPs...these are skills that are equally important to product developers, software engineers, marketing, manufacturing. | Y | 9 |