XLKitLearn

I wrote XLKitLearn to give students access to the full power of scikitlearn using an Excel interface. Students use it to fit random forests, boosted trees, and carry out Latent Dirichlet Allocation on large datasets, all in Excel. It has changed the way I teach data science and analytics, in my Business Analytics 2 class.

Why design a brand new tool when there are others out there? See below for my rationale. Ready to give it a try? See the installation instructions and quick demos below.

Here are the highlights

• Ability to fit a number of linear and tree-based predictive analytic models, as well as basic text analytic capability (encoding, and Latent Dirichlet Allocation)
• Each run of the add-in produces a full report
• Every run outputs a piece of python code that carries out the same analysis. The code is dynamically-generated for every run to use the simplest sklearn function for the specific model being fit
• Works natively on Windows and Mac computers - no need for any parallels environment of any kind

You are welcome to use XLKitLearn for free for any purpose, but every run will be logged, together with your email address, on a server. If you want to avoid this log, either reach out to me or simply run it while disconnected from the internet. You will need to run it while connected to the internet at least once every 10 days to check for updates.

Installing the add-in

Before you install the add-in, download the installation instructions and the add-in Excel file. If you are installing the add-in on a PC, you will also need this zip file.

Two quick demos

Here are two short videos that demonstrate the basics of the predictive analytics and text analytics capabilities of XLKitLearn. A more detailed user manual is forthcoming!

Why design a brand new tool?

Before designing XLKitLearn, I did a broad search to see what other approaches existed to teach non-technical students data science. I found three approaches, but none of them met my needs exactly, hence my decision to create something new.

• Other Excel Add-ins. A number of Excel add-ins that already exist; they function similarly to XLKitLearn, often with far more advanced functionality. I had two main concerns with using them, however, in order of importance (1) as far as I know, none of these add-ins are based on any widely-available ML libraries, and they do not make their code publicly available. This makes them very limiting - it's hard to know exactly what they're doing under the hood, and there's no way to go beyond the functionality they offer. The whole idea of my class is to provide my students the tool to communicate with data science teams in their respective companies - in that respect, it was essential for me to have a tool that would link directly back to scikitlearn (2) these add-ins typically only work natively on a PC, not on a Mac - an increasing number of my students use Mac computers, and find it cumbersome to install a parallels environment.
• No code or hands-on work. A perfectly respectable approach is simply to eschew all hands-on work, and simply present students with the results of the algorithms for discussion. I think this is a perfectly respectable approach, but I wanted my class to include the excitement - and experience - of actually working with data.
• Copy-pasted Code Segments. Some classes I've seen take a 'code snippet' approach - they provide bits of Python or R code, and get students to copy and paste them. I'm not a fan of this method, and I think it the worst of all worlds. Having to worry about syntax provides less guidance to students, and make it harder to focus on the underlying principles. It also gives students a false sense of confidence - being able to paste a piece of R code that does one specific task on one specific dataset does not a data scientist make. With the Excel approach, it is at least clear what is and isn't being taught, and students can go in-depth on the principles. A final concern is that for certain complex, highly-tuned models, the code can get pretty hairy.