Super cool. This is the opposite angle from the approach we’re taking at Mito [1].
Instead of extending Excel to support more advanced functionality, we’re extending Python environments to support more basic spreadsheet functionality. So: basic Python users doing data science can use a point and click environment that they are more familiar with.
Interesting difference in user base as a result of this opposite approach. I also don’t envy you building an Excel add-in… we tried that first and didn’t love it :-)
Great nonetheless, and not to take anything away from the project, but I guess what I was expecting from the title, however ambitious, was an Excel-native solution.
I've heard legends of people fitting shallow feedforward NNs with pure Excel formulas.
I once interned for an economist who was using recursive Excel formulas for some kind of iterative optimization algorithm. I don't remember what it was, probably some iterative least squares thing.
I also was a research assistant for an economist who fit linear regressions with hand-written routines in Matlab. For some reason they felt more comfortable using Matlab's QR decomposition instead of just using the built-in linear modeling tools in R or even Stata.
Some people just want to use their favorite tool even if it makes no sense to do so.
You think that’s funny, but from experience, I’d tell you this happens more often than you think.
A company I was with, that was a startup and then IPO’d, our pipeline involved an open office spreadsheet “deployed” (copy) onto the server, where then the Java webapp (war+tomcat) would read the inputs to calculate for millions of people, user classifier and what kind of robocall they should get in case of a utility event to prevent brownouts.
I would be curious what deployment would look like here…
Would it be uploading the model back to the cloud to run inference for other users - like a marketplace?
Perhaps Microsoft can add it to their certs list and start competing with the tier 1 grad schools.
I remember a newly hired ex-consultant have a slide in her deck about using ML across sparse/inaccurate company data and how this would turn “data into information.” I think eventually they realized that it takes more than PowerPoints and repeating that phrase.
Just a question about these type of apps/plugins, when it says something like this: "Add-in capabilities When this add-in is used, it Can read and make changes to your document Can send data over the Internet" is that isolated to just the pane you are currently on with the plugin?
Yes, the data is only read from the active worksheet. And your data is not sent over the internet. The trained model is stored on a server, but your data stays on your machine.
The spreadsheet is a killer app that nobody has yet figured out how it fits in the new era. Godspeed with your venture but keep in mind that scikit-learn has been built by an open source community.
Instead of extending Excel to support more advanced functionality, we’re extending Python environments to support more basic spreadsheet functionality. So: basic Python users doing data science can use a point and click environment that they are more familiar with.
Interesting difference in user base as a result of this opposite approach. I also don’t envy you building an Excel add-in… we tried that first and didn’t love it :-)
[1] https://trymito.io/hn