regression-js VS arquero

Modeling: We are currently exploring this space so our findings are not final yet but let me share what we've found so far. TensorFlow.js is absolutely amazing, it provides a native port from TensorFlow to JavaScript with CPU, WebGL, WebAssembly and NodeJS backends. We were able to write an LSTM to do time series prediction, so far so good. However, we started hitting issues when we started to do simpler models, like a linear regression. We tried Regression.js but we found it falls short, so we wrote our own script to handle single-variable regressions using TF.js and gradient decent. It actually worked quite well but exposed a flaw in this approach; basically, there is a lot of work to be done to bring many models into the web. We also found out Arquero.js does not play nicely with TF.js since well, Arquero.js does not support tensors; so we went on to explore Danfo.js, which integrates great with TF.js but we found out it's hard to scale it's transformations to +1M rows and found other rough edges. Since then, well, we started exploring Pyodide and perhaps contributing to Danfo.js, or perhaps involve more server-side compute, but that will be a story for another time.

In a similar vein is https://pbeshai.github.io/tidy/ which I've used for 3+ years. It's a really nice lightweight transformer.

I've also used https://github.com/uwdata/arquero once (better performance for large datasets).

Neat!

This is one of my favorite spaces, so I'll add some generic advice which may or may not be helpful.

I once had the privilege of working for Max Roser and Hannah Ritchie at Our World in Data, as one of the engineers on their Grapher library (https://github.com/owid/owid-grapher), and learned a ton from them (and others on the team) about making great charts.

My one piece of advice from looking at your examples would be: don't neglect title, subtitle, and caption! They would be so easy to do well because you've already created your "simple headers thingies". A few words go along way. Check out "Storytelling with Data" by Cole Nussbaumer Knaflic for a great read on the subject. Owid's Grapher does those the best, IMO (followed closely by DataWrapper.de -- but that's not open source).

At some point, if you keep up with this, you'll also want to add a dataflow library and DSL. Hadley Wickham's dplyr in R was the GOAT, and I copied that in my Ohayo tool and in OWID Grapher's CoreTable library (https://github.com/owid/owid-grapher/tree/master/packages/%4...). Jeffrey Heer's newish Arquero (https://idl.uw.edu/arquero/) library is also along those lines.

Lately I've delving into Mike Bostock's new thing Plot (https://observablehq.com/plot/). So far, excited by it, but only spent a day or two with it at this point.

I don't use emacs anymore, but hopefully something helpful in the comments above.

https://github.com/uwdata/arquero

Transformations: We found out that JavaScript in combination with D3.js has a pretty decent set of data transformation functions; however, it comes nowhere near to Pandas or dplyr. We found out about Tidy.js quite early, loved it, and adopted it. The combination of Tidy.js and D3.js and Plot.js is absolutely amazing for visualizations and data wrangling with small datasets, say 10-100K rows. We were very happy with this for a while; however, once you move away from visualizations into real-world data analysis, we found out 100K rows restrictive, which gets worse when having 100 or 1K columns. So we switched gears and started using Arquero.js, which happens to be columnar and enabled us to process +1M rows in the browser, descent size for real-world data analysis.

Take a look at the arquero library from a research group at University of Washington (the same group that D3 came out of). https://github.com/uwdata/arquero