Even in the Era AI, GGPlot's API is still the best charting API. The name "Grammar of Graphics" isn't just marketing, they literally sought to write a god damn grammar to was capable of expressing all possible qualitative graphics.
I actually stumbled upon this book when I was trying to look up how draftsmen (with pens and pencils on paper) did qualitative graphics as I found they had a lot of charm as opposed to modern charting libraries. It's something I noticed when looking through a bunch of historical RBA (Reserve bank of Australia) annual reports, the 1960-1980 charts had a lot of character, but then you go into the early 2000s and its a stale chart from excel.
Anyways ggplot doesn't really recapture the magic of those older charts, but it seems use quite a few of those as a baseline for how to communicate information. Like in figure 20.1 they talk about efforts to replicate older inforgraphics that showed Napoleon’s March on Russia, this graphic here (I think the example in the book is a bit nicer than the one in this blogpost IMO)
On top of the charts just look nicer than anything you could produce with pyplot (and any API built on top of it) as pyplot seems to be have some really limited raster based rendering or something and the text handling is incredibly limited, I've never had this issue in ggplot.
I feel like most software engineers aren't exposed to because it exists in the R ecosystem which is more so data scientist, econometricians, statisticians and other quantitative data professions, but it definitely one of the nicer APIs and I wish more people in the node and python ecosystem copied their homework. I see vega's full name is something to do with grammars, but idk it's for the same reason.
Yeah Grammar of Graphics semantics are already very ideal for both humans and agents alike. Ggsql is a recent project that implements it as SQL user defined functions which is even more agent friendly
It’s incomplete but it show cases some of what I mean. There’s just a lot more charm in these earlier graphics. I’m fairly certain they had to consider how the different colours of ink bleed on one another.
Like take 1978 for example, this might be some kind of highlighter ink, and when the colours over lap the grid they interact in ways most overlapping geometry don’t in charting libraries (but in CSS you can get a similar effect with mix-blend-mode: multiply)
If you go to page 159 of the most recent RBA report, you’ll see otherwise clean clear charts that do a good job of communicating the relevant information (perhaps that’s all there should do from such an institution), but it’s admittedly soulless compared to the charts from previous decades
not the parent, but one thing i find charming is that a lot of them have a really good feel for line weight and inking style, and vary it interestingly in various parts of the plot. computer generated charts tend to have a much more uniform line weight.
I have tried using Flint vs asking the AI to generate a Vega lite spec directly, and in my opinion Flint was not as nice of a solution.
Flint is fine for doing predetermined chat types, with very low customization. But I found using an agent or sub agent to create the Vega spec directly allowed for a lot more flexibility, and ultimately that means higher quality visualizations (stuff like adding points for min and max on a timeseries, or adding a callout marker for a date where some event happened).
That being said, with Vega lite you have to validate your chart specs, provide specific guidance, and play whack a mole with Vega bugs/idiosyncrasies. So Flint is more reliable if you don’t want to dedicate a whole skill to making charts and want to get running quickly.
Different charting backends support different kinds of charts. Being able to easily switch between "ECharts Sunburst" and "Vega-Lite faceted bar" - one of the examples on the project website - seems like a super useful ability.
Fair. How is this specific to the marketed AI integration? This could just be a unifying abstraction layer for a bunch of, in my opinion, verbose chart APIs. And then the LLM could just be the abstraction layer.
Yeah I was going to share it too. I watched a yt video yesterday where Tanner talks about how (in contrast to trad component-based libs) it takes a very different approach to composition, letting D3 do the heavy lifting and using its visual grammar, and staying grounded in putting the data (not the component tree) first.
What is the point of this. I already can get an llm to draw charts using plotly, matplotlib, echarts, etc all the time. There will be always a better way but what does this buy us?
This needs a side-by-side config comparison with something like echarts config schema format. I really don't see the point.
I'm 99% sure the verbosity required in the system prompt to teach non-M$ models this new ever-so-slightly-different-but-not-obviously-necessary chart def abstraction format, and the iterations required to get it right, will outweigh any supposed efficiency gains resulting from using it.
It's cool...but is it needed? I'm thinking of Apache echarts and any of many other mature charting libraries. Kinda seems like reinventing the wheel; thing gets released, thing gets more and more new features bolted onto it, eventually someone offers another thing that's basically the original thing with some slightly different design and syntax opinions...
Switch backends to use their native strengths: ECharts for hierarchical sunbursts, Plotly for statistical and analytical traces, or Excel for editable charts embedded in a workbook.
Or just find a charting library that you like and actually get to know what it can do, vs mixing and matching presets from different libraries but never tweaking them.
This is just json right? One issue with llms right now it's that if you give them a json specification, they're not always amazing at following it. I think it makes sense to have an agent tool that takes llm json, a file name and a spec path and only writes out the file if it conforms.
Every time I needed to do charts at work I had a conundrum. Libraries are aplenty but basically only plotly does everything, but it’s not the prettiest. I’d much rather developers contributed new features into existing libraries.
You can use skills or harnesses to have them validate their json (using plain old JSON validation or schemas) and they seem to have a very high success rate with that gate in place. It's slightly more tokens, but it's reliable
They even wrote a book about how they went about it (not that it speaks to the quality of the API) https://link.springer.com/book/10.1007/0-387-28695-0
I actually stumbled upon this book when I was trying to look up how draftsmen (with pens and pencils on paper) did qualitative graphics as I found they had a lot of charm as opposed to modern charting libraries. It's something I noticed when looking through a bunch of historical RBA (Reserve bank of Australia) annual reports, the 1960-1980 charts had a lot of character, but then you go into the early 2000s and its a stale chart from excel.
Anyways ggplot doesn't really recapture the magic of those older charts, but it seems use quite a few of those as a baseline for how to communicate information. Like in figure 20.1 they talk about efforts to replicate older inforgraphics that showed Napoleon’s March on Russia, this graphic here (I think the example in the book is a bit nicer than the one in this blogpost IMO)
https://www.andrewheiss.com/blog/2017/08/10/exploring-minard...
On top of the charts just look nicer than anything you could produce with pyplot (and any API built on top of it) as pyplot seems to be have some really limited raster based rendering or something and the text handling is incredibly limited, I've never had this issue in ggplot.
I feel like most software engineers aren't exposed to because it exists in the R ecosystem which is more so data scientist, econometricians, statisticians and other quantitative data professions, but it definitely one of the nicer APIs and I wish more people in the node and python ecosystem copied their homework. I see vega's full name is something to do with grammars, but idk it's for the same reason.
reply