We first chose to display only data points obtained after 20 or more epochs of training. Then, by slicing through the “loss” axis, we observed that larger learning rates led to better performance (perplexity). You can reproduce this example here:
These are known as https://en.wikipedia.org/wiki/Parallel_coordinates and are a pretty old technique. Although the wiki lists process control as one of the major users, their history has them as dormant until revived in their modern form in the 1960s, which doesn't quite match your timeline. Perhaps the discrepancy is resolved by whether parallel axes were used as a equation solving aid versus a visualization aid?
Don't know if it's what you're referring to, but psychrometric charts are always fun, and any of the thermodynamics charts (Pressure-Enthalpy being an example).
The best part? You say the "age of slide rules", but these are still used today.
Actually yeah, I wrote a script which 1. exports data from tensorboard.dev, then 2. prints it out in csv form that can be dropped into hiplot.
It was ok. Kind of lost interest when I saw the actual result. You can't smooth the data, so often times it's hard to tell what's going on.
Here's the awful script. It's specific to my own naming conventions, so it probably won't work for you. But the same idea would work: just parse the tensorboard data and spit it out as csv. https://gist.github.com/shawwn/b74d6e58da6496e2ade02bab61acc...
I also ran into a bug where multiple different datapoint types were getting merged into one. (It wasn't due to the csv code; the actual CSVs are fine.) I.e. the rows in the csv had a certain column that had a certain value, but in hiplot that column was nowhere to be found.
Edit: Just saw your name. Holy shit, it's Mike Bostock.
On a side tangent, I find it crazy that the New York Times for a long period had yourself, Jeremy Ashkenas, and Rich Harris all on staff. What an all-star team.
keanzu|6 years ago
https://facebookresearch.github.io/hiplot/_static/demo/ml1.c...
zetaben|6 years ago
https://doc.ubuntu-fr.org/picviz (Sorry I could only find it in French). Seems defunct nowadays.
rckoepke|6 years ago
Cybiote|6 years ago
They're also not outdated, you can find them in many libraries, like highcharts here: https://www.highcharts.com/docs/chart-and-series-types/paral...
Excellent technical coverage is: Parallel Coordinates for Visualizing Multi-Dimensional Geometry, from 1987 at https://link.springer.com/chapter/10.1007/978-4-431-68057-4_...
Wikipedia links to this 1880 chart: https://www.davidrumsey.com/luna/servlet/detail/RUMSEY~8~1~3...
chemeng|6 years ago
juliwu|6 years ago
sillysaurusx|6 years ago
It was ok. Kind of lost interest when I saw the actual result. You can't smooth the data, so often times it's hard to tell what's going on.
Here's the awful script. It's specific to my own naming conventions, so it probably won't work for you. But the same idea would work: just parse the tensorboard data and spit it out as csv. https://gist.github.com/shawwn/b74d6e58da6496e2ade02bab61acc...
I also ran into a bug where multiple different datapoint types were getting merged into one. (It wasn't due to the csv code; the actual CSVs are fine.) I.e. the rows in the csv had a certain column that had a certain value, but in hiplot that column was nowhere to be found.
Here's what it looks like: https://imgur.com/1n1eDXu
I deployed an example here: https://hiplot-subsim-demo.now.sh
Notice that the "run" column shows up in the data, but not in hiplot.
mbostock|6 years ago
http://bl.ocks.org/syntagmatic/3150059
It’s a shame Kai isn’t created in the README, LICENSE, or announcement.
gavinray|6 years ago
[0] https://bl.ocks.org/jasondavies/1341281
[1] https://observablehq.com/@d3/parallel-coordinates
Edit: Just saw your name. Holy shit, it's Mike Bostock.
On a side tangent, I find it crazy that the New York Times for a long period had yourself, Jeremy Ashkenas, and Rich Harris all on staff. What an all-star team.
snippyhollow|6 years ago
th0pe|6 years ago