5 ms·

I personally find the Python plotting landscape a mess and confusing. It always seems to be a lot harder than is necessary to do anything and make it look good

by joshlk 6y ago

I personally find the Python plotting landscape a mess and confusing. It always seems to be a lot harder than is necessary to do anything and make it look good.

zmmmmm 6y ago
yes ... I feel like everyone is stuck in a local minima (or maxima, depending on your optimisation direction!) of using matplotlib which is "good enough" but not very good. So there isn't great momentum to improve it, but nobody really likes it. My favorite plotting syntax is that exposed in BeakerX [1], but that's less common than everything else put together. [1] https://github.com/twosigma/beakerx https://github.com/twosigma/beakerx
captnswing 6y ago
I agree! I found this presentation invaluable in getting a good overview of the landscape and the different segments within: https://www.youtube.com/watch?v=FytuB8nFHPQ https://www.youtube.com/watch?v=FytuB8nFHPQ It helped me settle on https://altair-viz.github.io/ https://altair-viz.github.io/ (coming from matplotlib) and I never looked back
- pea 6y ago
  I find the big benefit of Altair is that the API is so nice and composable, and because vega provides a ton out of the box to build really high quality visualisations. For instance, if you take this (sample) report using Altair to plot Default Alive / Default Dead: https://datapane.com/leo/reports/startup_finance_report/ https://datapane.com/leo/reports/startup_finance_report/ - the interactive code is actually relatively small: https://gist.github.com/lanthias/5a41c1e4b21ae274ddb95cf5ad10b4ac https://gist.github.com/lanthias/5a41c1e4b21ae274ddb95cf5ad1... It's also great being able to add Altair shapes to Folium for geoplotting (as the vega geoplotting is a bit more low-level). What I really think is missing in the ecosystem is a "vega for tables", so you could be rich, interactive tables with a similar grammar. That would rock.
gmtx725 6y ago
yep- far too many libraries. Matplotlib, seaborn, bokeh, plotly, altair, ggplot2, vega, now this. And each library covers about 90% of your use cases on its own but you're constantly having to switch to another one for the remaining 10%. Nightmare.

by joshlk 6y ago

I personally find the Python plotting landscape a mess and confusing. It always seems to be a lot harder than is necessary to do anything and make it look good.

zmmmmm 6y ago

yes ... I feel like everyone is stuck in a local minima (or maxima, depending on your optimisation direction!) of using matplotlib which is "good enough" but not very good. So there isn't great momentum to improve it, but nobody really likes it. My favorite plotting syntax is that exposed in BeakerX [1], but that's less common than everything else put together. [1] https://github.com/twosigma/beakerx https://github.com/twosigma/beakerx

captnswing 6y ago

I agree! I found this presentation invaluable in getting a good overview of the landscape and the different segments within: https://www.youtube.com/watch?v=FytuB8nFHPQ https://www.youtube.com/watch?v=FytuB8nFHPQ It helped me settle on https://altair-viz.github.io/ https://altair-viz.github.io/ (coming from matplotlib) and I never looked back

pea 6y ago
I find the big benefit of Altair is that the API is so nice and composable, and because vega provides a ton out of the box to build really high quality visualisations. For instance, if you take this (sample) report using Altair to plot Default Alive / Default Dead: https://datapane.com/leo/reports/startup_finance_report/ https://datapane.com/leo/reports/startup_finance_report/ - the interactive code is actually relatively small: https://gist.github.com/lanthias/5a41c1e4b21ae274ddb95cf5ad10b4ac https://gist.github.com/lanthias/5a41c1e4b21ae274ddb95cf5ad1... It's also great being able to add Altair shapes to Folium for geoplotting (as the vega geoplotting is a bit more low-level). What I really think is missing in the ecosystem is a "vega for tables", so you could be rich, interactive tables with a similar grammar. That would rock.

gmtx725 6y ago

yep- far too many libraries. Matplotlib, seaborn, bokeh, plotly, altair, ggplot2, vega, now this. And each library covers about 90% of your use cases on its own but you're constantly having to switch to another one for the remaining 10%. Nightmare.