Hacker News new | past | comments | ask | show | jobs | submit login
Introducing Chartify: Easier chart creation in Python for data scientists (spotify.com)
57 points by homarp on Nov 16, 2018 | hide | past | favorite | 12 comments

Minor gripe - These sort of tools always deride matplotlib a bit, but often show outright incorrect examples of using it.

Admittedly, the "right" way of doing it is quite clunky as well:

The example shown teaches very bad habits. It doesn't change the tick label formatting at all. It makes the ticks and labels static. There's a big difference, and it's obvious as soon as you interact with the plot (i.e. zoom/pan around - you won't get new ticks/labels).

I agree that matplotlib isn't the right choice for web-based visualization, but let's not pretend it makes static plots. It's a very, very highly interactive desktop visualization package, and oriented towards building desktop applications that can produce publication-quality plots.

All that having been said, this looks pretty neat.

As someone who uses matplotlib to make plots pretty frequently (but also struggles getting it to do the right thing often, most likely due to accumulation of bad habits), can you recommend a good resource to learn the good habits? It seems like there are n different ways to do everything - I'd love to learn to work more efficiently.

Unless you're making high quality prints I would recommend staying away from matplotlib. That's it's purpose and if you ever use large amounts of data it slows to a crawl. It's a real shame that it's the default plotter for pandas. I imagine I've wasted a full 24 hours in total for plots to show up over the past few years.

Built on top of Bokeh, so if you do need more control you can always fall back on Bokeh's API.


APL2 licensed.

no relation with https://chartify.github.io/chartify/

I have not used a ton of Bokeh, but why would I want to learn yet another plotting library?

Add this to the end of a list of competing graphics libraries including ggplot2, seaborn, base matplotlib, etc...

Bokeh interactive tools are much nicer than matplotlib and its children. Also it has a js backend so you can more easily integrate into a website.

Finding Seaborn [1] after toiling for hours with Matplotlib for some convoluted plots was a godsend. Of course the moment you wanted anything off the beaten path you'd still have to fall back to Matplotlib API but Seaborn would provide an excellent starting point.

All in all I'm really happy to see high-level wrappers tailored to data-scientists.

[1]: https://seaborn.pydata.org/

I did some digging and am still somewhat confused. What does spotify have to do with these types of projects? Seems like it has nothing to do with their music streaming business.

Spotify gets heaps of data from users so I'm sure they do a bunch of data-science stuff in the background. For example their `Discover Weekly` feature is most likely powered by such analytics and possibly some ML.

I guess this is occupying the same space as Holoviews and Plotly (which are both bokeh wrappers, among other things).

Plotly isn't a bokeh wrapper. I believe it sits on top of D3.

Ah, my bad. I jumped to conclusions because HoloViews has both bokeh and plotly backends.

Guidelines | FAQ | Support | API | Security | Lists | Bookmarklet | Legal | Apply to YC | Contact