ENH: Expose symlog scaling in plotting API

[mdekstrand]

The default log-scaled axes, activated by the logx, logy, and loglog methods to the Pandas plotting API, do the straightforward thing and take the log of 0 values. It then attempt to plot with these infinite logs, and makes the entire plot unusable without warning in the presence of 0s.

For example:

draws = pd.DataFrame({'freq': np.random.zipf(1.7, 1000) - 1})
draws['rank'] = (-draws['freq']).rank()
draws.plot(x='rank', y='freq', kind='scatter', loglog=True)

Matplotlib provides another scale, the symlog scale, that makes a small region near 0 linear to avoid these problems. For quick-and-dirty 'look at my data on a log axis' plotting, symlog is significantly more useful.

I can access it like this:

draws = pd.DataFrame({'freq': np.random.zipf(1.7, 1000) - 1})
draws['rank'] = (-draws['freq']).rank()
p = draws.plot(x='rank', y='freq', kind='scatter', loglog=True)
p.set_xscale('symlog')
p.set_yscale('symlog')
p

Either making the symlog scale the default log scale for plotting, or supporting a loglog='sym' option, would make it significantly easier to do quick data inspection with Pandas' convenience plotting.

[TomAugspurger]

We probably wouldn't make it the default, unless matplotlib did as well. A PR allowing 'sym' for logx, logy, and loglog seems fine.

…

On Mon, Jan 21, 2019 at 2:04 PM Michael Ekstrand ***@***.***> wrote: The default log-scaled axes, activated by the logx, logy, and loglog methods to the Pandas plotting API, do the straightforward thing and take the log of 0 values. It then attempt to plot with these infinite logs, and makes the entire plot unusable without warning in the presence of 0s. Matplotlib provides another scale, the symlog scale, that makes a small region near 0 linear to avoid these problems. For quick-and-dirty 'look at my data on a log axis' plotting, symlog is significantly more useful. I can access it like this: `` dfp = nh_stat_df.plot(kind='scatter', x='nhelpful', y='count') dfp.set_xscale('symlog') dfp.set_yscale('symlog') dfp Either making the `symlog` scale the default log scale for plotting, or supporting a `loglog='sym'` option, would make it significantly easier to do quick data inspection with Pandas' convenience plotting. — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#24867>, or mute the thread <https://github.com/notifications/unsubscribe-auth/ABQHIo11veYq_-C7dTx_0QVZHArxnFggks5vFh1GgaJpZM4aLc9p> .

[TomAugspurger]

@mdekstrand can you edit your post to include a minimal example? http://matthewrocklin.com/blog/work/2018/02/28/minimal-bug-reports. That way it'll be obvious when this can be closed.

[mdekstrand]

@TomAugspurger done!