Add additional minor labels in log axis with a span less than two decades

[zblz]

For plots with log axis that have limits of the order or less than two decades, only one label is shown:

Which does not give a very good idea of the values of the x axis. Adding labels to the minor ticks results in a mess, including non-helpful labels such as 10^{4.30}:

ax.xaxis.set_minor_formatter(ticker.LogFormatterMathtext(labelOnlyBase=False))

And limiting them at [2,5]*base**i results in losing all other minor ticks:

ax.xaxis.set_minor_locator(ticker.LogLocator(subs=[2,5]))

The decimals in the exponent can be resolved using LogFormatter, but it is not a nice representation for axis and requires changing the major formatter to match:

ax.xaxis.set_minor_formatter(ticker.LogFormatter(labelOnlyBase=False))
ax.xaxis.set_major_formatter(ticker.LogFormatter())

Finally, the only way to get them in a consistent manner with the major tick labels, while preserving all minor ticks, is to set them manually, which is far from ideal:

ax.xaxis.set_ticklabels([r'$2\times10^4$','','',r'$5\times10^4$','','','','', 
        r'$2\times 10^4$','','',r'$5\times10^4$','','','','',
        r'$2\times 10^5$','','',r'$5\times10^5$','','','','',
        ],minor=True)

I think that there should be a LogFormatterMathtext with a subs options like the locator, and an option to factors instead of decimal exponents. I will take a look at the LogFormatterMathtext and try to modify it in the following days. If it looks good, I would propose to switch to it automatically for axis spanning less than two decades.

Note that this issue has been reported in a stackoverflow question where a custom formatter is proposed as an answer.
This is somewhat related to the issue #4730 because the lack of labels is particularly noticeable when the axis is around 0.1,1,10, where it would make a lot of sense to have labels at [0.1,0.2,2,5,20,50].

[tacaswell]

I am somewhat ambivalent on this as I am not convinced of the utility of log scaled axis with less than 2 decades of data.

[zblz]

In my field (high energy astrophysics) they are extremely common: the
gamma-ray photon spectra of an astrophysical source will almost always
follow a powerlaw behaviour, and the energy detection band for a single
instrument is rarely larger than two decades (often less than one), so the
plotted spectra from a single instrument are always like this. It might be
a corner case, but I brought it up as it comes up very often in my use of
matplotlib.

This would be useful for the gammapy and
naima packages, which deal with gamma-ray spectra
and use matplotlib for plotting.

2015-08-05 14:46 GMT+01:00 Thomas A Caswell notifications@github.com:

I am somewhat ambivalent on this as I am not convinced of the utility of
log scaled axis with less than 2 decades of data.

—
Reply to this email directly or view it on GitHub
#4867 (comment)
.

[tacaswell]

Fair enough. I come from a soft-matter physics background where we are looking for power laws in noisy data, with less than 2 decades just about any powerlaw can be made to fit, hence my skepticism.

[efiring]

Another case where it would be nice to have labels for a short span of
a log scale is when one zooms in on a feature in a spectrum.

On 8/5/15, Thomas A Caswell notifications@github.com wrote:

Fair enough. I come from a soft-matter physics background where we are
looking for power laws in noisy data, with less than 2 decades just about
any powerlaw can be made to fit, hence my skepticism.

---

Reply to this email directly or view it on GitHub:
#4867 (comment)

[QuLogic]

Is this fixed by the above PR?

[efiring]

It was, but that was reverted by #7000, and then a modified version was restored by #7419.