Matplotlib - scaling of the y-axis

Mri-monitoring / Mri-python-client

Python client for Mri server, supports easy plotting/updating

Apache License 2.0

4 stars 1 forks source link

Matplotlib - scaling of the y-axis #3

Open tf79 opened 9 years ago

tf79 commented 9 years ago

y-axis scales automatically but, because the loss is very high at the beginning you can't really figure out to which value the accuracy is converging. A two y-axis plot should fix this.
I noticed that in this config placing the legend in the lower left corner occludes the early iteration phase. If you change to a two axis plot, put some offset to scale the accuracy axis, that when the legend is placed in the upper right corner nothing gets occluded.

nharada1 commented 9 years ago

This one is a little tricky. The issue is that it's possible to have more than just two things showing in the plots (i.e. you could show learning rate, train error, test loss, etc), and so getting axes for each one could be hard.

I'll have to think about this one. Options are:

Just go from 0 to 1 like before
Log scaling
Dynamically re-scale all lines so that they go from [0 1]. This would make it really difficult to see the values though
Use two scales, one for [0 1] and one for everything else. Accept an argument stating which variables are percentage scale and which are real value. This is probably the best option, but may require some architecture weirdness.

tf79 commented 9 years ago

Last options sounds good. Might it be easier if the user would specify the output they want to see, as well as the scale? Not specified outputs could be ignored.

tf79 commented 9 years ago

Could the location of the legend be moved out of the plot?

This regarding to plotting so I hope it's fine to put it here: showing the max/peak and min values could help to compare plots with each other.

I have a thousands of plots with different scaling and resolution of the image plots. Though it's very hard to compare them when it comes down to +-3 % or so.

nharada1 commented 9 years ago

I've moved the legend to the bottom of the plot as well as added min and maxes in all the plots.