Plot Tensorflow's Summary event in a beautiful way (using seaborn actually) instead of using Tensorboard.
cd tensorboard2seaborn
python setup.py install
Example: tensorboard2seaborn --logdir=/experiments/logs --smoothing=0.7
usage: tensorboard2seaborn [-h] --logdir LOGDIR [--smoothing SMOOTHING]
[--no_title] [--no_legend] [--no_axis_labels]
optional arguments:
-h, --help show this help message and exit
--logdir LOGDIR Path to event files
--smoothing SMOOTHING
smoothing scale .It should be in [0,1] (default: 0.6)
--no_title Doesn't show title in any of the graphs
--no_legend Doesn't show legend in any of the graphs
--no_axis_labels Doesn't show axis labels in any of the graphs
import tensorboard2seaborn
tensorboard2seaborn.plot(logdir='/experiments/logs')
We expect following structure of the tensflow log events.
βββ results
βββ exp2
β βββ events.out.tfevents....0
βββ exp1
β βββ events.out.tfevents....0
βββ ....
Following structure is generated comprising of seaborn graphs
βββ results
βββ seaborn
βββ group1
β βββ plot1.png
β βββ plot2.png
β βββ ...
βββ group2
β βββ plot1.png
β βββ plot2.png
β βββ ...
Before | After |
---|---|