Post-processing

The optzer writes some messages to stdout and we recommend it redirected to out.optzer as,

optzer --nproc 4 --random-seed 42 | tee out.optzer

In addition to the stdout, optzer writes db.optzer.json file which contains all the trial parameters and their loss values.

The db.optzer.json can be loaded using pandas.read_json as,

import pandas as pd
db = pd.read_json('db.optzer.json',orient='records',lines=True)
print(db)

The options orient and lines are required to load it successfully.

You can convert the db.optzer.json to CSV format using the script optzer-db2csv as,

optzer-db2csv db.optzer.json out.csv

Then you can plot it with gnuplot, for example,

gnuplot
gnuplot> p 'out.csv' us 2:3 w p
gnuplot> p 'out.csv' us 4:3 w p pt 5, '' us 5:3 w p pt 7

The script optzer-out2md will extract loss information from out.optzer and tidy it up to markdown format.

optzer-out2md out.optzer

This will gives you the following table in markdown format.

|  Target  |  Weight |  Loss  |
|----------|---------|--------|
|  simple  |  1.000  |  0.0235  |
|  total  |  -----  |  0.0235  |

This table will be shown as below if the markdown is processed correctly…

To get the best data calculated during the optzer, the optzer-get-best script extracts and copy them to best_data directory as,

optzer-get-best
ls best_data/

The best parameters are written in in.vars.optzer.best file. And you can extract and map them into the parameter files that can be read from external programs as,

mkdir 00_test
cd 00_test
optzer-out2prms ../in.vars.optzer.best ../in.params.simple
or
optzer-db2prms ../db.optzer.json ../in.params.simple

• Above example is for the case of optzer/examples/01_simple_example. One can specify several in.params.XXX files that contain placeholders for the parameters.
• This script should be performed different directory from the working directory of the optzer to avoid replacing in.params.XXX files. (The script will raise an Error in such a case.)