scilog: record-keeping for computational experiments

Author

Soeren Wolfers <soeren.wolfers@gmail.com>

Organization

King Abdullah University of Science and Technology (KAUST)

This package provides a command line tool that executes numerical experiments and stores results along with auxiliary information about inputs, source code, hardware, installed packages, runtime, memory usage.

Getting started

In its easiest form, scilog 'foo.bar(x=1,y=2)', runs the function Python function bar in the module foo.

Since most computational research requires series of experiments, scilog supports such series through the specification of variable ranges. Using scilog 'foo.bar(x=var(1,2),y=2)' does the same as above, but foo.bar is run once with x=1 and once with x=2.

Scilog can also be used for numerical experiments that are not based on Python. Using scilog --variables p=[0,1] 'my_tool {p}' the command line tool my_tool is run twice, with arguments 1 and 2, respectively.

Information on previous scilog entries can be displayed using scilog --show 'my_tool' or by simple navigating scilog's directory hierarchy, where all entries are stored in the most human-readable form possible.

Currently running experiments can be deattached from and will continue in the background. They can be listed and reattached to with scilog -ls and scilog -r, respectively.

Installation

Run pip install scilog

Requirements

• Python 3.6+
• Unix-like operating system
• GNU screen