WARNING: This was an experiment. The answer is in the negative. Matching resource utilization over increasing input sizes is not possible in this way. This tool should not be used for serious purposes.

This module provides a CLI tool for approximate lower bounds according to a catalog of prescribed growth rates.

The tool currently features most well-known classes of growth rates and works by fitting each generic growth rate function to the given data.

Its primary use case is estimating time or space growth rates of various algorithms, but it can also be used on data coming from physical measurements.

To install from pypi:

pip3 install --user matchgrowth

The CLI tool expects --infile with a valid CSV file path, and --col1 and --col2 each with a column name present in the CSV file. Currently the CSV file needs to have headers for all columns, in the first row.

The --top parameter allows to limit the number of matched growth rates reported.

The --outfile parameter is optional and allows to write the generated plot to a PNG file on disk. If the --outfile parameter is not passed, the default GUI from matplotlib will be used to display the plot.

usage: match-growth.py [-h] --infile INFILE [--outfile OUTFILE] --col1 COL1
                       --col2 COL2 [--top TOP] [--catalog CATALOG] [--debug]
                       [--plot PLOT_TYPE]

Tool for estimating growth rates

optional arguments:
  -h, --help         show this help message and exit
  --infile INFILE    CSV input file with measurements
  --outfile OUTFILE  PNG file to write the plot to
  --col1 COL1        data column 1 from the CSV
  --col2 COL2        data column 2 from the CSV
  --top TOP          only print top N closest growth rates
  --catalog CATALOG  select the function catalog to use
  --debug            enable debug mode
  --plot PLOT_TYPE   plot type (normal|loglog)

Example usage:

match-growth.py --infile ./tests/modif7_bench.txt --outfile r1.png --col1 N --col2 memory --top 2

In this example we're estimating the closest growth rate for an algorithm that has a single input N and for which we've already recorded the memory usage for different values of N.

For questions or requests for paid support, please send an e-mail to [email protected]