This package provides standardized plotting routines. It is part of beNNch, however, it can also be used in a standalone fashion.
It is designed to work with performance benchmarking results stored in .csv files that adhere to a common naming convention.
On the top level, run
pip install .Alternatively, if you want to make changes on the fly (e.g. change plotting parameters), execute
pip install -e .Both commands install beNNch-plot as package such that it can be used anywhere on your system via import bennchplot.
Conceptually, beNNch-plot adheres to a modular design philosophy. This means that arrangement of the created figure is not done internally but defineable on the surface level by the user. Concretely, this means that the user can provide their own axis objects, thereby retaining customization options such as setting titles and labels.
For measures that are used by default in beNNch, default colors and labels are provided. This can be extended if desired.
Examples of how to use beNNch-plot are provided in ./examples. Here you can find examples for two models, the microcircuit and the multi-area-model. The necessary underlying performance results are given alongside.
The microcircuit serves as an example of a benchmark model that can be run across different numbers of virtual processes on a single node. After defining custom axes, microcircuit.py calls the two main plotting functions of beNNch-plot: plot_main for plotting simple line or error plots as plot_fractions to create a fill_between-style plot.
In contrast to the microcircuit, the multi-area model showcases benchmarks across multiple number of nodes. While the basics are the same as for the microcircuit, an additional panel is created for showing the network construction time together with the state propagation time. Additionally, multi-area-model_ram.py gives a minimal example of how to plot other measurements than times.