Performance

Although parsnip is optimized for accuracy rather than performace, our parsing strategy results in file reads and unit-cell reconstructions that are 3-500x faster than comparable tools. The following data makes use of ltalirz/cif-parsing-benchmark, a benchmark that profiles the parsing throughput of several open-source CIF libraries.

While not the fastest CIF parser around, parsnip achieves competetive performance when reading files in addition to our class-leading accuracy.

Increasing Performance

In some cases, particularly when constructing thousands of unit cells, the performance of parsnip’s build_unit_cell may become a bottleneck. parsnip includes several tools for resolving this: first, parse_mode="python_float" attempts to build unit cells using floating point arithmetic rather than rational expression. This is less accurate, but is still sufficient for high-quality databases and stuctures. For the best combination of performance and accuracy, installing the cfractions library lets parsnip use more optimized code for unit cell reconstruction. This is functionally equivalent to the default mode, but several times faster.

>>> # uv pip install cfractions
>>> from parsnip import CifFile
>>> cif = CifFile("hP3.cif")

>>> # If `cfractions` is available it is used by the default `parse_mode="rational"`
>>> faster = cif.build_unit_cell(n_decimal_places=4)
>>> faster
array([[0.2254    , 0.        , 0.33333333],
       [0.        , 0.2254    , 0.66666667],
       [0.7746    , 0.7746    , 0.        ]])
>>> assert faster.shape == (3, 3)

Reproducing these Benchmarks

All benchmarks in this file were obtained using Python 3.13.2 on an M1 Macbook Pro, with parsnip version 0.6.1 and the uv.lock file associated with that tag. To reproduce the results on your own hardware, run the following commands from the root of the repository:

uv sync --group tables

# Measure parsnip's performance reading CIF files
./doc/generate_benchmark_plots/cif_parsing_benchmark.sh

# Compare the efficiency of various parsing modes
./doc/generate_benchmark_plots/parse_mode_benchmark_plot.py

# Measure the space group and unit cell reconstruction accuracy
python _joss/generate_table_1.py
python _joss/generate_table_2.py