Bottleneck

1.6.0 · active · verified Thu Apr 09

Bottleneck is a Python library that provides a collection of fast, C-optimized functions for NumPy arrays, particularly useful for operations involving NaNs (Not a Number). It aims to offer significant performance improvements over equivalent NumPy functions for large arrays by implementing them in C. The current version is 1.6.0, with releases generally following significant NumPy updates or critical bug fixes, ensuring compatibility and performance.

Warnings

gotcha Results from bottleneck functions might differ slightly from NumPy equivalents due to distinct algorithms and floating-point arithmetic. This is generally within acceptable numerical precision but could be a concern for highly sensitive applications.
Fix: Always test `bottleneck` functions against `numpy` for critical numerical precision requirements to ensure the minor differences are acceptable for your application.
gotcha Performance gains from bottleneck are primarily noticeable for large arrays and specific operations (especially those handling NaNs). For small arrays, the overhead of calling the C extension might negate or even reverse performance benefits compared to pure NumPy.
Fix: Profile your code with both `bottleneck` and `numpy` for your specific array sizes and operations to confirm `bottleneck` provides a benefit in your use case.
breaking Bottleneck 1.x series (including 1.6.0) requires Python 3.10 or newer. Installing on older Python versions will fail.
Fix: Ensure your Python environment is version 3.10 or newer. If you must use an older Python version, refer to bottleneck's historical documentation for compatible versions (e.g., 1.3.2 for Python 3.7+).

Install

pip install bottleneck Install stable release

Imports

bottleneck
```
import bottleneck as bn
```

Quickstart

This example demonstrates how to import bottleneck as 'bn' and use its NaN-aware functions, such as `nansum` and `nanmean`, on NumPy arrays for potentially faster computation.

import numpy as np
import bottleneck as bn

a = np.array([1.0, 2.0, np.nan, 4.0])
b = np.arange(12.0).reshape(3, 4)
b[0, 0] = np.nan

print(f"Original array a: {a}")
print(f"bn.nansum(a): {bn.nansum(a)}")

print(f"\nOriginal array b:\n{b}")
print(f"bn.nanmean(b, axis=1): {bn.nanmean(b, axis=1)}")

view raw JSON →