ferray-numpy-interop

Zero-copy conversions between ferray arrays and Python/Arrow/Polars for the ferray scientific computing library.

What's in this crate

• NumPy (feature python): AsFerray and IntoNumPy traits via PyO3
• Apache Arrow (feature arrow): FromArrow/ToArrow traits + RecordBatch round-trip + null/missing handling
• Polars (feature polars): FromPolars/ToPolars DataFrame round-trip
• Dtype coverage: f32/f64, i8-i128, u8-u128, bool, f16, bf16, complex, complex — all with dtype validation on conversion
• DynArray integration: convert Array<T, IxDyn> to/from PyArrayDyn / Arrow dynamic-typed columns
• Zero-copy when arrays are C-contiguous, safe copy otherwise

All backends are feature-gated and disabled by default.

Usage

[dependencies.ferray-numpy-interop]
version = "0.1"
features = ["python"]  # or "arrow", "polars"

use ferray_numpy_interop::{AsFerray, IntoNumPy};
use pyo3::prelude::*;
use numpy::PyReadonlyArray1;

fn process(py: Python<'_>, arr: PyReadonlyArray1<'_, f64>) -> PyResult<()> {
    let ferray_arr = arr.as_ferray()?;
    // ... process ...
    Ok(())
}

This crate is re-exported through the main ferray crate with the numpy feature.

License

MIT OR Apache-2.0