Working with NumPy and Pandas

NumPy Best Practices

Pyvorin compiles the Python orchestration code around NumPy calls. The NumPy C operations themselves already run at native speed.

import numpy as np

def process(arr):
    result = []
    for i in range(len(arr)):
        val = arr[i] * 2 + 1
        result.append(val)
    return np.array(result)

The Python loop above will be compiled natively; the np.array call falls back to CPython gracefully.

Pandas DataFrames

DataFrame operations inside loops can benefit from compilation:

def transform_rows(df):
    results = []
    for idx, row in df.iterrows():
        results.append(row['price'] * row['quantity'])
    return results

SciPy

SciPy functions are C-backed and do not need compilation. Focus Pyvorin on the Python glue code that calls SciPy.

Interop Limitations

• Direct mutation of NumPy C arrays from compiled code is unsupported.
• Pandas ExtensionArrays may fall back to CPython.
• Numba-jitted functions cannot be further compiled by Pyvorin.