Polars Series have support for NumPy universal functions (ufuncs). Element-wise functions such as np.exp(), np.cos(), np.div(), etc. all work with almost zero overhead.

However, as a Polars-specific remark: missing values are a separate bitmask and are not visible by NumPy. It can yield to a window function or a np.convolve() giving flawed or incomplete results.

Convert a Polars Series to a NumPy array with the .to_numpy() method. Missing values will be replaced by np.nan during the conversion. If the Series does not include missing values, or those values are not desired anymore, the .view() method can be used instead, providing a zero-copy NumPy array of the data.