Create Your Own ufunc

🛠️ Creating Your Own ufunc in NumPy

NumPy allows you to create custom universal functions (ufuncs), which can then operate element-wise on NumPy arrays just like built-in ufuncs.
There are two main ways:

1. Using np.vectorize() (Python-level, slower, but easy)

2. Using NumPy C-API or Numba (fast, compiled, advanced)

Here, we focus on np.vectorize(), which is sufficient for most Python use-cases.

1. Using np.vectorize()

np.vectorize() wraps a Python function so it can be applied element-wise to arrays.

• Works element-wise

• Can handle any Python function

 2. Using Lambda Function

• Compact and convenient for simple functions

 3. Using otypes Parameter

You can specify output type with otypes to avoid type inference issues.

4. Using Multiple Inputs

• Supports multiple array inputs

• Element-wise operations are performed

5. Example: Conditional Function ufunc

• Useful for categorical transformations

🧠 Notes

• np.vectorize() is not truly compiled; it’s essentially a for loop in Python.

• For high-performance ufuncs, consider Numba’s @vectorize decorator or Cython.

• Can handle scalars, arrays, multiple inputs, and different output types.

🎯 Practice Exercises

1. Create a ufunc that computes f(x) = x^3 – x + 2 for an array.

2. Create a ufunc that classifies numbers as even/odd.

3. Create a ufunc with two inputs that computes x^2 + y^2.