I’m frequently running into problems where I run a for-loop and subset an array. Here’s an example of the kind of calculation I’m doing.

import numpy as np

def get_betas(x, y):
    
    # Calculate parts of OLS formula
    xy = x.T @ y
    xx = x.T @ x
    xx_inv = np.linalg.inv(xx)
    
    # Calculate beta
    beta = xx_inv @ xy
    
    return beta

def calc_oos_error(x, y, k):
    
    errors = []
    
    for t in range(k, len(y)):
        
        beta = get_betas(x[:t], y[:t])
        
        error = y[t] - x[t] @ beta
        
        errors.append(error)
        
    return np.mean(errors), np.std(errors)

x = np.random.normal(size = 1000)
y = 10 * x + np.random.normal(scale = 1e-2, size = len(x))

x = x.reshape(-1, 1)

a, b = calc_oos_error(x, y, 10)

I would like to vectorize out the loop in calc_oos_error. However, when I place the array into the vectorization function an error is thrown.

One option would be to use a global reference to x and y.

def calc(t):
    
    beta = get_betas(x[:t], y[:t])
        
    error = y[t] - x[t] @ beta
    
    return error

calc = np.vectorize(calc)    
    

def calc_oos_error_vec(x, y, k):
    
    errors = calc(np.arange(k, len(y)))
          
    return np.mean(errors), np.std(errors)

However, this strikes me as confusing and just asking for errors.