Gradient DescentEasy
Spot Gradient Descent Divergence
~14 mincode completion
Implement bowl_descent(curvatures, theta0, lr, steps) that runs the update steps times and returns the array [final_loss, diverged], with diverged equal to 1.0 or 0.0.
Where this shows up
~14 min
••••••
3 employers weight this skill
1 big tech firm, 1 frontier lab, 1 quant fund. Top match scores 80.
Python
import numpy as np
def bowl_descent(curvatures, theta0, lr, steps):
"""
Gradient descent on L = 0.5 * (c1 * t1^2 + c2 * t2^2).
Args:
curvatures: array [c1, c2], both positive
theta0: starting point [t1, t2]
lr: learning rate
steps: number of updates k (0 or more)
Returns:
array [final_loss, diverged], diverged 1.0 if final_loss > starting loss else 0.0
"""
# YOUR CODE HERE
pass