sgd-update — Snowflake case

Snowflake-inspired ML platform team is validating a small optimizer implementation before using it inside a larger training workflow. You need to compute one exact SGD parameter update from the current parameters, learning rate, and gradients.

Compute one gradient-descent step: $\theta \leftarrow \theta -\eta g\backslash theta\backslash leftarrow\backslash theta-\backslash eta\backslash ,g$.

Implement solve(theta,grad,lr).

Returns. Return a NumPy array with the updated parameters; the inputs are not modified.

Examples

Example 1

Input:

solve([1, 2], [0.5, -1], 0.1)

Output:

[0.95, 2.1]

Example 2

Input:

solve([0.0], [10.0], 0.01)

Output:

[-0.1]