He Initialization

Create a weight matrix with He (Kaiming) normal initialisation: draw each entry from $\mathcal{N}(0,2\mathrm{/}\text{fan\_in})\backslash mathcal\; N(0,\backslash ,2/\backslash text\{fan\backslash \_in\})$ using np.random.default_rng(seed).normal(0, sqrt(2/fan_in), (fan_in, fan_out)).

Implement solve(fan_in,fan_out,seed=0).

Returns. Return a float NumPy array of shape (fan_in, fan_out).

Examples

Example 1

Input:

solve(2, 3, 0)

Output:

[[0.12573, -0.132105, 0.640423], [0.1049, -0.535669, 0.361595]]

Example 2

Input:

solve(4, 1, 3)

Output:

[[1.443148], [-1.807128], [0.295641], [-0.401474]]