A mistake(?) in Std implementation

[EngineerKhan]

I have made a simple snippet:

key = rand.PRNGKey(0)
x = rand.uniform(key, (1,100))
y = jnp.arange(1,101)

stdv = lambda x: jnp.linalg.norm(x-jnp.average(x))/(jnp.sqrt(len(x)))

When I call this stdv for both vectors and compare it with the JAX intrinsic one jnp.std(), it gives the same answers for y, but different for x. Why is it so?

[jakevdp]

Your x is 2-dimensional and y is one-dimensional. This means that y.size and len(y) are the same, but x.size and len(x) are not. If you switch jnp.sqrt(len(x)) to jnp.sqrt(x.size), your implementation will be correct for these inputs.