jax.nn.initializers.he_normal

Contents

jax.nn.initializers.he_normal#

jax.nn.initializers.he_normal(in_axis=-2, out_axis=-1, batch_axis=(), dtype=<class 'jax.numpy.float64'>)[source]#

Builds a He normal initializer (aka Kaiming normal initializer).

A He normal initializer is a specialization of jax.nn.initializers.variance_scaling() where scale = 2.0, mode="fan_in", and distribution="truncated_normal".

Parameters:

  • in_axis (int | Sequence[int]) – axis or sequence of axes of the input dimension in the weights array.

  • out_axis (int | Sequence[int]) – axis or sequence of axes of the output dimension in the weights array.

  • batch_axis (Sequence[int]) – axis or sequence of axes in the weight array that should be ignored.

  • dtype (DTypeLikeInexact) – the dtype of the weights.

Returns:

An initializer.

Return type:

Initializer

Examples:

>>> import jax, jax.numpy as jnp
>>> initializer = jax.nn.initializers.he_normal()
>>> initializer(jax.random.key(42), (2, 3), jnp.float32)  
Array([[ 0.6604483 ,  1.1900088 ,  1.2047218 ],
       [-0.87225807, -0.95321447,  0.1369438 ]], dtype=float32)
Contents