bonni.InitType

class bonni.InitType(*values)

Enumeration of initialization strategies for model parameters.

These values define how weights or biases are initialized before training begins. Used primarily for bias_init in the model configuration.

zeros

Initializes parameters to exactly 0. This is the standard practice for bias terms in most neural network layers.

ones

Initializes parameters to exactly 1.

uniform

Initializes parameters with values drawn from a uniform distribution. The range is typically determined by the specific layer implementation.

normal

Initializes parameters with values drawn from a normal (Gaussian) distribution.

__init__(*args, **kwds)

Attributes

zeros
ones
uniform
normal