twodlearn.core.initializers module¶

class twodlearn.core.initializers.FrobeniusNormal(scale=1.0, shape=None)[source]¶

Initializer that creates a matrix sampled from Normal(0, scale**2/(fan_in * fan_out))

class twodlearn.core.initializers.KernelInitializer(scale=1.0, shape=None)[source]¶: Bases: object

class twodlearn.core.initializers.MaxNormal(scale=1.0, shape=None)[source]¶

Initializer that creates a matrix sampled from Normal(0, scale**2/(max(fan_in, fan_out)))

class twodlearn.core.initializers.SingularNormal(scale=1.0, shape=None)[source]¶

Initializer that creates a matrix

class twodlearn.core.initializers.SumFanConstant(scale=1.0)[source]¶

Bases: object

Initializer that creates a matrix fill with a constant value equal to the SumFan rule

The sumfan rule is:

output = sqrt(scale**2/(fain_in  + fan_out)) eye(shape)

class twodlearn.core.initializers.SumNormal(scale=1.0, shape=None)[source]¶

Initializer that creates a matrix sampled from Normal(0, scale**2/(fan_in + fan_out))

twodlearn.core.initializers.compute_fans(kernel_shape)[source]¶

Computes the number of input and output units for a weight shape. :param shape: Integer shape tuple or TF tensor shape.