init
26 memberslucid.nn.initnn.init: parameter initialization functions. All functions operate in-place and return the tensor.
Functions
uniform_→ TensorInitialise tensor in-place with samples from a uniform distribution.
normal_→ TensorInitialise tensor in-place with samples from a Gaussian distribution.
constant_→ TensorFill tensor in-place with a single scalar value.
ones_→ TensorFill tensor in-place with ones.
zeros_→ TensorFill tensor in-place with zeros.
eye_→ TensorFill a 2-D tensor in-place with the identity matrix.
xavier_uniform_→ TensorInitialise tensor in-place with Xavier (Glorot) uniform initialisation.
xavier_normal_→ TensorInitialise tensor in-place with Xavier (Glorot) normal initialisation.
kaiming_uniform_→ TensorInitialise tensor in-place with Kaiming (He) uniform initialisation.
kaiming_normal_→ TensorInitialise tensor in-place with Kaiming (He) normal initialisation.
trunc_normal_→ TensorInitialise tensor in-place with truncated normal samples.
orthogonal_→ TensorInitialise tensor in-place with a (semi-)orthogonal matrix.
sparse_→ TensorInitialise a 2-D tensor in-place with a sparse random matrix.
dirac_→ TensorInitialise a 3/4/5-D convolution weight in-place as a Dirac delta.
calculate_gain→ floatReturn the recommended variance-preserving gain for an activation.
constant→ TensorFill tensor in-place with a single scalar value.
dirac→ TensorInitialise a 3/4/5-D convolution weight in-place as a Dirac delta.
eye→ TensorFill a 2-D tensor in-place with the identity matrix.
kaiming_normal→ TensorInitialise tensor in-place with Kaiming (He) normal initialisation.
kaiming_uniform→ TensorInitialise tensor in-place with Kaiming (He) uniform initialisation.
normal→ TensorInitialise tensor in-place with samples from a Gaussian distribution.
orthogonal→ TensorInitialise tensor in-place with a (semi-)orthogonal matrix.
sparse→ TensorInitialise a 2-D tensor in-place with a sparse random matrix.
uniform→ TensorInitialise tensor in-place with samples from a uniform distribution.
xavier_normal→ TensorInitialise tensor in-place with Xavier (Glorot) normal initialisation.
xavier_uniform→ TensorInitialise tensor in-place with Xavier (Glorot) uniform initialisation.