A library for generating mathematical objects from parameterisations.
E.g. you would like to build a dataclass like:
class MyNetwork(BaseModel):
weights: Tensor
with weights sampled from a Gaussian distribution (mean
and variance
).
Cast would let you call something like:
network = MyNetwork.build("{ 'weights': {'mean': 0.0, 'var': 0.1} }")
to build an instance of MyNetwork
using a parameterisation of the matrix.