Kevin P Murphy
Kevin P Murphy
See also https://github.com/probml/dynamax/issues/130
This notebook also has a unit test for monotoncally increaseing, and seems to pass, at least on a CPU. https://github.com/probml/dynamax/blob/main/docs/notebooks/linear_gaussian_ssm/lgssm_learning.ipynb
See https://github.com/google/jax/issues/13224
Correct. The input vector u_t at each time step must be a D-dimensional vector. So `inputs` has shape `(T,D)` (or `None`). You can always flatten your 3d inputs outside of...
See also https://github.com/probml/dynamax/issues/50
If we pass the valid length off each sequence, we can lax.scan only over that prefix. missing data at random times could be handled with an if statement for conditional...
Also need to replace `print((torch.symeig(X).eigenvalues > 0.).all())` with `print((torch.linalg.eigh(X).eigenvalues > 0.).all()) `
And the "right inverse" section needs updating in the same way ``` class CayleyMap(nn.Module): def __init__(self, n): super().__init__() self.register_buffer("Id", torch.eye(n)) def forward(self, X): # Assume X skew-symmetric # (I +...
Sec E.4 of paper
We should also add a bernoulli bandit, not just the Gaussian bandit. We can use CMGF method which we didn't have back then.