rmsander

class CompositeBatchedGP(ExactGP):
    """Class for creating batched Gaussian Process Regression models.  Ideal candidate if
    using GPU-based acceleration such as CUDA for training.

    This kernel produces a composite kernel that multiplies actions times states,
    i.e. we have a different kernel for both the actions and states.  In turn,
    the composite kernel is then multiplied by a Scale kernel.

    Parameters:
        train_x (torch.tensor): The training features used for Gaussian Process

rmsander

class BatchedGP(ExactGP):
    """Class for creating batched Gaussian Process Regression models.  Ideal candidate if
    using GPU-based acceleration such as CUDA for training.

    Parameters:
        train_x (torch.tensor): The training features used for Gaussian Process
            Regression.  These features will take shape (B * YD, N, XD), where:
                (i) B is the batch dimension - minibatch size
                (ii) N is the number of data points per GPR - the neighbors considered
                (iii) XD is the dimension of the features (d_state + d_action)