An identity layer f(x) = x

Load event sequences via minbatch

The class of generalized Hawkes process model contains most of necessary function.

The class of inhomogeneous Poisson process

The negative log-likelihood loss of events of point processes nll = sum_{i in batch}[ -log lambda_ci(ti) + sum_c Lambda_c(ti) ]

The cross entropy loss that maximize the conditional probability of current event given its intensity ls = -sum_{i in batch} log p(c_i | t_

Calculate Gromov-Wasserstein discrepancy given optimal transport and cost matrix

The least-square loss of events of point processes ls = || Lambda_c(t) - N(t) ||_F^2

The negative log-likelihood loss of events of point processes nll = [ -log lambda_ci(ti) + sum_c Lambda_c(ti) ]

The class of a mixture model of generalized Hawkes processes contains most of necessary function.

Calculate Wasserstein discrepancy given optimal transport and

The parent class of decay functions, which actually an exponential kernel

The parent class of endogenous impact functions sum_i phi_{kk_i}(t-t_i) for k = 1,...,C, which actually a simple endogenous impact with phi

The parent class of exogenous intensity function mu(t), which actually a constant exogenous intensity

The class of bilinear endogenous impact functions phi_{cc'}(t) = sum_m (f_{c}^T * W_m * f_{c'}) * kernel_m(t) Here W_m for m=1,...,M,

The class of exponential kernel, which is inherited from the parent class directly.

The class of factorized endogenous impact functions phi_{cc'}(t) = sum_m (u_{cm}^T * v_{c'm}) * kernel_m(t) Here, U_m=[u_{cm}] and V_m

The class of gate kernel.

The class of Gaussian kernel

The class of generalized Hawkes process model contains most of necessary function.

The class of linear endogenous impact functions phi_{cc'}(t) = sum_m (w_{cm}^T * f_{c'}) * kernel_m(t) Here W_m = [w_{cm}], for m=1,..

The class of linear exogenous intensity function mu_c(t) = w_c^T * f. Here f is nonnegative feature vector of a sequence.

The class of multi-gaussian kernel.

The class of naive endogenous impact functions sum_i phi_{kk_i}(t-t_i) for k = 1,...,C, which actually a simple endogenous impact with phi_

The class of constant exogenous intensity function mu(t) = mu

The class of neural exogenous intensity function mu_c(t) = F(c, f), where F is a 3-layer neural network, c is event type, and f is the feat

The class of generalized Hawkes process model contains most of necessary function.

The class of powerlaw kernel.

The class of Rayleigh kernel

Load event sequences with labels via minbatch