Hi there.
I am reading the original KDD paper. Sec3.4 described an incremental online training approach, that is, if a new edge (u,v) comes , update the random walk set with new walks containing (u,v). However, when I run the code, the code just adds random walks starting from node u and v into the random walk set and retains the previous walks which is not started with u or v. And I also can not find whether there is an incremental online training started with triple loss (I'm not sure if I really understand discussion 1). What I really find is a total retaining process starting from newly constructed random walk set. Can anyone explain why? Thks
OK I figured out problem 1 how to update the random walk. The implementation is consisted with the states in the original paper. But I still can not find a incremental online training
Hi there. I am reading the original KDD paper. Sec3.4 described an incremental online training approach, that is, if a new edge (u,v) comes , update the random walk set with new walks containing (u,v). However, when I run the code, the code just adds random walks starting from node u and v into the random walk set and retains the previous walks which is not started with u or v. And I also can not find whether there is an incremental online training started with triple loss (I'm not sure if I really understand discussion 1). What I really find is a total retaining process starting from newly constructed random walk set. Can anyone explain why? Thks