Closed devvrit closed 5 years ago
Fot 1. It's just an alternative implementation of the algorithm. We try to ensure the neighbor selection insensitive to the order of processing. For 2 & 3. We are testing the robustness of this part of the code. Later we will update it. For 4. We actually do not need an end mark. For 5. NSG is an approximation of MRNG. We make some trade-offs to ensure the memory and time efficiency. In NSG, we do not try to ensure all monotonic paths. Instead, most nodes will not exceed the limitation of the degree. For the long-tail part who exceeds the limitation, we cut off distant neighbors to ensure memory efficiency. The cost is some monotonic paths are not guaranteed and a little detour will be needed at search time. We have tested that there is no impact on the search performance.
Hi. I have some questions regarding the algorithm and the code: