As mentioned in class, a good way to think about an operation like this is as a shader in OpenGL. Shaders compute the value of a particular mesh element given specific input data (ie color of vertex, colors of neighboring vertices, value in a uv texture map, etc). Here, we are using the values of the neighboring vertices (the in-neighbors) of a vertex to compute the rank of the vertex.
This comment was marked helpful 0 times.
dfarrow
@arjunh - That is a good analogy. The per-vertex kernel (analogous to a shader program) is essentially the inner block of a for loop that is called by the runtime, probably in parallel. I think that this "thinking like a vertex"* approach is used in both Graph Lab and in Google's Pregel.
As mentioned in class, a good way to think about an operation like this is as a shader in OpenGL. Shaders compute the value of a particular mesh element given specific input data (ie color of vertex, colors of neighboring vertices, value in a uv texture map, etc). Here, we are using the values of the neighboring vertices (the in-neighbors) of a vertex to compute the rank of the vertex.
This comment was marked helpful 0 times.
@arjunh - That is a good analogy. The per-vertex kernel (analogous to a shader program) is essentially the inner block of a for loop that is called by the runtime, probably in parallel. I think that this "thinking like a vertex"* approach is used in both Graph Lab and in Google's Pregel.
*source
This comment was marked helpful 0 times.