docs: add personalized PageRank on a grid tutorial #821
Conversation
BeaMarton13
changed the title
There was a problem hiding this comment.
Nice work, well done!
Could you add a very short explanation, perhaps a "note", about what PageRank is? The PageRank score of a vertex is the frequency with which a random walker traverses that vertex. After each step, the walker has a 1 - damping chance to restart the walk, picking a starting vertices with probabilities given in the reset vector.
| plt.title("Graph with Personalized PageRank") | ||
| plt.show() |
There was a problem hiding this comment.
Let's make the plot not be stretched vertically or horizontally. Something like plt.axis("equal") can achieve this.
| # %% | ||
| # Then we set the nodes to prioritize, for example nodes with indices ``0`` and ``6``: | ||
| reset_vector[0] = 1 | ||
| reset_vector[6] = 1 |
There was a problem hiding this comment.
Let's choose two nodes that are a bit further from each other, and give one of them a higher weight than the other. This will make a more interesting plot.
Alternatively, use the reset_vertices parameter and pass in the vertices. They will have equal weights in this case.
It's your choice which pattern you demonstrate.
There was a problem hiding this comment.
@ntamas In IGraph/M I allow reset_vertices to be a dictionary. They keys are vertices and the values are reset weights. What do you think about implementing the same for python-igraph, assuming that @BeaMarton13 is interested in doing it?
There was a problem hiding this comment.
Absolutely, we can try to do it.
|
|
||
| # %% | ||
| # Then we calculate the personalized PageRank: | ||
| personalized_page_rank = g.personalized_pagerank(weights=None, damping=0.85, reset=reset_vector) |
There was a problem hiding this comment.
Let's remove weights here since there are no weights to think about. You can choose to tune the damping to make the plot look more interesting. Even if you leave it at the default of 0.85, do keep the damping parameter, as this is something people may want to play with.
You could even choose to show two plots with two different damping values, to illustrate the effect of this parameter.
Co-authored-by: Szabolcs Horvát <[email protected]>
This PR implements a tutorial for calculating and visualizing personalized PageRank on a grid, as suggested here.