Graph summary

I thiiiink that summary is only set when you use that graph inside another graph?

I thought that when I set its value, it should be serialized and deserialized. And of course I'm using it in cases where pragh contains multiple SimpleVcectorIndex.

I think it is not used, so it is not serialized.

The summary of the graph is only needed when creating a graph that contains that graph

Well, I haven't tried to do that. You didn't give an example of that, so I didn't know that a graph could contain another graph.

There's an example provided here (it's pretty new) 💪 https://gpt-index.readthedocs.io/en/latest/guides/tutorials/graph.html

As I said before,I'm using a GPTListIndex on top of many simple vector indexes to build a graph,these vector indexes relate to many different topics.And this graph will be the only one graph.Newly vector index will be append to this graph too.I use this graph to answer any questions about the knowledge of all these vector indexes. I want to know whether I'm using them correctly and efficient?
I'm actually confused about how to choose and how to combine them (such like GPTListIndex,GPTSimpleKeywordTableIndex,GPTTreeIndex).

For my use case, different ducument of index may be irrelevant topics, and I also wish that my graph can take on the role of routing to find the right index to answer the question, How should I combine and utilize them in order to achieve their maximum potential?

@Logan M I would like to get your suggestions and thoughts on this matter.

A list index at the top level will always check ever index below it. If you keep adding more, this process will get pretty slow

You could have a keyword, tree, or even another vector index for the top level instead. Just need to make sure you generated a good summary for each sub index

OK, I'll replace the top level GPTListIndex with a keyword or tree index.