CAGRA: variable graph degree for CPU-based algorithm#2031
Conversation
|
Auto-sync is disabled for draft pull requests in this repository. Workflows must be run manually. Contributors can view more details about this message here. |
achirkin
added
feature request
|
/ok to test |
|
/ok to test |
|
/ok to test |
mfoerste4
left a comment
mfoerste4
left a comment
There was a problem hiding this comment.
graph_core.cuh looks good to me.
| if (variable_graph_degree) { | ||
| RAFT_LOG_INFO("# Pruning kNN graph (size=%lu, degree=%lu, target_pruned_degree=%lu)", | ||
| graph_size, | ||
| knn_graph_degree, | ||
| target_pruned_degree); | ||
| } |
There was a problem hiding this comment.
You might want to move down the legacy RAFT_LOG here as well and combine it.
| 0.0, | ||
| false, | ||
| normalize_mean); | ||
| raft::copy(res, raft::make_host_scalar_view(&avg_natural), d_avg_natural.view()); |
There was a problem hiding this comment.
Why did you not choose map_reduce on the vector directly?
| if constexpr (VariableDegree) { | ||
| if (i + 1 == target_pruned_degree) { | ||
| // Freeze the detour level after we've placed exactly target_pruned_degree edges. | ||
| target_detour_level = warp_min_count; | ||
| } else if (i >= target_pruned_degree && warp_min_count > target_detour_level && | ||
| natural_degree == output_graph_degree) { | ||
| // The detour level just rose above the target band. Record the natural degree once. | ||
| natural_degree = i; | ||
| } | ||
| } |
There was a problem hiding this comment.
So we just track the 'natural_degree' here but continue to fill up the output graph - is this required?
There was a problem hiding this comment.
Yes, because we need the full graph in the other steps; we only use the natural degree later during the merging step.
|
/ok to test |
|
@achirkin can we get some benchmarks attached to this PR please demonstrating the improvements? It's a lot easier to point users to PRs to see benefits of changes than it is to have to look up slide decks or internal convos. |
|
Here are results on couple datasets. We look at red line (CAGRA graph with default parameters) vs broun line (HNSW graph). I'm not yet happy with the performance and want to tweak the merging logic just a little bit more. Apart from this, the PR is ready for review. 
(the build time is virtually unchanged with/without variable degree option) |
4 participants
Modify
optimizeroutine of CAGRA build process to allow variable graph degree.Introduce
variable_graph_degree_fractionparameter (by default = 1.0 = normal / constant degree behavior). This parameter defines the minimum allowed graph degree for any graph node.