feat: power iteration clustering

.gitignore

@@ -15,6 +15,7 @@ gf_df_tmp*
 gf_checkpoints*
 gf_workdir*
 test-test*
+email-*
 .#*
 /graph500*
 /wiki-Talk*

Cargo.toml

@@ -19,6 +19,7 @@ futures = "0.3"
 uuid = {version = "1", features = ["v4"] }
 log = "0.4"
 rand = { version = "0.9", features = ["std_rng"] }
+wide = "0.7"
 datasketches = { version = "0.3.0", default-features = false, features = ["hll"] }
 
 # CLI dependencies

README.md

@@ -45,7 +45,9 @@ Available algorithms:
 - `kcore`: K-Core decomposition;
 - `hyperanf`: Approximate Neighbor Function;
 - `shortest-path`: Multi-Source Shortest Path;
-- `classical-lp`: Classical (Raghavan) Label Propagaion
+- `classical-lp`: Classical (Raghavan) Label Propagaion;
+- `pic`: Power Iteration Clustering;
+- `mllib kmeans`: raw K-Means over a `List<Float32>` feature column of the vertices file (no edges read).
 
 Each has its own arguments — run `graphframes <algorithm> --help` for the full list. For what each algorithm computes, see [References](#references).
 
@@ -103,6 +105,7 @@ TBD
 ### Community Detection
 
 - **Classical Label Propagation**: _Raghavan, Usha Nandini, Réka Albert, and Soundar Kumara. "Near linear time algorithm to detect community structures in large-scale networks." Physical Review E—Statistical, Nonlinear, and Soft Matter Physics 76.3 (2007): 036106._
+- **Power Iteration Clustering**: _Lin, Frank, and William W. Cohen. "Power iteration clustering." (2010)._
 
 ### Subgraphs
 

benches/python/main.py

@@ -57,6 +57,9 @@
     "sp":       {"cli": "shortest-path", "args": [], "undirected": True},
     "cdlp":     {"cli": "classical-lp", "args": ["--max-iter", "10"], "undirected": False},
     "mis":      {"cli": "mis", "args": [], "undirected": False},
+    # PIC symmetrizes internally (the affinity matrix is symmetric by
+    # definition), so it needs no --symmetrize input handling.
+    "pic":      {"cli": "pic", "args": ["--k", "2", "--max-iter", "20"], "undirected": False},
 }
 
 

src/algorithm.rs

@@ -1,5 +1,5 @@
 mod centrality;
-mod community;
+pub(crate) mod community;
 mod connectivity;
 mod pregel;
 mod subgraph;

src/algorithm/centrality/hyperanf.rs

@@ -95,6 +95,7 @@ impl<'a> HyperANFBuilder<'a> {
                     .clone()
                     .select_columns(&[EDGE_SRC, EDGE_DST])?,
                 true,
+                None,
             )?
         };
 

src/algorithm/centrality/k_core.rs

@@ -87,6 +87,7 @@ impl<'a> KCoreBuilder<'a> {
                 .clone()
                 .select_columns(&[EDGE_SRC, EDGE_DST])?,
             true,
+            None,
         )?;
 
         // The undirected degree is the out-degree of the symmetrized graph.

src/algorithm/community.rs

@@ -1 +1,2 @@
 mod classical_lp;
+pub(crate) mod power_iteration_clustering;

src/algorithm/community/classical_lp.rs

@@ -74,6 +74,7 @@ impl<'a> ClassicalLPBuilder<'a> {
                     .clone()
                     .select_columns(&[EDGE_SRC, EDGE_DST])?,
                 false,
+                None,
             )?
         };