Wrapper for Clustal Omega.

[jabaws.git] / binaries / src / clustalo / src / kmpp / KmTree.h

diff --git a/binaries/src/clustalo/src/kmpp/KmTree.h b/binaries/src/clustalo/src/kmpp/KmTree.h
new file mode 100644 (file)
index 0000000..48cf53e
--- /dev/null
+++ b/binaries/src/clustalo/src/kmpp/KmTree.h
@@ -0,0 +1,77 @@
+// BEWARE: BETA VERSION\r
+// --------------------\r
+//\r
+// A k-d tree that vastly speeds up an iteration of k-means (in any number of dimensions). The main\r
+// idea for this data structure is from Kanungo/Mount. This is used internally by Kmeans.cpp, and\r
+// will most likely not need to be used directly.\r
+//\r
+// The stucture works as follows:\r
+//   - All data points are placed into a tree where we choose child nodes by partitioning all data\r
+//     points along a plane parallel to the axis.\r
+//   - We maintain for each node, the bounding box of all data points stored at that node.\r
+//   - To do a k-means iteration, we need to assign points to clusters and calculate the sum and\r
+//     the number of points assigned to each cluster. For each node in the tree, we can rule out\r
+//     some cluster centers as being too far away from every single point in that bounding box.\r
+//     Once only one cluster is left, all points in the node can be assigned to that cluster in\r
+//     batch.\r
+//\r
+// Author: David Arthur (darthur@gmail.com), 2009\r
+\r
+#ifndef KM_TREE_H__\r
+#define KM_TREE_H__\r
+\r
+// Includes\r
+#include "KmUtils.h"\r
+\r
+// KmTree class definition\r
+class KmTree {\r
+ public:\r
+  // Constructs a tree out of the given n data points living in R^d.\r
+  KmTree(int n, int d, Scalar *points);\r
+  ~KmTree();\r
+\r
+  // Given k cluster centers, this runs a full k-means iterations, choosing the next set of\r
+  // centers and returning the cost function for this set of centers. If assignment is not null,\r
+  // it should be an array of size n that will be filled with the index of the cluster (0 - k-1)\r
+  // that each data point is assigned to. The new center values will overwrite the old ones.\r
+  Scalar DoKMeansStep(int k, Scalar *centers, int *assignment) const;\r
+\r
+  // Choose k initial centers for k-means using the kmeans++ seeding procedure. The resulting\r
+  // centers are returned via the centers variable, which should be pre-allocated to size k*d.\r
+  // The cost of the initial clustering is returned.\r
+  Scalar SeedKMeansPlusPlus(int k, Scalar *centers) const;\r
+\r
+ private:\r
+  struct Node {\r
+    int num_points;                 // Number of points stored in this node\r
+    int first_point_index;          // The smallest point index stored in this node\r
+    Scalar *median, *radius;        // Bounding box center and half side-lengths\r
+    Scalar *sum;                    // Sum of the points stored in this node\r
+    Scalar opt_cost;                // Min cost for putting all points in this node in 1 cluster\r
+    Node *lower_node, *upper_node;  // Child nodes\r
+    mutable int kmpp_cluster_index; // The cluster these points are assigned to or -1 if variable\r
+  };\r
+\r
+  // Helper functions for constructor\r
+  Node *BuildNodes(Scalar *points, int first_index, int last_index, char **next_node_data);\r
+  Scalar GetNodeCost(const Node *node, Scalar *center) const;\r
+\r
+  // Helper functions for DoKMeans step\r
+  Scalar DoKMeansStepAtNode(const Node *node, int k, int *candidates, Scalar *centers,\r
+                            Scalar *sums, int *counts, int *assignment) const;\r
+  bool ShouldBePruned(Scalar *box_median, Scalar *box_radius, Scalar *centers, int best_index,\r
+                      int test_index) const;\r
+\r
+  // Helper functions for SeedKMeansPlusPlus\r
+  void SeedKmppSetClusterIndex(const Node *node, int index) const;\r
+  Scalar SeedKmppUpdateAssignment(const Node *node, int new_cluster, Scalar *centers,\r
+                                  Scalar *dist_sq) const;\r
+\r
+  int n_, d_;\r
+  Scalar *points_;\r
+  Node *top_node_;\r
+  char *node_data_;\r
+  int *point_indices_;\r
+};\r
+\r
+#endif\r