The example demonstrates how to use vector data for dimensionality reduction and clustering, using Principal Component Analysis (PCA) and k-Means.

1. Assume that there is a data set called datavec containing one ID column and a vector column with 100 dimensions.

   Name                   Null?    Type
   ---------------------- -------- ----------------------------
   ID                              NUMBER
   PROD_DATA                       VECTOR(100, FLOAT32, DENSE)

2. Build a PCA feature extraction model. The following step creates a model that uses PCA scoring to reduce dimensionality.

   DECLARE
     v_setlst DBMS_DATA_MINING.SETTING_LIST;
   BEGIN
     v_setlst('ALGO_NAME')         := 'ALGO_SINGULAR_VALUE_DECOMP';
     v_setlst('SVDS_SCORING_MODE') := 'SVDS_SCORING_PCA';
     
     DBMS_DATA_MINING.CREATE_MODEL2(
           MODEL_NAME          => 'pca_model',
           MINING_FUNCTION     => 'FEATURE_EXTRACTION',
           DATA_QUERY          => 'SELECT * FROM DATAVEC',
           CASE_ID_COLUMN_NAME => 'id',
           SET_LIST            => v_setlst);    
   END;
   /

3. Transform PCA results into a vector table pca_data with reduced dimensions by using the VECTOR_EMBEDDING() operator.

   CREATE table pca_data as SELECT id, VECTOR_EMBEDDING(pca_model using *) embedding FROM datavec;

4. The new pca_data contains one ID column and one vector with 10 dimensions based on the data characteristics.

   DESC pca_data;
   Name             Null?    Type
   ---------------- -------- ----------------------------
   ID                    	  NUMBER
   EMBEDDING                 VECTOR(10, FLOAT64, DENSE)

5. Build a k-Means clustering model on pca_data, leveraging its reduced dimensions.

   DECLARE
     v_setlst DBMS_DATA_MINING.SETTING_LIST;
   BEGIN
     v_setlst('ALGO_NAME')         := 'ALGO_KMEANS';
     v_setlst('KMNS_DETAILS')      := 'KMNS_DETAILS_ALL';
     v_setlst('CLUS_NUM_CLUSTERS') := '2';
   
     DBMS_DATA_MINING.CREATE_MODEL2(
           MODEL_NAME          => 'km_model',
           MINING_FUNCTION     => 'CLUSTERING',
           DATA_QUERY          => 'SELECT * FROM PCA_DATA',
           CASE_ID_COLUMN_NAME => 'id',
           SET_LIST            => v_setlst);    
   END;
   /

6. Check the data dictionary settings.

   SELECT model_name, attribute_name, data_type, target, vector_info 
   FROM   USER_MINING_MODEL_ATTRIBUTES
   WHERE  model_name='KM_MODEL’ ORDER BY attribute_name;
    
   MODEL_NAME ATTRIBUTE_NAME  DATA_TYPE	        TAR VECTOR_INFO
   ---------- --------------- ------------------------ --- --------------------
   KM_MODEL   EMBEDDING	   VECTOR		         NO  VECTOR(10,FLOAT64)

7. You can check the model detail views for KM_MODEL model.

   SELECT model_name, view_name, view_type
   FROM USER_MINING_MODEL_VIEWS
   WHERE model_name='KM_MODEL’ ORDER BY view_name; 
    
   MODEL_NAME VIEW_NAME	     VIEW_TYPE
   ---------- --------------- ----------------------------------------
   KM_MODEL   DM$VAKM_MODEL   Clustering Attribute Statistics
   KM_MODEL   DM$VCKM_MODEL   k-Means Scoring Centroids
   KM_MODEL   DM$VDKM_MODEL   Clustering Description
   KM_MODEL   DM$VGKM_MODEL   Global Name-Value Pairs
   KM_MODEL   DM$VHKM_MODEL   Clustering Histograms
   KM_MODEL   DM$VNKM_MODEL   Normalization and Missing Value Handling
   KM_MODEL   DM$VRKM_MODEL   Clustering Rules
   KM_MODEL   DM$VSKM_MODEL   Computed Settings
   KM_MODEL   DM$VWKM_MODEL   Model Build Alerts 

8. You can also view each vector dimension as a predictor from the model details.

   SELECT * FROM(SELECT cluster_id, attribute_name, attribute_subname, 
          mean, variance, mode_value 
   FROM DM$VAKM_MODEL ORDER BY cluster_id, attribute_name,attribute_subname) 
   
   CLUSTER_ID ATTRIBUTE_NAME  ATTRIBUTE_SUBNAME    MEAN	        VARIANCE       MODE_VALUE
   ---------- --------------- -------------------- ------------- ------------- --------------------
            1 EMBEDDING	      DM$$VEC1             28.9538       3.4382
            2 EMBEDDING	      DM$$VEC1             27.9580       5.5661
            3 EMBEDDING	      DM$$VEC1             29.9495       2.1698

9. Use scoring operators CLUSTER_ID and CLUSTER_PROBABILITY to find cluster assignments and probabilities for each record in pca_data.

   SELECT id, cluster_id(km_model using *) cluster_id, cluster_probability(km_model using *)probability FROM pca_data ORDER BY id;
   
   ID 		CLUSTER_ID      PROBABILITY
   ----------      ----------      -----------
   1                1              .617
   2                2              .584
   3                1              .579
   4                1              .605
   5                1              .621
   6                1              .642
   7                2              .598
   8                2              .614
   9                2              .650
   10               2              .618