Demo Script: Naive Bayes (works fine on Greenplum)

examples/demo/naive-bayes.sql

@@ -4,30 +4,93 @@
 \qecho 
 \qecho === Refresh table: bayes ========================================
 
-DROP TABLE bayes;
-CREATE TABLE bayes (id INT, class INT, attributes INT[]) distributed by (ID);
-INSERT INTO bayes VALUES(   1 ,     1 , array[1,2,3]);
-INSERT INTO bayes VALUES(   3 ,     1 , array[1,4,3]);
-INSERT INTO bayes VALUES(   5 ,     2 , array[0,2,2]);
-INSERT INTO bayes VALUES(   2 ,     1 , array[1,2,1]);
-INSERT INTO bayes VALUES(   4 ,     2 , array[1,2,2]);
-INSERT INTO bayes VALUES(   6 ,     2 , array[0,1,3]);
+SET client_min_messages = error;
+
+DROP TABLE IF EXISTS bayes CASCADE;
+DROP TABLE IF EXISTS toclassify CASCADE;
+
+CREATE TABLE bayes
+(
+	id integer NOT NULL,
+	class INTEGER,
+	attributes INTEGER[],
+	CONSTRAINT pk_bayes PRIMARY KEY (id)
+);
+
+COPY bayes (id, class, attributes) FROM stdin;
+1	1	{1, 2, 3}
+3	1	{1, 4, 3}
+5	2	{0, 2, 2}
+2	1	{1, 2, 1}
+4	2	{1, 2, 2}
+6	2	{0, 1, 3}
+\.
+
+
+\qecho === Show training data ==========================================
+
+SELECT * FROM bayes;
+
+
+\qecho === Refresh table: toclassify ===================================
+
+CREATE TABLE toclassify
+(
+	id SERIAL NOT NULL,
+	attributes INTEGER[],
+	CONSTRAINT pk_toclassify PRIMARY KEY (id)
+);
+
+COPY toclassify (attributes) FROM stdin;
+{0, 2, 1}
+{1, 2, 3}
+\.
+
+
+\qecho === Show Data we want to run Naive Bayes classification on ======
+
+SELECT * FROM toclassify;
 
-SET CLIENT_MIN_MESSAGES=WARNING;
 
 \qecho === Precompute feature probabilities and class priors ===========
-DROP TABLE IF EXISTS nb_class_priors;
-DROP TABLE IF EXISTS nb_feature_probs;
+
+DROP TABLE IF EXISTS nb_class_priors CASCADE;
+DROP TABLE IF EXISTS nb_feature_probs CASCADE;
+
 SELECT madlib.create_nb_prepared_data_tables(
-     'bayes', 'class', 'attributes', 3,
-     'nb_feature_probs', 'nb_class_priors');
+    'bayes', 'class', 'attributes', 3,
+    'nb_feature_probs', 'nb_class_priors');
+
+
+\qecho === Show feature probabilities and class priors =================
+
+SELECT * FROM nb_feature_probs ORDER BY class, attr, value;
 
-select * from nb_class_priors;
+SELECT * FROM nb_class_priors ORDER BY class;
 
-select * from toclassify;
 
 \qecho === Run Naive Bayes =============================================
-select madlib.create_nb_classify_view(
+
+DROP VIEW IF EXISTS nb_classify_view_fast;
+
+SELECT madlib.create_nb_classify_view(
       'nb_feature_probs', 'nb_class_priors',
       'toclassify', 'id', 'attributes', 3,
       'nb_classify_view_fast');
+
+SELECT * from nb_classify_view_fast;
+
+
+\qecho === Look at the probabilities for each class
+\qecho === (Note we use Laplacian Smoothing):
+
+DROP VIEW IF EXISTS nb_probs_view_fast;
+
+SELECT madlib.create_nb_probs_view(
+    'nb_feature_probs', 'nb_class_priors',
+    'toclassify', 'id', 'attributes', 3,
+    'nb_probs_view_fast');
+
+SELECT * FROM nb_probs_view_fast;
+
+RESET client_min_messages;