Update music_by_gender_rank_example.sql

Author: ipeirotis

session7/music_by_gender_rank_example.sql

@@ -1,34 +1,59 @@
-DROP VIEW IF EXISTS popular_music;
-CREATE VIEW popular_music AS
+USE facebook;
+
+# We create first a table of popular music across everyone
+# For efficiency we only keep music liked by more than 10 people
+DROP TEMPORARY TABLE IF EXISTS popular_music;
+CREATE TEMPORARY TABLE popular_music AS
 	SELECT M.Music AS music, COUNT(M.ProfileID) AS cnt    
 	FROM FavoriteMusic M 
 	GROUP BY M.Music
+    HAVING cnt > 10
 	ORDER BY cnt DESC;
 
 
-DROP VIEW IF EXISTS popular_music_by_sex;
-CREATE VIEW popular_music_by_sex AS
+# We now calculate popularity of each music broken down by sex
+DROP TEMPORARY TABLE IF EXISTS popular_music_by_sex;
+CREATE TEMPORARY TABLE popular_music_by_sex AS
 	SELECT M.Music AS music, P.Sex AS gender, COUNT(P.ProfileID) AS cnt    
 	FROM FavoriteMusic M JOIN Profiles P ON P.ProfileID = M.ProfileID
 	WHERE P.Sex IS NOT NULL
 	GROUP BY M.Music, P.Sex 
 	ORDER BY cnt DESC;
 
-DROP VIEW IF EXISTS chart;
-CREATE VIEW chart AS 
-	SELECT music, gender, cnt,
-		RANK() OVER (PARTITION BY gender ORDER BY cnt DESC) AS music_rank 	
+# We will now create two tables with music rank, one per gender
+# In principle, we could also do a window OVER (PARTITION BY gender ORDER BY cnt DESC)
+# However, MySQL has a bug that does not allow temporary tables to 
+# join with itself, so we end up creating one temp table for males and another for females
+DROP TEMPORARY TABLE IF EXISTS chart_male;
+CREATE TEMPORARY TABLE chart_male AS 
+	SELECT music, cnt,
+		RANK() OVER (ORDER BY cnt DESC) AS music_rank 	
 	FROM popular_music_by_sex
-	ORDER BY music_rank, gender;
-
-
+    WHERE gender = 'Male' 
+	ORDER BY music_rank;
 
+DROP TEMPORARY TABLE IF EXISTS chart_female;
+CREATE TEMPORARY TABLE chart_female AS 
+	SELECT music, cnt,
+		RANK() OVER (ORDER BY cnt DESC) AS music_rank 	
+	FROM popular_music_by_sex
+    WHERE gender = 'Female'
+	ORDER BY music_rank;
+
+# Finally we bring everything together.
+# Note that we start with popular_music as a reference table
+# and we left join the other two tables, as there is no guarantee
+# that we will encounter a music in both males and females tables
+# To estimate the difference between males and females we take the
+# log of the rank, and then the difference; the reason we do that 
+# is because a difference of 5 between No1 and No6 is very different
+# than a difference of 5 between No605 and No610. With LOG we kind
+# of estimate difference in "orders of magnitude"
 SELECT S.music, S.cnt, 
 	M.cnt AS male_cnt, M.music_rank AS male_rank,
-	F.cnt AS female_cnt, F.music_rank AS female_rank
+	F.cnt AS female_cnt, F.music_rank AS female_rank,
+    ROUND(-LOG(F.music_rank / M.music_rank),2) AS diff_females
 FROM popular_music S 
-	LEFT JOIN chart M ON (S.music = M.music AND M.gender='Male')
-	LEFT JOIN chart F ON (S.music = F.music AND F.gender='Female');
-    
-
-
+	LEFT JOIN chart_female M ON (S.music = M.music)
+	LEFT JOIN chart_male F ON (S.music = F.music)
+ORDER BY diff_females DESC;