Moveing Average and Weighted Moving Average

Author: ptyadana

Advanced SQL for Data Science - Time Series/03.Time Series Analysis/04.Moving Average.sql

@@ -0,0 +1,16 @@
+/********** Moving Average *************/
+
+/* 
+We want to know the hourly cpu utilization.
+So we can create a sliding windows to get the average cpu utilization of 1 hour.
+In hour case, each log is inserted per 5 minutes. So for 1 hour, we need to gather 12 logs (transactions).
+
+NOTE:
+OVER statements basically said order by the event time, then given the current row, go back 12 rows.
+With that set of date, apply the average function.
+*/
+
+SELECT
+	event_time, server_id,
+	AVG(cpu_utilization) OVER (ORDER BY event_time ROWS BETWEEN 12 PRECEDING AND CURRENT ROW) AS hourly_cpu_util
+FROM time_series.utilization;

Advanced SQL for Data Science - Time Series/03.Time Series Analysis/05.Weighted Moving Average.sql

@@ -0,0 +1,29 @@
+/********* Weighted Moving Average ************/
+
+/*
+The idea is you want to give MORE weight to the MORE RECENT events than the past events.
+
+We want to weight the average temp of last 3 days. so we need to sum up of those last 3 days.
+*/
+WITH daily_avg_temp AS (
+	SELECT
+		DATE_TRUNC('day', event_time) as event_date,
+		ROUND(AVG(temp_celcius),2) AS avg_temp
+	FROM time_series.location_temp
+	GROUP BY DATE_TRUNC('day', event_time)
+)
+SELECT
+	event_date, avg_temp,
+	(SELECT ROUND(avg_temp,2) * 0.5
+	 FROM daily_avg_temp d2
+	 WHERE d2.event_date = d1.event_date - INTERVAL '1' day
+	) + 
+	(SELECT ROUND(avg_temp,2) * 0.333
+	 FROM daily_avg_temp d3
+	 WHERE d3.event_date = d1.event_date - INTERVAL '2' day
+	) +
+	(SELECT ROUND(avg_temp,2) * 0.167
+	 FROM daily_avg_temp d4
+	 WHERE d4.event_date = d1.event_date - INTERVAL '3' day
+	) AS three_days_weighted_avg_temp
+FROM daily_avg_temp d1