versatica / TrendCalculator

An algorithm to ease off decreasing values based on a constant factor, an elapsed time and (probably) also on currently highest value seen

TrendCalculator

An algorithm to ease off decreasing values based on a constant factor, an elapsed time and (probably) also on currently highest value seen.

Overview

The TrendCalculator is provided with input values (via the update() method). Upon updated values, the TrendCalculator computes its new "current value".

• If the new given value is higher or equal than the current value, the current value must be set to it. This is, higher values must automatically update the current value.
• Otherwise, if the new given value is less than the current value, the algorithm must ease it off to produce a value slightly lower than the current value. And such a relaxed value must depend on the given "now" (in ms).

Expected Results

Let's assume we have trendA.

• Initially it's provided with a value of 1000.
• The current value of trendA must be 1000.
• After 1 second, trendA.Update() is called with value 500.
• The new current value (trendA.GetValue()) must be something lower than 1000. Let's say 950.
• After another second, trendA.Update() is called with value 500 again.
• The new current value (trendA.GetValue()) must be something lower than 950. Let's say 900.

Let's assume we have trendB.

• Initially it's provided with a value of 1000.
• The current value of trendB must be 1000.
• After 2 seconds, trendB.Update() is called with value 500.
• The new current value (trendB.GetValue()) must be something lower than 1000. Let's say 980.

This is, in both cases the initial values were the same (1000) and we have then called Update() on both with value 500. In the former, twice in 2 seconds. In the latter, once in 2 seconds.

The algorithm is supposed to produce the same current value in both cases. This is, it should not matter that Update() is called multiple times with the same value or just once.

Important: Those requirements may be wrong. And most probably this algorithm needs to keep a histogram of previous values and analyze its current trending in order to compute the new value.

Usage

Just run:

$ ./run-tests.sh

It will run some tests (see src/main.cpp file) that, right now, produce the following results:

$ ./run-tests.sh

>>> test 1:
    trend.Update(value:1000, now:0) [trend value: 1000]
    trend.Update(value:500, now:500) [trend value: 975]
    trend.Update(value:500, now:1000) [trend value: 950]
    trend.Update(value:500, now:1500) [trend value: 925]
    trend.Update(value:500, now:2000) [trend value: 900]
    trend.Update(value:500, now:2500) [trend value: 875]
    trend.Update(value:500, now:3000) [trend value: 850]
    trend.Update(value:500, now:3500) [trend value: 825]
    trend.Update(value:500, now:4000) [trend value: 800]
    trend.Update(value:500, now:4500) [trend value: 775]
    trend.Update(value:1000, now:5000) [trend value: 1000]

>>> test 2:
    trend.Update(value:1000, now:0) [trend value: 1000]
    trend.Update(value:500, now:1000) [trend value: 950]
    trend.Update(value:500, now:2000) [trend value: 900]
    trend.Update(value:500, now:3000) [trend value: 850]
    trend.Update(value:500, now:4000) [trend value: 800]
    trend.Update(value:1000, now:5000) [trend value: 1000]

>>> test 3:
    trend.Update(value:1000, now:0) [trend value: 1000]
    trend.Update(value:900, now:1000) [trend value: 950]
    trend.Update(value:800, now:2000) [trend value: 900]
    trend.Update(value:700, now:3000) [trend value: 850]
    trend.Update(value:600, now:4000) [trend value: 800]
    trend.Update(value:500, now:5000) [trend value: 750]
    trend.Update(value:400, now:6000) [trend value: 700]
    trend.Update(value:300, now:7000) [trend value: 650]
    trend.Update(value:200, now:8000) [trend value: 600]
    trend.Update(value:100, now:9000) [trend value: 550]
    trend.Update(value:100, now:10000) [trend value: 500]
    trend.Update(value:100, now:11000) [trend value: 450]
    trend.Update(value:100, now:12000) [trend value: 400]
    trend.Update(value:100, now:13000) [trend value: 350]
    trend.Update(value:100, now:14000) [trend value: 300]
    trend.Update(value:100, now:15000) [trend value: 250]
    trend.Update(value:100, now:16000) [trend value: 200]
    trend.Update(value:100, now:17000) [trend value: 150]
    trend.Update(value:100, now:18000) [trend value: 100]
    trend.Update(value:100, now:19000) [trend value: 100]
    trend.Update(value:100, now:20000) [trend value: 100]

>>> done