Back
Close
  • 19

Learning Opportunities

This puzzle can be solved using the following concepts. Practice using these concepts and improve your skills.

Statement

 Goal

There is a pile of data that needs to be modelled, things like motion of the stars, stock price projections, and your chances of getting a date next Wednesday.

Unfortunately your graphing calculator broke and you've lost internet connection so you are going to have to calculate the best model for each data set.

Sometimes data is best modelled as a straight line, a parabola, a sine wave, or an exponential function. You want to know which model best describes the data.

Here are the models you are considering:
LINEAR f(x) = a*x + b
PARABOLA f(x) = a*x^2 + b*x + c
SINE f(x) = a*sin(b*x+c) + d
EXPONENTIAL f(x) = abs(a)^(x+b) + c

You measure how good a function is at describing the data by measuring the square of its error, with a lower value being better.

For example, with the LINEAR model f(x) = 1*x+1, and the data set (0,1.5) (1,2) (2,3) (3,5) we can calculate the total squared error like so
(0,1.5) : f(0) = 0+1 = 1.  error = (1.5-f(0))^2 = (1.5-1)^2 = 0.25
(1,2.0) : f(1) = 1+1 = 2. error = (2.0-f(1))^2 = (2.0-2)^2 = 0.0
(2,3.0) : f(2) = 2+1 = 3. error = (3.0-f(2))^2 = (3.0-3)^2 = 0.0
(3,5.0) : f(3) = 3+1 = 4. error = (5.0-f(3))^2 = (4.0-5)^2 = 1.0
Total = 0.25 + 0 + 0 + 1.0 = 1.25


You don't have the values a,b,c,d. You must find the best values for each model that has the lowest square error using something like gradient descent, simulated annealing, or some other sampling technique.

Any good model should 'predict' unseen data, so you have divided your data into training and test.
Use the training data to find the best a,b,c,d. Use the test data to see how well it does as a predictor.

The square error of the test data is what you will use to pick the model.

GOAL Pick the model with the lowest square error on the test data


REFERENCES
Intro to machine learning (specifically function fitting) https://www.youtube.com/watch?v=wUPFOFbCeFI
From Blokops https://www.cs.utep.edu/ofuentes/cs4361/fall19/grad_desc.pdf

HINTS
Use rot13.com to decode a hint
Hint 1 : Vg vf cbffvoyr gb svaq gur cnenzrgref sbe rnpu zbqry jvgu gra gubhfnaq genvavat fgrcf
Hint 2 : Znxvat lbhe bja grfg qngn rafher gung lbhe nytbevguz pna svg n zbqry gb qngn pbeerpgyl
Hint 3 : Gb trarengr grfg qngn hfr gur zbqry jvgu enaqbz cnenzrgref naq gur nqq n fznyy nzbhag bs reebe gb gur bhgchg
Hint 4 : Orpnhfr gurer vf fbzr punapr vaibyirq va svaqvat gur orfg inyhrf sbe rnpu zbqry, lbh znl unir gb er-fhozvg bapr be gjvpr. N tbbq fbyhgvba fubhyq abg arrq zber guna bar be gjb fhozvgf.
Input
Line 1: T The training set. Space separated list of floating point x y pairs like : 0.0 1.5 1.0 2.0 2.0 3.0 3.0 5.0
Line 2: U The test set. Space separated list of floating point x y pairs like : 4.0 5.0 5.0 4.8
Output
One of LINEAR, PARABOLA, SINE, or EXPONENTIAL
Constraints
-1e20 < x,y < 1e20
The training and test set will have at least 4 sets of points (8 numbers) and at most 20 sets of points (40 numbers)
-20 <= a,b,c,d <= 20

Except for exponential where 0 <= a <= 4
Example
Input
1 -5.4396 3 -1.7278 5 1.3968 7 5.0468 9 8.6818 11 11.9518 13 15.7351 15 19.2545 17 22.6826 19 25.6735
0 -7.1129 2 -3.6857 4 -0.2995 6 3.4321 8 6.4681 10 10.072 12 13.9492 14 17.09 16 20.8774 18 23.7027 20 28.4941
Output
LINEAR

A higher resolution is required to access the IDE