The distances given aren\u2019t correct. And, as a result, there\u2019s usually several attempts by various people to figure out where the year end function will be in advance.<\/p>\n
I thought I\u2019d join in this year, using some machine learning on those distances.<\/p>\n
Now, I should mention that this is a very poor use for ML. Mainly because of a lack of data. I should have hundreds of data points for a decent prediction. I have 2 or 3 data points, for 4 different locations. Still, it\u2019s what I have to work with.<\/p>\n
First, the starting data. The distances for this year are:<\/p>\n
- \n
- Clearwater mall: 63 KM<\/li>\n
- Centurion: 56 KM<\/li>\n
- Fourways: 43 KM<\/li>\n
- HQ: 20 KM<\/li>\n
- Cape Town: 1447 KM<\/li>\n<\/ul>\n
I\u2019m going to ignore Cape Town for training, as it only had a distance previously specified in 2015, and so I only have one piece of data.<\/p>\n
Plotting this on a map makes it clear that the distances have been \u2018massaged\u2019 (I\u2019m plotting \u2018as the bird flies\u2019, not driving distance for ease of plotting, I\u2019ll use driving distances for the training)<\/p>\n
![\"yef_720\"](\"https:\/\/www.sqlinthewild.co.za\/wp-content\/uploads\/2017\/10\/yef_720_thumb.png\" "\\"yef_720\\"")
<\/a><\/p>\nLet\u2019s look at previous years.<\/p>\n
2016<\/strong><\/p>\n
Actual location: Seasons Sport and Spa (pin on the map below)<\/p>\n
Actual distances calculated with Google Maps, driving distance, shortest route.<\/p>\n
- \n
- Clearwater: Given distance – 80KM. Actual distance – 67KM<\/li>\n
- Centurion : Given distance – 88KM. Actual distance – 47KM<\/li>\n
- Fourways : Given distance – 68KM. Actual distance – 52KM<\/li>\n
- HQ: Given distance – 115KM. Actual distance – 75KM<\/li>\n<\/ul>\n
![\"YEF2016\"](\"https:\/\/www.sqlinthewild.co.za\/wp-content\/uploads\/2017\/10\/YEF2016_thumb.png\" "\\"YEF2016\\"")
<\/a><\/p>\n2015<\/strong><\/p>\n
Actual location: Vaal River Country Lodge.<\/p>\n
Actual distances calculated with Google Maps, driving distance, shortest route.<\/p>\n
- \n
- Clearwater Mall: Given distance –\u00a0 51KM. Actual distance \u2013 79KM<\/li>\n
- Centurion: Given distance – 110KM. Actual distance – 118KM<\/li>\n
- Fourways: Given distance – 89KM. Actual distance – 97KM<\/li>\n<\/ul>\n
![\"YEF2015\"](\"https:\/\/www.sqlinthewild.co.za\/wp-content\/uploads\/2017\/10\/YEF2015_thumb.png\" "\\"YEF2015\\"")
<\/a><\/p>\n2014<\/strong><\/p>\n
Actual location: Askari Game Lodge.<\/p>\n
Actual distances calculated with Google Maps, driving distance, shortest route.<\/p>\n
- \n
- Clearwater Mall: Given distance –\u00a0 52KM. Actual distance \u2013 79KM<\/li>\n
- HQ: Given distance – 90KM. Actual distance \u2013 118KM<\/li>\n<\/ul>\n
![\"YEF2014\"](\"https:\/\/www.sqlinthewild.co.za\/wp-content\/uploads\/2017\/10\/YEF2014_thumb.png\" "\\"YEF2014\\"")
<\/a><\/p>\nWith that, I have the following training data:<\/p>\n
\n\n
\n Location<\/td>\n Given Distance<\/td>\n Error in Distance (Given – Actual)<\/td>\n<\/tr>\n \n Clearwater Mall<\/td>\n 80<\/td>\n 13<\/td>\n<\/tr>\n \n Clearwater Mall<\/td>\n 51<\/td>\n -28<\/td>\n<\/tr>\n \n Clearwater Mall<\/td>\n 52<\/td>\n -27<\/td>\n<\/tr>\n \n Centurion<\/td>\n 88<\/td>\n 41<\/td>\n<\/tr>\n \n Centurion<\/td>\n 110<\/td>\n -8<\/td>\n<\/tr>\n \n Fourways<\/td>\n 68<\/td>\n -6<\/td>\n<\/tr>\n \n Fourways<\/td>\n 89<\/td>\n -8<\/td>\n<\/tr>\n \n HQ<\/td>\n 115<\/td>\n 40<\/td>\n<\/tr>\n \n HQ<\/td>\n 90<\/td>\n -28<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n Now to stick those into a linear regression and see if I can predict the error on this year\u2019s measurements.<\/p>\n
I need to mention that with so little data, the accuracy of the linear regression is going to be very low. I\u2019m as likely to get the correct results from linear regression as I am to get correct results from rolling a couple of d20s.<\/p>\n
That said, onwards to untrustworthy results.<\/p>\n
Once the starting values are loaded into R, creating a simple model is as easy as<\/p>\n
m <- lm(Error ~ Location + Distance, data=YEF)<\/pre>\n
Then load up this year\u2019s values into another data frame, and predict.<\/p>\n
predict(m, YEFPredict)<\/pre>\n
The errors come out as:<\/p>\n
- \n
- Clearwater Mall: -12<\/li>\n
- Centurion: -18<\/li>\n
- Fourways: -35<\/li>\n
- HQ: -60<\/li>\n<\/ul>\n
Giving final estimated distances (Given – Error) as<\/p>\n
- \n
- Clearwater Mall: 75KM<\/li>\n
- Centurion: 74KM<\/li>\n
- Fourways: 78KM<\/li>\n
- HQ: 80KM<\/li>\n<\/ul>\n
![\"yef_720\"](\"https:\/\/www.sqlinthewild.co.za\/wp-content\/uploads\/2017\/10\/yef_720.png\" "\\"yef_720\\"")
<\/a><\/p>\n