複雑なのが常によくない(A simple methods better than complex)

Photo by JESHOOTS.COM on Unsplash

Bordeaux Wine Vintage Quality and The Weather

Eli Kitagawa

eli@gracekyoto.com

"Nor let thy vineyard bend toward the sun when setting"

Original Analysis from liquidasset

http://www.liquidasset.com/orley.htm

Take 5 values from the dataset
Year Price WinterRain AGST HarvestRain Age FrancePop
0 1952 7.4950 600 17.1167 160 31 43183.569
1 1953 8.0393 690 16.7333 80 30 43495.030
2 1955 7.6858 502 17.1500 130 28 44217.857
3 1957 6.9845 420 16.1333 110 26 45152.252
4 1958 6.7772 582 16.4167 187 25 45653.805
Year Price WinterRain AGST HarvestRain Age FrancePop
count 25.000000 25.000000 25.000000 25.000000 25.000000 25.000000 25.000000
mean 1965.800000 7.067224 605.280000 16.509336 148.560000 17.200000 49694.436760
std 7.691987 0.650341 132.277965 0.675397 74.419464 7.691987 3665.270243
min 1952.000000 6.204900 376.000000 14.983300 38.000000 5.000000 43183.569000
25% 1960.000000 6.518800 536.000000 16.200000 89.000000 11.000000 46583.995000
50% 1966.000000 7.121100 600.000000 16.533300 130.000000 17.000000 50254.966000
75% 1972.000000 7.495000 697.000000 17.066700 187.000000 23.000000 52894.183000
max 1978.000000 8.493700 830.000000 17.650000 292.000000 31.000000 54602.193000

Single Value Linear Regression Models vs Price

Gradient Boosting and XGBoost Predict Perfect in Training Set

Surprised, Linear Regression did better analysis.

In this analysis, although, training data linear regression was not very good, at real prediction it works better than other newer algorithm.

Root Mean Square error of Linear Regression : 0.20247428671657527 
Root Mean Square error of GradientBoosting : 0.5837740667697905 
Root Mean Square error of XGBoost : 0.5107839576846731

REFERENCES

Data and Results for Wine Regressions as Reported in Chance Magazine by Ashenfelter, Ashmore and Lalonde

Haraszthy, A. Grape Culture, Wines, and Wine-making, New York: Harper & Bros., 1862.

Liquid Assets: The International Guide To Fine Wines, 169 Nassau St., Princeton, NJ 08542, various issues.

Penning-Rowsell, Edmund. The Wines of Bordeaux, Fifth Edition, San Francisco: The Wine Appreciation Guild, 1985.