該數據集包含來自西班牙的 7500 種不同類型的紅葡萄酒，具有 11 個特徵，描述了它們的價格、評級，甚至一些風味描述，目的為預測價格或品質。[1]
數據包含11個類型:
1. winery: 酒莊名稱
2. wine: 葡萄酒名稱
3. year: 葡萄收穫年分
4. rating: 葡萄酒的平均評分
5. num_reviews: 評論數量。
6. country: 原產地，即西班牙
7. region: 葡萄酒產區
8. price: 價格
9. type: 葡萄酒品種。
10. body: 酒體評分，代表口中葡萄酒的豐富度和重量。
11. acidity: 酸度評分，代表葡萄酒的「皺褶」或酸味。
data在kaggle上Spanish Wine Quality Dataset page上取得。
Index: 6070 entries, 0 to 7499
Data columns (total 11 columns):
# Column Non-Null Count Dtype
--- ------ -------------- -----
0 winery 6070 non-null object
1 wine 6070 non-null object
2 year 6070 non-null int64
3 rating 6070 non-null float64
4 num_reviews 6070 non-null int64
5 country 6070 non-null object
6 region 6070 non-null object
7 price 6070 non-null float64
8 type 6070 non-null object
9 body 6070 non-null float64
10 acidity 6070 non-null float64
dtypes: float64(4), int64(2), object(5)
memory usage: 569.1+ KB
| winery | wine | year | rating | num_reviews | region | price | type | body | acidity | |
|---|---|---|---|---|---|---|---|---|---|---|
| 0 | Teso La Monja | Tinto | 2013 | 4.9 | 58 | Toro | 995.00 | Toro Red | 5.0 | 3.0 |
| 1 | Artadi | Vina El Pison | 2018 | 4.9 | 31 | Vino de Espana | 313.50 | Tempranillo | 4.0 | 2.0 |
| 2 | Vega Sicilia | Unico | 2009 | 4.8 | 1793 | Ribera del Duero | 324.95 | Ribera Del Duero Red | 5.0 | 3.0 |
| 3 | Vega Sicilia | Unico | 1999 | 4.8 | 1705 | Ribera del Duero | 692.96 | Ribera Del Duero Red | 5.0 | 3.0 |
| 4 | Vega Sicilia | Unico | 1996 | 4.8 | 1309 | Ribera del Duero | 778.06 | Ribera Del Duero Red | 5.0 | 3.0 |
summary: strong relation between price and rating, and specific type of wine will affect price.
summary: spefic wine have high price, and next step will be finding the model to predict the price.
Index: 6070 entries, 0 to 7499
Data columns (total 10 columns):
# Column Non-Null Count Dtype
--- ------ -------------- -----
0 winery 6070 non-null int32
1 wine 6070 non-null int32
2 year 6070 non-null int64
3 rating 6070 non-null float64
4 num_reviews 6070 non-null int64
5 region 6070 non-null int32
6 price 6070 non-null float64
7 type 6070 non-null int32
8 body 6070 non-null float64
9 acidity 6070 non-null float64
dtypes: float64(4), int32(4), int64(2)
memory usage: 426.8 KB
{'LinearRegression': 0.3997098329164317,
'Lasso': -0.0002825894551166108,
'Ridge': 0.3997079007354649,
'BayesianRidge': 0.3996679858749155,
'DecisionTreeRegressor': 0.3963314565148496,
'LinearSVR': 0.2089817504376078,
'KNeighborsRegressor': 0.5949966180649249,
'RandomForestRegressor': 0.7376611282475856}
summary: KNeighborsRegressor and RandomForestRegressor have higher r2 factor, hence we use these two models to generate std price vs. index plots.
index y_prediction y_actual
0 1764 -0.295041 -0.369379
1 1821 -0.178520 -0.353428
2 1835 -0.256382 -0.348233
3 1595 -0.226673 -0.346843
4 1731 -0.208753 -0.345574
... ... ... ...
1209 42 5.391731 7.383273
1210 100 7.117485 7.882926
1211 595 3.849200 7.892109
1212 199 8.146660 9.494925
1213 332 6.879295 9.622104
[1214 rows x 3 columns]
index y_prediction y_actual
0 1764 -0.251923 -0.369379
1 1947 -0.257671 -0.364968
2 1029 -0.267733 -0.359531
3 1805 -0.183921 -0.355422
4 1821 -0.184619 -0.353428
... ... ... ...
1209 595 0.269403 7.892109
1210 94 13.041085 7.957844
1211 97 4.603715 8.655424
1212 188 6.941651 10.383123
1213 343 7.828997 16.207611
[1214 rows x 3 columns]
首先經由sort_values(by='y_actual')和reset_index()兩指令使得x軸呈現標準化的low price到high price，y軸則為deviation。可見KNeighborsRegressor model 和 RandomForestRegressor model在index 600下都能有效、正確的預測標準化後的wine price。當index超過800甚至到1000以上時兩model得到的prediction value與actual value相差很大，說明此時model無法預測high price下的actual price。其背後的原因為：從box chart中能發現大部分的wine都屬於low price，當資料量夠大時能得到較準確預測pricce的模型。而high price的wine數量非常少，不容易生成準確的model
因此，我們能用這兩個model來預測特定type的wine的價格(特別在價格不高時)。此練習使用了各式的model，其各自的模型原理還需要再細看。