进行预测

R 中的回归入门

Richie Cotton

Data Evangelist at DataCamp

鱼类数据集：鲂鱼

bream <- fish %>% 
  filter(species == "Bream")

species	length_cm	mass_g
Bream	23.2	242
Bream	24.0	290
Bream	23.9	340
Bream	26.3	363
Bream	26.5	430
...	...	...

绘制质量与长度

ggplot(bream, aes(length_cm, mass_g)) +
  geom_point() +
  geom_smooth(method = "lm", se = FALSE)

鲂鱼质量与长度的散点图，带线性趋势线。各点均接近趋势线。

运行模型

mdl_mass_vs_length <- lm(mass_g ~ length_cm, data = bream)

Call:
lm(formula = mass_g ~ length_cm, data = bream)

Coefficients:
(Intercept)    length_cm  
   -1035.35        54.55

用于预测的自变量数据

如果将自变量设为这些值，
因变量会是多少？

library(dplyr)
explanatory_data <- tibble(length_cm = 20:40)

调用 predict()

library(tibble)
explanatory_data <- tibble(length_cm = 20:40)

predict(mdl_mass_vs_length, explanatory_data)

         1          2          3          4          5          6 
  55.65205  110.20203  164.75202  219.30200  273.85198  328.40196 
         7          8          9         10         11         12 
 382.95194  437.50192  492.05190  546.60188  601.15186  655.70184 
        13         14         15         16         17         18 
 710.25182  764.80181  819.35179  873.90177  928.45175  983.00173 
        19         20         21 
1037.55171 1092.10169 1146.65167

在数据框内预测

library(dplyr)
explanatory_data <- tibble(length_cm = 20:40)

prediction_data <- explanatory_data %>% 
  mutate(
    mass_g = predict(
      mdl_mass_vs_length, explanatory_data
    )
  )

# A tibble: 21 x 2
   length_cm mass_g
       <int>  <dbl>
 1        20   55.7
 2        21  110. 
 3        22  165. 
 4        23  219. 
 5        24  274. 
 6        25  328. 
 7        26  383. 
 8        27  438. 
 9        28  492. 
10        29  547. 
# ... with 11 more rows

展示预测结果

ggplot(bream, aes(length_cm, mass_g)) +
  geom_point() +
  geom_smooth(method = "lm", se = FALSE) +
  geom_point(
    data = prediction_data, 
    color = "blue"
  )

鲂鱼质量与长度的散点图，带线性趋势线。图上标注了用 predict() 计算的点，这些点完全落在趋势线上。

外推

"外推"指在观测数据范围之外进行预测。

explanatory_little_bream <- tibble(length_cm = 10)
explanatory_little_bream %>% 
  mutate(
    mass_g = predict(
      mdl_mass_vs_length, explanatory_little_bream
    )
  )

# A tibble: 1 x 2
  length_cm mass_g
      <dbl>  <dbl>
1        10  -490.

开始练习吧！

R 中的回归入门