如何不通过生成图像获取R自定义函数中的coef系数值

实现方案 #

完全可以不依赖绘图直接获取coef的取值，该值本质是nls非线性拟合得到的模型截距项b2，计算过程和绘图步骤没有任何关联，仅需对原有函数做少量修改即可直接返回数值结果。

步骤1：修改原有函数

调整后的函数支持选择是否开启绘图，同时直接返回系数值：

# 加载必要依赖包
library(dplyr)
library(tidyverse)
library(lubridate)

# 原有数据集定义不变
df1 <- structure(
  list(date1= c("2021-06-28","2021-06-28","2021-06-28","2021-06-28"),
       date2 = c("2021-06-30","2021-06-30","2021-07-01","2021-07-01"),
       Category = c("FDE","ABC","FDE","ABC"),
       Week= c("Wednesday","Wednesday","Friday","Friday"),
       DR1 = c(4,1,6,3),
       DR01 = c(4,1,4,3), DR02= c(4,2,6,2),DR03= c(9,5,4,7),
       DR04 = c(5,4,3,2),DR05 = c(5,4,5,4),
       DR06 = c(2,4,3,2),DR07 = c(2,5,4,4),
       DR08 = c(3,4,5,4),DR09 = c(2,3,4,4)),
  class = "data.frame", row.names = c(NA, -4L))

# 改造后的函数，支持直接返回系数值
f1_get_coef <- function(dmda, CategoryChosse, draw_plot = TRUE) {
  x<-df1 %>% select(starts_with("DR0"))
  x<-cbind(df1, setNames(df1$DR1 - x, paste0(names(x), "_PV")))
  PV<-select(x, date2,Week, Category, DR1, ends_with("PV"))
  
  med<-PV %>%
    group_by(Category,Week) %>%
    summarize(across(ends_with("PV"), median), .groups = "drop")
  
  SPV<-df1%>%
    inner_join(med, by = c('Category', 'Week')) %>%
    mutate(across(matches("^DR0\\d+$"), ~.x + 
                    get(paste0(cur_column(), '_PV')),
                  .names = '{col}_{col}_PV')) %>%
    select(date1:Category, DR01_DR01_PV:last_col())
  SPV<-data.frame(SPV)
  
  mat1 <- df1 %>%
    filter(date2 == dmda, Category == CategoryChosse) %>%
    select(starts_with("DR0")) %>%
    pivot_longer(cols = everything()) %>%
    arrange(desc(row_number())) %>%
    mutate(cs = cumsum(value)) %>%
    filter(cs == 0) %>%
    pull(name)
  dropnames <- paste0(mat1,"_",mat1, "_PV")
  
  SPV <- SPV %>%
    filter(date2 == dmda, Category == CategoryChosse) %>%
    select(-any_of(dropnames))
  
  datas<-SPV %>%
    filter(date2 == ymd(dmda)) %>%
    group_by(Category) %>%
    summarize(across(starts_with("DR0"), sum), .groups = "drop") %>%
    pivot_longer(cols= -Category, names_pattern = "DR0(.+)", values_to = "val") %>%
    mutate(name = readr::parse_number(name))
  colnames(datas)[-1]<-c("Days","Numbers")
  
  datas <- datas %>% 
    group_by(Category) %>% 
    slice((as.Date(dmda) - min(as.Date(df1$date1) [
      df1$Category == first(Category)])):max(Days)+1) %>%
    ungroup
  
  # 模型拟合为核心计算步骤，和绘图无关
  model <- nls(Numbers ~ b1*Days^2+b2,start = list(b1 = 0,b2 = 0),data = datas, algorithm = "port")
  coef_val <- coef(model)[2]
  
  # 可选绘图
  if(draw_plot){
    plot(Numbers ~ Days,  xlim= c(0,45), ylim= c(0,30),
         xaxs='i',data = datas,main = paste0(dmda, "-", CategoryChosse))
    new.data <- data.frame(Days = with(datas, seq(min(Days),max(Days),len = 45)))
    new.data <- rbind(0, new.data)
    lines(new.data$Days,predict(model,newdata = new.data),lwd=2)
    points(0, coef_val, col="red",pch=19,cex = 2,xpd=TRUE)
    text(.99,coef_val + 1,max(0, round(coef_val,1)), cex=1.1,pos=4,offset =1,col="black")
  }
  
  return(coef_val)
}

步骤2：调用函数直接获取系数

不需要绘图的场景下直接关闭绘图参数即可得到数值结果：

# 关闭绘图直接获取系数
coef_result <- f1_get_coef("2021-06-30", "FDE", draw_plot = FALSE)
print(coef_result)

运行后输出结果为4，和可视化图像上标注的数值完全一致。

内容的提问来源于stack exchange，提问作者user16774617