使用 R 在两年间隔内对时间序列进行子集化

Question

我想使用 autoarima 开发模型，我有 1970 年到 2015 年的时间序列数据。我只想看看它如何使用一年数据（1970 年）进行预测，然后预测 2005 年，使用三年数据(1970, 1971,1972) 预测 2005 年或 2006 年，使用五年数据 (1970, 1971,1972,1973, 1974 (训练集)...并以两年间隔继续 (如 1,3,5, 7..years of data）直到 2015 年预测一年（测试集）。我在编码时遇到困难，但我在 Stackoverflow 上有以下重要链接，我无法修改它。这个子集问题让我发疯,请帮帮我。我从 StackOverflow 获得了这个很好的来源，但我如何根据我的情况定制它呢？ R multivariate one step ahead forecasts and accuracy

pred2<-numeric(0)
rmse2<-numeric(0)

for(i in 1:15){

DF.train2<-DF[DF$YEAR < 2000+i & DF$YEAR > 1989+i,]
DF.test2<-DF[DF$YEAR == 2000+i,]
lmod2 <- lm(TEMP ~ PRESSURE + RAINFALL, data = DF.train2)
pred2[i]<- predict(lmod2, newdata = DF.test2)
rmse2[i]<-sqrt(mean((DF.test2$TEMP-pred2[i])^2))
} 

pred2
rmse2  
mean(rmse2) 

Answer 1

考虑在两对火车年份和测试年份之间使用Map（包装到mapply）的元素循环进行子集化：

年份值

test_yrs <- lapply(c(2005:2015), function(i) c(2005:i) )

train_yrs <- lapply(seq(1,(2015-1970), by=2), function(i) 
     c(1970:2004)[c(1:i)] 
)[seq_along(test_yrs)]


train_yrs    

# [[1]]
# [1] 1970
# 
# [[2]]
# [1] 1970 1971 1972
# 
# [[3]]
# [1] 1970 1971 1972 1973 1974
# 
# [[4]]
# [1] 1970 1971 1972 1973 1974 1975 1976
# 
# [[5]]
# [1] 1970 1971 1972 1973 1974 1975 1976 1977 1978
# 
# [[6]]
# [1] 1970 1971 1972 1973 1974 1975 1976 1977 1978 1979 1980
# 
# [[7]]
# [1] 1970 1971 1972 1973 1974 1975 1976 1977 1978 1979 1980 1981 1982
# 
# [[8]]
# [1] 1970 1971 1972 1973 1974 1975 1976 1977 1978 1979 1980 1981 1982 1983 1984
# 
# [[9]]
# [1] 1970 1971 1972 1973 1974 1975 1976 1977 1978 1979 1980 1981 1982 1983 1984 1985 1986
# 
# [[10]]
# [1] 1970 1971 1972 1973 1974 1975 1976 1977 1978 1979 1980 1981 1982 1983 1984 1985 1986 1987 1988
# 
# [[11]]
# [1] 1970 1971 1972 1973 1974 1975 1976 1977 1978 1979 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990


test_yrs

# [[1]]
# [1] 2005
# 
# [[2]]
# [1] 2005 2006
# 
# [[3]]
# [1] 2005 2006 2007
# 
# [[4]]
# [1] 2005 2006 2007 2008
# 
# [[5]]
# [1] 2005 2006 2007 2008 2009
# 
# [[6]]
# [1] 2005 2006 2007 2008 2009 2010
# 
# [[7]]
# [1] 2005 2006 2007 2008 2009 2010 2011
# 
# [[8]]
# [1] 2005 2006 2007 2008 2009 2010 2011 2012
# 
# [[9]]
# [1] 2005 2006 2007 2008 2009 2010 2011 2012 2013
# 
# [[10]]
# [1] 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014
# 
# [[11]]
# [1] 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015

建模

model_func <- function(train_yrs, test_yrs) {

   train_df <- df[df$Year %in% train_yrs,]
   test_df <- df[df$Year %in% test_yrs,]

   mod <- lm(DepVar ~ IndepVar1 + IndepVar2 + IndepVar3, data = train_df)

   pred <- predict(mod, newdata = DF.test2),
   rmse <- sqrt(mean((test_df$DepVar - pred)^2))

   return(list(pred = pred, rmse = rmse))
}

results_list <- Map(model_func, train_yrs, test_yrs)

# ALL ELEMENTS
results_list

# ALL pred ELEMENTS
lapply(results_list, `[`, "pred")

# ALL rmse ELEMENTS
lapply(results_list, `[`, "rmse")    

# INDIVIDUAL ELEMENTS
results_list[[1]]$pred
results_list[[1]]$rmse

results_list[[2]]$pred
results_list[[2]]$rmse
...
results_list[[11]]$pred
results_list[[11]]$rmse

---

要跨多个数据帧运行上述进程，请将进程包装到一个方法中，并在数据帧列表上使用lapply 调用它。结果将嵌套在顶层下

proc_df <- function(df) {
    # NEW PARAM
    model_func <- function(train_yrs, test_yrs, df) {    
       ...
    }

    # ADDED ARG
    results_list <- Map(model_func, train_yrs, test_yrs, MoreArgs=list(df))

    return(results_list)
}

dfs <- list(df1, df2, df3)
df_results_list <- lapply(dfs, proc_df)

# ALL ELEMENTS BY DATA FRAME
df_results_list[[1]]
df_results_list[[2]]
df_results_list[[3]]