Question

我很乐意为stan数据块创建大型的预测矩阵。

我想使用以下数据中的变量w_1至w_K作为模型中的预测变量“矩阵” real<lower=0> weights[N, W];。 K=W是变量权重（权重的列）的数量，N是观察数（权重的行），因此K和N是{{1} }。

下面的

int适用于几列（例如my current approach），但是我还有更多K=10列，因此，鉴于下面的数据，我需要一个函数来提供有效的和可扩展的方法来做到这一点：

K>100

我从tidybayes探索了#for the desired data block dat1 <- list (N = N, ncases = ncases, A = A, B = B, id = id, P = imput, nn = nn, W = 10, weights = cbind(w_1, w_2, w_3, w_4, w_5, w_6, w_7, w_8, w_9, w_10))，但看不到如何使用它来完成我想要的compose_data。因此，我们将不胜感激。

#sample数据

desired data block

＃我目前的做法

dat <- data.frame(
id = c(1, 1, 1, 1, 1, 2, 2, 2, 2, 2, 3, 3, 3, 3, 3, 4, 4, 4, 4, 4),
imput = c(1, 2, 3, 4, 5, 1, 2, 3, 4, 5, 1, 2, 3, 4, 5, 1, 2, 3, 4, 5),
A = c(1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0), 
B = c(1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0),
Pass = c(278, 278, 278, 278, 278, 100, 100, 100, 100, 100, 153, 153, 153, 153, 153, 79, 79, 79, 79, 79), 
Fail = c(740, 743, 742, 743, 740, 7581, 7581, 7581, 7581, 7581, 1231, 1232, 1235, 1235, 1232, 1731, 1732, 1731, 1731, 1731), 
W_1= c(4, 3, 4, 3, 3, 1, 2, 1, 2, 1, 12, 12, 11, 12, 12, 3, 5, 3, 3, 3),
W_2= c(3, 3, 3, 3, 3, 1, 1, 1, 1, 1, 12, 12, 12, 12, 12, 3, 3, 3, 3, 3),
W_3= c(4, 3, 3, 3, 3, 1, 2, 1, 1, 1, 12, 12, 11, 12, 12, 3, 3, 3, 3, 3),
W_4= c(3, 3, 4, 3, 3, 1, 1, 1, 2, 1, 12, 12, 13, 12, 12, 3, 2, 3, 3, 3),
W_5= c(3, 3, 3, 3, 3, 1, 0, 1, 1, 1, 12, 12, 12, 12, 12, 3, 3, 3, 3, 3),
W_6= c(4, 3, 3, 3, 3, 1, 1, 1, 1, 1, 12, 12, 12, 12, 12, 3, 3, 3, 3, 3),
W_7= c(3, 3, 3, 3, 3, 1, 1, 1, 1, 1, 12, 12, 12, 12, 12, 3, 3, 3, 3, 3),
W_8= c(3, 3, 3, 3, 3, 1, 1, 1, 1, 1, 15, 12, 12, 12, 12, 3, 3, 3, 3, 3),
W_9= c(3, 3, 3, 4, 3, 1, 1, 1, 1, 1, 12, 12, 12, 12, 12, 2, 3, 3, 3, 3),
W_10= c(3, 3, 4, 3, 3, 1, 1, 1, 1, 1, 12, 10, 12, 12, 12, 3, 3, 3, 3, 3)
      )

＃当前数据块

N <- nrow(dat)
ncases <- dat$Pass
nn <- dat$Fail + dat$Pass
A <- dat$A
B <- dat$B
id <- dat$id
imput <- dat$imput
w_1 <- dat$W_1
w_2 <- dat$W_2
w_3 <- dat$W_3
w_4 <- dat$W_4
w_5 <- dat$W_5
w_6 <- dat$W_6
w_7 <- dat$W_7
w_8 <- dat$W_8
w_9 <- dat$W_9
w_10 <- dat$W_10

#for current data block
    dat_list <-dat %>%compose_data(.n_name = n_prefix("N"))

#for desired data block
    dat1 <- list (N = N, 
              ncases = ncases, A = A, B = B, id = id, P = imput, nn = nn, W = 10,
              weights = cbind(w_1, w_2, w_3, w_4, w_5, w_6, w_7, w_8, w_9, w_10))

＃所需的数据块

    data{
    int N;                                    // number of observations
    int ncases[N];                        
    int A[N];                                 
    int B[N];                                
    int nn[N];                               
    int id[N];                                
    real<lower=0> w_1[N];                     // variable w_1
    real<lower=0> w_2[N];                     // variable w_2       
    real<lower=0> w_3[N];                     // variable w_3      
    real<lower=0> w_4[N];                     // variable w_4       
    real<lower=0> w_5[N];                     // variable w_5       
    real<lower=0> w_6[N];                     // variable w_6       
    real<lower=0> w_7[N];                     // variable w_7       
    real<lower=0> w_8[N];                     // variable w_8       
    real<lower=0> w_9[N];                     // variable w_9       
    real<lower=0> w_10[N];                    // variable w_10
    }

此问题也已发布here。预先感谢您的帮助。

Answer 1

如果dat中的所有预测变量列均以W_开头，那么我认为这应该可以解决问题：

w.matrix = as.matrix(dat[,grepl("^W_", colnames(dat))])
dat1 <- list (N = N, ncases = ncases, A = A, B = B, id = id, P = imput, nn = nn,
    W = ncol(w.matrix), weights = w.matrix)

如何有效地为stan数据块指定大型预测矩阵

1 个答案: