使用`interp2 {pracma}`进行双线性插值时出错;任何更好的2D插值方法?

时间:2016-12-08 05:13:57

标签: r interpolation bilinear-interpolation

我正在尝试对名为vol_coarse的表进行二维插值。

install.packages("install.load")
install.load::load_package("pracma", "data.table")

vol_coarse <- data.table(V1 = c(3 / 8, 1 / 2, 3 / 4, 1, 1 + 1 / 2, 2, 3, 6),
V2 = c(0.50, 0.59, 0.66, 0.71, 0.75, 0.78, 0.82, 0.87),
V3 = c(0.48, 0.57, 0.64, 0.69, 0.73, 0.76, 0.80, 0.85),
V4 = c(0.44, 0.53, 0.60, 0.65, 0.69, 0.72, 0.76, 0.81))
setnames(vol_coarse, c("Maximum size of aggregate (in)", "2.40", "2.60", "2.80"))

x <- vol_coarse[, 2][[1]]

y <- as.numeric(colnames(vol_coarse[, 2:ncol(vol_coarse)]))

z <- meshgrid(x, y)

xp <- 3 / 4

yp <- 2.70

interp2(x = x, y = y, Z = z, xp = xp, yp = yp, method = "linear")

这是返回的错误消息:

  

错误:is.numeric(Z)不为TRUE

我在?interp2中读到:

length(x) = nrow(Z) = 8 and length(y) = ncol(Z) = 3 must be satisfied.

如何创建一个8乘3的矩阵,以便我可以使用interp2

或者有更好的方法来执行此类插值吗?

谢谢。

1 个答案:

答案 0 :(得分:1)

如果我没错你,你想要:

x <- c(3 / 8, 1 / 2, 3 / 4, 1, 1 + 1 / 2, 2, 3, 6)  ## V1
y <- c(2.4, 2.6, 2.8)  ## column names
Z <- cbind(c(0.50, 0.59, 0.66, 0.71, 0.75, 0.78, 0.82, 0.87),  ## V2
           c(0.48, 0.57, 0.64, 0.69, 0.73, 0.76, 0.80, 0.85),  ## V3
           c(0.44, 0.53, 0.60, 0.65, 0.69, 0.72, 0.76, 0.81))  ## V4
xp <- 3 / 4
yp <- 2.70

您已经在网格上有一个定义良好的矩阵。例如,您可以通过以下方式调查您的3D数据:

persp(x, y, Z)
image(x, y, Z)
contour(x, y, Z)

我不推荐pracma,因为interp2函数有错误。我建议在interp.surface包中使用fields函数进行网格插值。

library(fields)
## the list MUST has name `x`, `y`, `x`!
## i.e., unnamed list `list(x, y, Z)` does not work!
interp.surface(list(x = x, y = y, z = Z), cbind(xp, yp))
# [1] 0.62
来自interp2

pracma不一致。手册说Z矩阵length(x)位于length(y),但该函数确实Z必须length(y) length(x)

## from manual

   Z: numeric ‘length(x)’-by-‘length(y)’ matrix.

## from source code of `interp2`

   lx <- length(x)
   ly <- length(y)
   if (ncol(Z) != lx || nrow(Z) != ly) 
       stop("Required: 'length(x) = ncol(Z)' and 'length(y) = nrow(Z)'.")

因此,为了使interp2有效,您必须传递Z的转置:

interp2(x, y, t(Z), xp, yp)
# [1] 0.62

或反转xy(以及xpyp也是如此!!):

interp2(y, x, Z, yp, xp)
# [1] 0.62

这与我们使用imagecontourpersp的方式非常不一致。