在 r 中读取二进制映射文件

Reading a binary map file in r

我正在尝试读取 R 中的二进制文件,其中包含一个简单的 360x180 值二维数组。作为参考,可以在此处找到二进制文件:

http://transcom.project.asu.edu/download/transcom03/smoothmap.fix.2.bin

这是此 .bin 的自述文件内容:

The file 'smoothmap.fix.2.bin' contains a single real, binary array dimensioned 360 x 180. The array contains the numbers 1 through 22, denoting each of the 22 basis functions in the TransCom 3 experiment. This file was written on an SGI Origin 2000 hosting UNIX.

我的代码:

to.read <- file("smoothmap.fix.2.bin", "rb")
raw.transcom <- readBin(to.read, integer(), n = 360*180, size = 4, endian = "big")
transcom <- matrix(raw.transcom, 180, 360, byrow = F)

现在 raw.transcom 只包含垃圾值:

unique(raw.transcom)
 [1]     259200          0 1101004800 1082130432 1092616192 1097859072 1100480512 1102053376 1086324736
[10] 1077936128 1101529088 1095761920 1096810496 1099956224 1091567616 1084227584 1090519040 1094713344
[19] 1099431936 1073741824 1093664768 1088421888 1065353216 1098907648

为什么会这样?

我已经看了一个小时了,我被难住了。试过字节序设置和 readBin 中的 'size',但这没有帮助。

我怎样才能正确读入这个文件?

好吧,我没有时间研究 "R" 的方法,但我确实可以访问 GDL 并找到 this,所以我拼凑起来:

Data  = read_binary('smoothmap.fix.2.bin',DATA_TYPE=4,ENDIAN='big');
Data = Data[1:64800]
Data = reform(Data,[360,180])

openw,unit,'testfile.dat',/get_lun
printf,unit,Data
free_lun,unit

并成功生成:http://rud.is/dl/testfile.dat.gz

如果你抓住它然后做:

x <- as.numeric(scan("testfile.dat.gz", "numeric"))

length(x)
## [1] 64800

table(x)
##   0    1    2    3    4    5    6    7    8    9   10   11   12   13   14   15   16   17   18   19   20   21   22 
## 7951 1643 1189  796  868 1688  864 2345 2487  509  733 1410 5144 2388 2433 4111 7617 2450 1671 2058 9161 2334 2950 

看起来它确实为您指定的定义提供了正确的值,您可以将其转换为矩阵。

请回来查看,因为我现在需要弄清楚如何在 R 中执行此操作:-)


更新

知道了!

我很高兴我找到了 IDL 代码来验证 R 结果。

x <- readBin("smoothmap.fix.2.bin", "raw", file.size("smoothmap.fix.2.bin"))
x <- x[-(1:4)]
x <- x[-((length(x)-3):length(x))]

table(readBin(rawConnection(x), "numeric", 360*180, 4, endian="big"))
##    0    1    2    3    4    5    6    7    8    9   10   11   12   13   14   15   16   17   18   19   20   21   22 
## 7951 1643 1189  796  868 1688  864 2345 2487  509  733 1410 5144 2388 2433 4111 7617 2450 1671 2058 9161 2334 2950 

理想情况下,我们会检查前 4 个字节和后 4 个字节是否相等,但这个 hack 应该能让你通过。


综合起来

添加了代码的验证位…

#' Read in a binary array, likely written with IDL
#' 
#' @param x path to file (auto-expanded & tested for existence)
#' @param n number of `float` elements to read in
#' @param endian endian-ness (default `big`)
#' @return numeric vector of length `n`
read_binary_float <- function(x, n, endian="big") {

  x <- normalizePath(path.expand(x))

  x <- readBin(con = x, what = "raw", n = file.size(x))

  first4 <- x[1:4] # extract front bits
  last4 <- x[(length(x)-3):length(x)] # extract back bits

  # convert both to long ints      

  f4c <- rawConnection(first4)
  on.exit(close(f4c), add=TRUE)
  f4 <- readBin(con = f4c, what = "integer", n = 1, size = 4L, endian=endian)

  l4c <- rawConnection(last4)      
  on.exit(close(l4c), add=TRUE)      
  l4 <- readBin(con = l4c, what = "integer", n = 1, size = 4L, endian=endian)

  # validation

  stopifnot(f4 == l4) # check front/back are equal
  stopifnot(f4 == n*4) # check if `n` matches expected record count

  # strip off front and back bits

  x <- x[-(1:4)]
  x <- x[-((length(x)-3):length(x))]

  # slurp it all in

  rc <- rawConnection(x)      
  on.exit(close(rc), add=TRUE)

  readBin(con = rc, what = "numeric", n = n, size = 4L, endian=endian)

}

快速示例:

library(magrittr)

read_binary_float("smoothmap.fix.2.bin", 360*180) %>% 
  matrix(nrow = 360, ncol = 180) %>% 
  image()

此文件似乎符合 Fortran "unformatted I/O" 规范:https://docs.oracle.com/cd/E19957-01/805-4939/6j4m0vnc4/index.html:这证实了

"# records" | record | record | … | record | "# records"

我们看到了。因此,该函数可以推广到支持的不仅仅是 float 转换:

read_binary_array <- function(x, type=c("byte", "integer", "float"), endian="big") {

  type <- match.arg(trimws(tolower(type)), c("byte", "integer", "float"))
  type_size <- unname(c("byte"=1, "integer"=4, "float"=4)[type])

  x <- normalizePath(path.expand(x))

  x <- readBin(con = x, what = "raw", n = file.size(x))

  first4 <- x[1:4]
  last4 <- x[(length(x)-3):length(x)]

  f4c <- rawConnection(first4)
  on.exit(close(f4c), add=TRUE)
  f4 <- readBin(con = f4c, what = "integer", n = 1, size = 4L, endian=endian)

  l4c <- rawConnection(last4)
  on.exit(close(l4c), add=TRUE)
  l4 <- readBin(con = l4c, what = "integer", n = 1, size = 4L, endian=endian)

  stopifnot(f4 == l4) # check front/back are equal
  stopifnot((f4 %% type_size == 0)) # shld have nothing left over

  n_rec <- f4 / type_size
  message(sprintf("Reading in %s records...", scales::comma(n_rec)))

  x <- x[-(1:4)]
  x <- x[-((length(x)-3):length(x))]

  rc <- rawConnection(x)
  on.exit(close(rc), add=TRUE)

  what <- switch(type, byte="raw", integer="integer", float="numeric")
  dat <- readBin(con = rc, what = what, n = n_rec, size = type_size, endian=endian)

  dat

}

此文未完成,已发布以取得进展。

数据文件中可能存在未记录的 "feature",因为前八个字节不是数据的一部分。 (该文件是 259208,但 360*180*4==259200。)不过,我确实发现了其他有趣的东西:

d <- readBin(file("~/Downloads/smoothmap.fix.2.bin", "rb"), integer(), n = 360*180, size = 4, endian = "big")

head(d)
# [1] 259200      0      0      0      0      0

我要推断第一个4字节整数(259200)表示数据的大小,所以我建议我们可以丢弃它。您可能会争辩说您的矢量长度是否合适,但那是因为您强制 readBin 停止加载数据。来自 ?readBin:

   n: integer.  The (maximal) number of records to be read.  You
      can use an over-estimate here, but not too large as storage
      is reserved for 'n' items.

因此读取超出您预期的文件大小应该是安全的,它会自行处理 EOF。我随便加10:

length(d)
# [1] 64800
d <- readBin(file("~/Downloads/smoothmap.fix.2.bin", "rb"), integer(), n = 360*180+10, size = 4, endian = "big")
length(d)
# [1] 64802
tail(d)
# [1] 1098907648 1098907648 1098907648 1098907648 1098907648     259200

(请注意,尽管我建议再读取 10 个字节,但只有两个可用。所以你知道,n 参数的基本原理是为了预分配内存,仅此而已。) 259200 又出现了,我会推断确认数据结束,所以我们应该能够安全地丢弃这两个 (first/last) 数字。

d <- d[-c(1, length(d))]

第一个非零数是:

head(which(d>0))
# [1] 4321 4322 4323 4324 4325 4326
d[4321]
# [1] 1101004800

并查看位:

intToBits(d[4321])
#  [1] 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 01 00 01 01
# [26] 00 00 00 00 00 01 00

因此,如果您推断出直接二进制解释,则该值为 2820,与可用值的 smoothmap.readme 描述不匹配。此外,我们期待看到的是:

intToBits(22)
#  [1] 00 01 01 00 01 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00

看来您的位……顺序不对,或类似的问题。如果你 intToBits 所有唯一值,你会注意到所有位 1-19(最低有效位)都是零。

从这里开始,我不知所措了...

sapply(unique(d), function(a) packBits(rev(intToBits(a)), type="integer"))
#  [1]    0 1410  258 1154 3714 6530 3458  770  514 5506 2690 1666 2434 2178 1282  130  642 4482    2 3202 1794  508  386