CODEWITHSUNDEEP
rbase64type-conversion
Rorschach
Rorschach

Reputation: 32446

Convert raw vector to R object

How do you convert a raw vector back to an R object without writing to disk? I want to read a stream of base64 data and convert it to its R object representation. Here is an example - how would I get back the lm object from the raw vector?

## some rdata -- writes to temp file!
mod <- lm(mpg ~ cyl, data=mtcars)
f1 <- tempfile()
save(mod, file=f1, compress="bzip2")

library(base64enc)
r1 <- readBin(f1, "raw", n=file.info(f1)[1, "size"])
r2 <- base64decode(base64encode(file(f1, "rb")))  # emulate input base64
identical(r1, r2)

## I can get mod back by writing to file and loading, but how to just
## load from a raw vector?
rm(mod)  # get rid of mod
f2 <- tempfile()
writeBin(r2, f2)
load(f2)  # mod is back

Upvotes: 8

Views: 1352

Answers (2)

Martin Morgan
Martin Morgan

Reputation: 46876

For R-level access use unserialize(serialize(mod, NULL)) to round-trip from an R object to a raw vector and back.

Upvotes: 3

Dirk is no longer here
Dirk is no longer here

Reputation: 368409

Inside my RcppRedis package I use the RApiSerialize package (which is based on base R code initially borrowed in the Rhpc package) to do these conversions on the fly:

R> mod <- lm(mpg ~ cyl, data=mtcars)     # your example
R>
R> library(RApiSerialize)
R> modraw <- serializeToRaw(mod)         # serialized
R> str(modraw)                           # really just a raw vector now
 raw [1:6819] 58 0a 00 00 ...
R>

So at this point you can do whatever you want with the raw vector. Write it to disk, write it to a database (as we do with RcppRedis), ....

But importantly, you also get your model back:

R> summary( unserializeFromRaw(modraw) )

Call:
lm(formula = mpg ~ cyl, data = mtcars)

Residuals:
   Min     1Q Median     3Q    Max 
-4.981 -2.119  0.222  1.072  7.519 

Coefficients:
            Estimate Std. Error t value Pr(>|t|)    
(Intercept)   37.885      2.074   18.27  < 2e-16 ***
cyl           -2.876      0.322   -8.92  6.1e-10 ***
---
Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

Residual standard error: 3.21 on 30 degrees of freedom
Multiple R-squared:  0.726, Adjusted R-squared:  0.717 
F-statistic: 79.6 on 1 and 30 DF,  p-value: 6.11e-10

R>

Upvotes: 5

Related Questions