在Azure ML Studio上为Microsoft R Open 3.4.4安装textshape程序包

Question

I'm trying to use the R sentimentr package on Azure ML Studio. As this package is not supported, I'm trying to install it and its dependencies as described in the documentation .

The steps that I have performed are:

• downloaded Windows binaries from the R Open 3.4.4 snapshot at CRAN time machine

  • sentimentr_2.2.3.zip
  • syuzhet_1.0.4.zip
  • textclean_0.6.3.zip
  • lexicon_0.7.4.zip
  • textshape_1.5.0.zip

• zipped those zip files into a zipped folder packages.zip

• uploaded packages.zip as a dataset to Microsoft Azure ML Studio

In my ML experiment I connect the packages.zip dataset to the "Script Bundle (Zip)" input port on "Execute R Script" and include this code:

# install R package contained in src  
install.packages("src/lexicon_0.7.4.zip", 
                 lib = ".", 
                 repos = NULL, 
                 verbose = TRUE)  

install.packages("src/textclean_0.6.3.zip", 
                 lib = ".", 
                 repos = NULL, 
                 verbose = TRUE)  

install.packages("src/textshape_1.5.0.zip", 
                 lib = ".", 
                 repos = NULL, 
                 verbose = TRUE)  

install.packages("src/syuzhet_1.0.4.zip", 
                 lib = ".", 
                 repos = NULL, 
                 verbose = TRUE)  

install.packages("src/sentimentr_2.2.3.zip", 
                 lib = ".", 
                 repos = NULL, 
                 verbose = TRUE)  

# load libraries
library(sentimentr, lib.loc = ".", verbose = TRUE)

The experiment runs successfully, until I include a function from sentimentr :

mydata <- mydata %>%
  get_sentences() %>%
  sentiment()

This gives the error:

there is no package called 'textshape'

Which is difficult to understand given that the output log does not indicate an issue with the packages:

[Information]         The following files have been unzipped for sourcing in path=["src"]:
[Information]                           Name  Length                Date
[Information]         1 sentimentr_2.2.3.zip 3366245 2019-08-07 14:57:00
[Information]         2    syuzhet_1.0.4.zip 2918474 2019-08-07 15:05:00
[Information]         3  textclean_0.6.3.zip 1154814 2019-08-07 15:13:00
[Information]         4    lexicon_0.7.4.zip 4551995 2019-08-07 15:17:00
[Information]         5  textshape_1.5.0.zip  463095 2019-08-07 15:42:00
[Information]         Loading objects:
[Information]           port1
[Information]         [1] "Loading variable port1..."
[Information]         package 'lexicon' successfully unpacked and MD5 sums checked   
[Information]         package 'textclean' successfully unpacked and MD5 sums checked
[Information]         package 'textshape' successfully unpacked and MD5 sums checked
[Information]         package 'syuzhet' successfully unpacked and MD5 sums checked
[Information]         package 'sentimentr' successfully unpacked and MD5 sums checked

Has anyone seen this, or similar issues? Is it possible that "successfully unpacked" is not the same as successfully installed and usable?

Answer 1

I can now answer my own question thanks to a hint on Twitter from @bryan_hepworth.

The R packages were installed correctly, but not in the standard library location. So when a function from sentimentr runs, R tries to load the dependency package textshape :

library(textshape)

Which of course does not exist in the standard location as Azure ML does not support it.

The solution is to load textshape explicitly from its installed location:

library(textshape, lib.loc = ".")

So the solution is: explicitly load packages that you installed at the start of your R code, rather than letting R try to load them as dependencies, which will fail.