Cannot load AzureML library into an experiment

Question

Hello,

I cannot load AzureML library into my experiment and as far as I know this is the only way to be able to have access to data from Datasets.
I created a package with all the libraries needed for AzureML and attached it as a ZIP file to Execure R Script module. I get the following error:

[ModuleOutput] package 'AzureML' successfully unpacked and MD5 sums checked
[ModuleOutput] 
[ModuleOutput] Error in inDL(x, as.logical(local), as.logical(now), ...) : 
[ModuleOutput] 
[ModuleOutput]   unable to load shared object 'C:/ThirdParty/site-library/curl/libs/x64/curl.dll':
[ModuleOutput] 
[ModuleOutput]   LoadLibrary failure:  A dynamic link library (DLL) initialization routine failed.

I was able to use it easily inside Jupyter (library("AzureML") worked fine).

Do you have any ideas how to make it work? Or are there other ways to access Datasets? I would love to be able to just write load("mydata.RData") and take the RData file from Datasets.

Thanks in advance,

Piotr

Answer 1

This is a known limitation of the sandboxed environment where Execute R Script runs, some libraries such as curl.dll fail to load.

What's the reason you'd need to access the dataset from the script directly? Could you put your RData into a zip file and pass it in through the zip input port instead?

-Roope

 

Answer 2

Thanks for your answer, Roope.

What I need is to be able to pass my RData (available in Datasets, often updated) and the scripts (fixed set) to my Execute R Script (ERS). The ERS has only one Script Bundle input, and without the AzureML library I don't see a way to access my RData from Datasets.

In fact, I don't even see a way to access the RData file from Datasets, what I do right now is to zip it and download it with the AzureML lib.

In my perfect world I'd love to have my RData available in some place to all my ERSs.

I tried to pass my data from one ERS to another, but both serializing and casting it to data.frame changed something inside the data so my functions reported later that the data is not valid (probably some change in encoding of strings).

I put all my scripts and the RData file into one ZIP and it worked well, but as I mentioned earlier the RData is often changed and I'd like to separate them.

-Piotr

Answer 3

I don't think that's currently supported, unfortunately. We'll consider it for inclusion in future iterations.