Azure Machine Learning - using CLI to run R code - Stack Overflow

admin管理员组

文章数量:1350343

I'm attempting to run a Azure ML job to train and save a model using R. It seems as if my pipeline runs, but it doesn't save the output. I'm using a very simple script first of all as a proof before I move onto the actual R workload I plan to deploy.

Due to lack of MS documentation on running R code in Azure ML (there was documentation up until around 2 weeks ago, although this has been removed - I've raised a query with MS about this), I'm struggling to find examples on how to accomplish this.

There are some code examples of Github which are of some use. These examples include the use of mlflow, however, speaking to the MS rep as well as other documentation I've seen, I don't think the use of mlflow is imperative for running R code (it's only necessary if you want to rely on its ability to log metrics etc).

.R

;t=644s

AZURE-ML-IRIS
- docker-context
---- Dockerfile this is the Dockerfile from the MS Github azureml-examples for R
- src
---- train.R
- job.yml

train.R

library(optparse)
library(rpart)

parser <- OptionParser()

parser <- add_option(
    parser, "--data_folder",
    type="character", 
    action="store", 
    default = "./data", 
    help="data folder")

parser <- add_option(
  parser,
  "--data_output",
  type = "character",
  action = "store",
  default = "./data_output"
)

args <- parse_args(parser)

file_name = file.path(args$data_folder)

iris <- read.csv(file_name)
iris_head <- head(iris)

write.csv(iris_head, file = paste0(args$data_output, "/iris_head.csv"))

job.yml

$schema: .schema.json
command: >
  Rscript train.R 
  --data_folder ${{inputs.iris}}
  --data_output ${{outputs.data_output}}
code: src
inputs:
  iris: 
    type: uri_file
    path: .csv
outputs:
  data_output:
environment:
  build:
    path: docker-context
display_name: r-iris-example
compute: azureml:noel001
experiment_name: r-iris-example
description: Get a subset of Iris data. 

I execute the creation of the job with the az ml job create command. The job runs, and completes according to Azure ML. However, it doesn't seem as if the iris_head.csv file actually get's saved anyway. The outputs data asset url the job suggests outputs are saved to contains no files.

I've ran the hello world example for data outputs:

$schema: .schema.json
command: echo "hello world" > ${{outputs.hello_output}}/helloworld.txt
outputs:
  hello_output:
environment:
  image: python

And that runs as expected producing a small .txt file. What I can't seem to do is move from this hello world example through to the R example.

I've also tried the full end to end examples from the Github repos above (including the mlflow elements) and run into the same problems for each.

Any help would be greatly appreciated.

本文标签： Azure Machine Learningusing CLI to run R codeStack Overflow