Interactive Metric Logging not writing back to AzureML Experiments (not sure if this is a bug / not a feature)

[buswrecker]

Bug/Problem Description

I'm trying to log metrics interactively; as i'm writing code interactively on a local machine; ie not within an estimator or a submit_experiment with local compute. I understand that this can be done if wrapped within an estimator; but that's not what i'm looking for; there are instances, where i don't want to wrap my interactive code as an estimator, but i definitely would like to log items while i'm testing / building - ie a cool graph/metric etc.

To Reproduce

library(azuremlsdk)
ws <- load_workspace_from_config()

experiment_name <- "train-r-local-test"
exp <- experiment(ws, experiment_name)
runid <- start_logging_run(exp)
message("set x to 100")
x <- 100

message("logging accuracy")
log_metric_to_run("Accuracy", 0.23)
log_metric_to_run("RMSE", 0.00001)

complete_run(runid)

Expected Behaviour

Accuracy and RMSE values to pop up in my experimentation page in AzureML; this doesn't happen - see below

Additional Context

We are trying to integrate AzureML R SDK within a Shiny Application; and we would want metrics of the ShinyApp to be logged into AzureML's experimentation page; and subsequently registering models into the AzureML Model Management.

Yes - we could design the Shiny app to trigger AzureML Submit_Experiments into a Remote Compute cluster; but for some of the interactive work on the App; this is not appropriate, the latency to submit experiments into a remote compute for logging one or two metrics exceeds a threshold that would be rather non-interactive for the ShinyApp user.

[buswrecker]

The RunID is required in the log_metric_to_run function

The correct code for this to run

runid <- start_logging_run(exp)
message("set x to 100")
x <- 100

message("logging accuracy")
log_metric_to_run("Accuracy", 0.23, runid)
log_metric_to_run("RMSE", 0.00001, runid)

This should be documented in ?start_logging_run

Examples:

     ## Not run:
     
     ws <- load_workspace_from_config()
     exp <- experiment(ws, name = 'myexperiment')
     run <- start_logging_run(exp)
     log_metric_to_run("Accuracy", 0.9)
     complete_run(run)
     ## End(Not run)