LoadTransform and LinearClassificationTrainer doesnot work in AzureFunctions

[Anipik]

System information

• OS version/distro: Windows
• .NET Version (eg., dotnet --info):

.NET Core SDK (reflecting any global.json):
 Version:   2.1.400-preview-009063
 Commit:    dd0179a67c

Runtime Environment:
 OS Name:     Windows
 OS Version:  10.0.17134
 OS Platform: Windows
 RID:         win10-x64
 Base Path:   C:\Program Files\dotnet\sdk\2.1.400-preview-009063\

Host (useful for support):
  Version: 2.1.1
  Commit:  6985b9f684

Issue

• What did you do?
  I am trying to use the ML.net package in azure functions. I am running this https://github.com/dotnet/machinelearning-samples/tree/master/samples/end-to-end-apps/github-labeler inside an azure function.

• What happened?
  The entry point map for these Microsoft.ML.Runtime.Data.LoadTransform and Microsoft.ML.Runtime.Learners.LinearClassificationTrainer types are not getting loaded properly. As a result I am not able to train or test using ML.net inside the Azure Function.

Errors

System.Private.CoreLib: Exception while executing function: GithubIssueLabeler. Microsoft.ML.Data: Couldn't load model: 'DataLoaderModel\Transform_001
System.Private.CoreLib: Exception while executing function: GithubIssueLabeler. System.Private.CoreLib: Exception has been thrown by the target of an invocation. Microsoft.ML.Data: Couldn't load model: 'DataLoaderModel\Transform_020\SchemaBindableMapper\InnerMapper\Predictor'.

• What did you expect?
  Expect to work properly.

Source code / logs

        [FunctionName("GithubIssueLabeler")]
        public static async Task<IActionResult> Run([HttpTrigger(AuthorizationLevel.Function, "get", "post", Route = null)]HttpRequest req, TraceWriter log)
        {
            //var type = typeof(Microsoft.ML.Runtime.Data.LoadTransform);
            //var type1 = typeof(Microsoft.ML.Runtime.Learners.LinearClassificationTrainer);
            log.Info("Http Issue Webhook Request is Being Processed");

            string requestBody = new StreamReader(req.Body).ReadToEnd();
            dynamic data = JsonConvert.DeserializeObject(requestBody);

            string Action = data?.action;
            dynamic issue = data?.issue;
            dynamic labels = issue?.labels;

            if (Action == "opened" && labels.Count == 0)
            {
                string title = issue?.title;
                int number = issue?.number;
                string body = issue?.body;
                log.Info($"A {number.ToString()} issue with {title} has been opened.");

                Configuration = new ConfigurationBuilder()
                    .SetBasePath(Directory.GetCurrentDirectory())
                    .AddJsonFile("appsettings.json").Build();

                var labeler = new Labeler(Configuration["GitHubRepoOwner"], Configuration["GitHubRepoName"], Configuration["GitHubToken"]);
                await labeler.PredictAndApplyLabelAsync(number, title, body, log); // can do training or prediting using already load model
                log.Info("Labeling completed");
            }
            else
            {
                log.Info("The issue is already opened or it already has a label");
            }

            Console.ReadLine();
            log.Info($"Issue Label request handled");
            return Action != null
                ? (ActionResult)new OkObjectResult($"Issue Label request handled")
                : new BadRequestObjectResult("Please pass a name on the query string or in the request body");
        }

WorkArounds

Adding these lines to the function helps the app to run perfectly fine.

var type = typeof(Microsoft.ML.Runtime.Data.LoadTransform);
var type1 = typeof(Microsoft.ML.Runtime.Learners.LinearClassificationTrainer);

cc @eerhardt

[eerhardt]

This should no longer be necessary once #371 is completed. We will no longer need to to front load the assembly loading, as the API usage will be directly against the underlying C# assemblies.

[eerhardt]

This now works as expected in 0.6 after #970. Closing.