无效的输出扩展名导致Azure ML实验批处理Web服务调用失败

Question

I have an Azure webjob that is calling a ML training experiment via HttpRequests, leveraging the code generated in the ML webportal:

var request = new BatchExecutionRequest()
            {
                Inputs = new Dictionary<string, AzureBlobDataReference>() {
                    {
                        "input1",
                        new AzureBlobDataReference()
                        {
                            ConnectionString = _connectionString,
                            RelativeLocation = $"{_containerName}/{experimentId}/{tenantId}/{trainingDataFileName}"
                        }
                    },
                },

                Outputs = new Dictionary<string, AzureBlobDataReference>() {
                    {
                        "output1",
                        new AzureBlobDataReference()
                        {
                            ConnectionString = "azureStorageConnectionString",
                            RelativeLocation = $"{_containerName}/{experimentId}/{tenantId}/Model_2018421.ilearner"
                        }
                    },
                },

                GlobalParameters = new Dictionary<string, string>()
                {
                }
            };

However, the request fails with the following message:

The blob reference: experiments/experimentId/TenantId/Model_2018421.ilearner has an invalid or missing file extension. Supported file extensions for this output type are: \\\\".csv, .tsv, .arff\\\\"

I'm pretty confused about this, since it's written right the documentation all over the place that if I'm expecting a trained model to use ".ilearner" as the file extension for the model.

I've seen this question asking about the same error leveraging the DataFactory, and also this question on datascience.stackexchange . Neither one had any clues, answers, or other follow up.

Any insight on what I'm missing would be greatly appreciated!

Answer 1

For anyone looking for your "Don't Overthink It" moment of the day:

I needed to provide TWO output blob file references:

var request = new BatchExecutionRequest()
            {
                Inputs = new Dictionary<string, AzureBlobDataReference>() {
                    {
                        "input1",
                        new AzureBlobDataReference()
                        {
                            ConnectionString = _connectionString,
                            RelativeLocation = $"{_containerName}/{experimentId}/{tenantId}/{trainingDataFileName}.csv"
                        }
                    },
                },

                Outputs = new Dictionary<string, AzureBlobDataReference>() {
                    {
                        "output1",
                        new AzureBlobDataReference()
                        {
                            ConnectionString = _connectionString,
                            RelativeLocation = $"{_containerName}/{experimentId}/{tenantId}/{outputFileNameCsv}.csv"
                        }
                    },
                    {
                        "output2",
                        new AzureBlobDataReference()
                        {
                            ConnectionString = _connectionString,
                            RelativeLocation = $"{_containerName}/{experimentId}/{tenantId}/{outputFileNameIlearner}.ilearner"
                        }
                    },
                },

                GlobalParameters = new Dictionary<string, string>()
                {
                }
            };

There's an old saying in American English about not making assumptions, and I assumed the second output was an optional parameter used in batch operations. Since I'm not actually looking for more than one result from each call, I thought I was safe to remove the second output parameter.

TL/DR: Keep all the parameters the webservice portal's "Consume" tab generates, and make sure the first one is a .csv file reference.