public ModelData CreateModel(string userId, string name, int supportThreshhold, int maxItemSetSize)
{
// Create a new model
var model = new ModelData()
{
PartitionKey = userId,
RowKey = name,
MaxItemSetSize = maxItemSetSize,
SupportThreshhold = supportThreshhold,
Status = "No Training Data"
};
// insert the entity and return it if successfull
return tableClient.InsertEntity(ModelTableName, model) ? model : null;
}
public bool UpdateModel(ModelData model)
{
// insert the entity and return it if successfull
return tableClient.UpdateEntity(ModelTableName, model);
}
public void TrainModel(ModelData model)
{
// How the model will be identified in Cloud ML
var modelIdentifier = Guid.NewGuid().ToString().Replace("-", "").Substring(0, 24);//userId + "-" + modelName;
// HACK: temporary workaround issue of webservices failing after 12 hours by extending lease time to 1 year
// The fix will be to have a seperate scoring experiment
string dataReadUrl = modelProvider.GetTemporaryDataFileReadUrl(model.DataFileBlobName, TimeSpan.FromHours(9000));
// Run the training experiment
var request = new BatchRequest()
{
GlobalParameters = new Dictionary <string, string>()
};
request.GlobalParameters["URL"] = dataReadUrl;
request.GlobalParameters["MaxItemSetSize"] = model.MaxItemSetSize.ToString();
request.GlobalParameters["SupportThreshhold"] = model.SupportThreshhold.ToString();
request.GlobalParameters["NumberofSets"] = "1";
var trainingJobId = trainingEndpoint.SubmitBatch(request).Result;
UpdateModelStatus("Training the Model",
m => m.ExperimentId = trainingJobId);
// wait for it to finish
var currentStatus = WaitForCompletion(trainingJobId);
// if it failed stop
if (currentStatus.StatusCode != BatchScoreStatusCode.Finished)
{
UpdateModelStatus("Error Building Training Model: " + currentStatus.StatusCode + currentStatus.Details);
return;
}
var trainedModel = currentStatus.Results["TrainedModel"];
UpdateModelStatus("Training the Model");
// if this is a new endpoint then just retrain it
Endpoint scoringEndpoint;
if (model.WebServiceGroupId != null && model.WebServiceGroupId.Length == 24)
{
scoringEndpoint = new Endpoint()
{
ApiLocation = model.ScoringUrl,
PrimaryKey = model.ScoringAccessKey,
WorkspaceId = scoringWebService.WorkspaceId,
WebServiceId = scoringWebService.Id,
Name = model.WebServiceGroupId
};
}
else
{
// Create the scoring endpoint
var epCreate = new EndpointCreate()
{
Description = "",
MaxConcurrentCalls = 10,
ThrottleLevel = ThrottleLevel.High
};
scoringEndpoint = scoringWebService.CreateEndpoint(modelIdentifier, epCreate, scoringWorkspaceEndpointToken).Result;
if (scoringEndpoint == null)
{
UpdateModelStatus("Error Creating Scoring Endpoint");
return;
}
}
// configure it to use the trained model
if (!scoringEndpoint.UpdateResource(new ResourceLocations()
{
Resources = new [] { new EndpointResource()
{
Name = "FBT Model",
Location = trainedModel
} }
}).Result)
{
UpdateModelStatus("Error Configuring Scoring Endpoint");
return;
}
// test the scoring service
ValidateScoringEndpoint(scoringEndpoint.ApiLocation, scoringEndpoint.PrimaryKey);
// All Done
UpdateModelStatus("Ready",
m => m.ScoringUrl = scoringEndpoint.ApiLocation,
m => m.WebServiceGroupId = modelIdentifier,
m => m.ScoringAccessKey = scoringEndpoint.PrimaryKey,
m => m.TrainingInvalidated = false,
m => m.IsTraining = false,
m => m.TrainingEndTime = DateTimeOffset.Now);
}
private Model GetModelFromData(ModelData model)
{
return new Model()
{
Name = model.RowKey,
Status = model.Status,
MaxItemsSetSize = model.MaxItemSetSize,
MinumumOccurrences = model.SupportThreshhold,
ReadyForScoring = !string.IsNullOrEmpty(model.ScoringUrl),
HasData = !string.IsNullOrEmpty(model.DataFileBlobName),
NeedsTraining = model.TrainingInvalidated,
IsTraining = model.IsTraining,
DataUploadTime = model.DataUploadTime,
TrainingStartTime = model.TrainingStartTime,
TrainingEndTime = model.TrainingEndTime
};
}
public bool UpdateModel(ModelData model)
{
// insert the entity and return it if successfull
return(tableClient.UpdateEntity(ModelTableName, model));
}