static void Main(string[] args)
{
var inputFile = $"..{sep}..{sep}..{sep}Data{sep}{_fileName}.csv";
var outputFile = $"{_fileName}_withActions.csv";
Console.WriteLine($"{DateTime.Now} Testing the ML model by calling the Web API for a set of test claims...");
Console.WriteLine($"Input file containing test claims: {Path.GetFullPath(inputFile)}");
Console.WriteLine($"Output file with added predicted actions: {Path.GetFullPath(outputFile)}");
using (var reader = File.OpenText(inputFile))
using (var writer = File.CreateText(outputFile))
{
string currLine;
bool currLineIsHeader = true;
// currentLine will be null when the StreamReader reaches the end of file
while ((currLine = reader.ReadLine()) != null)
{
if (currLineIsHeader)
{
const string expectedHeader = "MemberID,ClaimID,DateReceived,providerNPI,Diagnosis1,Diagnosis2,POS,ProcedureCode,Units,Price,PendReason,Action,Status";
Debug.Assert(currLine == expectedHeader);
writer.WriteLine(currLine + ",ActionScore,ReleaseScore,DenyScore,ReprocessScore,MedReviewScore"); // the actual prediction will be placed in Action column (11)
currLineIsHeader = false;
continue;
}
var elems = currLine.Split(',').ToList();
Debug.Assert(elems.Count == 13);
var modelInput = new ModelInput()
{
Diagnosis1 = elems[4],
Diagnosis2 = elems[5],
POS = elems[6],
ProcedureCode = elems[7],
Units = int.Parse(elems[8]),
Price = float.Parse(elems[9]),
PendReason = elems[10]
};
var predictedOutput = PendPredictor.PredictEx(modelInput);
elems[11] = predictedOutput.Action;
elems.Add(GetScoreForAction(predictedOutput, predictedOutput.Action));
elems.Add(GetScoreForAction(predictedOutput, "Release"));
elems.Add(GetScoreForAction(predictedOutput, "Deny"));
elems.Add(GetScoreForAction(predictedOutput, "Reprocess"));
elems.Add(GetScoreForAction(predictedOutput, "MedReview"));
writer.WriteLine(string.Join(',', elems.ToArray()));
}
}
Console.WriteLine();
Console.WriteLine($"{DateTime.Now} ALL DONE!");
if (Debugger.IsAttached)
{
Console.Write("Press any key to continue . . . ");
Console.ReadKey();
}
}
public IActionResult ObtainPendAdvice([FromBody] ClaimData claim)
{
if (!ModelState.IsValid)
{
return(BadRequest());
}
//For simplicity, the entire client logic is implemented inside the controller
var modelInput = _mapper.Map <ModelInput>(claim);
var modelOutput = PendPredictor.PredictEx(modelInput);
var adviceData = _mapper.Map <AdviceData>(modelOutput);
adviceData.ClaimID = claim.ClaimID;
return(Ok(adviceData));
}
public void Predict_SampleInput_CorrectOutput()
{
//arrange
var sampleInput = new ModelInput()
{
Diagnosis1 = "E10",
Diagnosis2 = "I11",
POS = "21",
ProcedureCode = "34907",
Units = 1,
Price = 100f,
PendReason = "PAUT"
};
//act
var predictedOutput = PendPredictor.Predict(sampleInput);
//assert
predictedOutput.Action.Should().BeOneOf("RePend", "Release", "Deny", "MedReview");
predictedOutput.Scores.Count().Should().Be(4);
predictedOutput.Scores.ToList().ForEach(s => s.Should().BeGreaterThan(0f).And.BeLessThan(1f));
}
public void PredictEx_SampleInput_CorrectOutput()
{
//arrange
var sampleInput = new ModelInput()
{
Diagnosis1 = "E10",
Diagnosis2 = "I11",
POS = "21",
ProcedureCode = "34907",
Units = 1,
Price = 100f,
PendReason = "PAUT"
};
//act
var predictedOutput = PendPredictor.PredictEx(sampleInput);
//assert
predictedOutput.ActionsAndScores.Count().Should().Be(4);
predictedOutput.ActionsAndScores.Select(t => t.Score).Should().BeInDescendingOrder();
predictedOutput.ActionsAndScores.Select(t => t.Action).Should().BeEquivalentTo("RePend", "Release", "Deny", "MedReview");
}
static void Main(string[] args)
{
// Load the data
var allData = _mlContext.Data.LoadFromTextFile <ModelInput>(_dataPath, hasHeader: true, separatorChar: ',');
// Split the data into a training set (80%) and a test set (20%)
var trainTestData = _mlContext.Data.TrainTestSplit(allData, testFraction: 0.2, seed: 0);
var trainData = trainTestData.TrainSet;
var testData = trainTestData.TestSet;
//Build training pipeline
(IEstimator <ITransformer> trainingPipeline, string algorithmDesc) = BuildTrainingPipeline();
//Train the model
Console.WriteLine($"{DateTime.Now} Training the model...");
Console.WriteLine($"Training algorithm used: {algorithmDesc}");
ITransformer mlModel = TrainModel(trainingPipeline, trainData);
// Evaluate quality of the model
Console.WriteLine();
Console.WriteLine($"{DateTime.Now} Evaluating the model...");
(var metrics, var crossValidationMetrics) = EvaluateModel(mlModel, testData, trainingPipeline, allData);
Console.WriteLine($"Macro accuracy = {metrics.MacroAccuracy:P2}");
Console.WriteLine($"Micro accuracy = {metrics.MicroAccuracy:P2}");
Console.WriteLine(metrics.ConfusionMatrix.GetFormattedConfusionTable());
if (crossValidationMetrics == null)
{
Console.WriteLine("Cross-validation skipped.");
}
else
{
Console.WriteLine($"Mean cross-validated macro accuracy = {crossValidationMetrics.Average(s => s.Metrics.MacroAccuracy):P2}");
Console.WriteLine($"Mean cross-validated micro accuracy = {crossValidationMetrics.Average(s => s.Metrics.MicroAccuracy):P2}");
}
// Use the model to make a sample prediction
Console.WriteLine();
Console.WriteLine($"{DateTime.Now} Using the model to make prediction...");
var input = new ModelInput
{
Diagnosis1 = "E10",
Diagnosis2 = "I11",
POS = "21",
ProcedureCode = "34907",
Units = 1,
Price = 100f,
PendReason = "PAUT"
};
var predictor = _mlContext.Model.CreatePredictionEngine <ModelInput, ModelOutput>(mlModel);
var prediction = predictor.Predict(input);
var labelMap = PendPredictor.GetLabelMap(predictor); //translates 0-based index to the actual value in the Label column.
Console.WriteLine("Predictions and their scores (high to low):");
prediction.Scores.Zip(Enumerable.Range(0, int.MaxValue))
.OrderByDescending(t => t.First).ToList() //First = Score, Second = index
.ForEach(t => Console.WriteLine($"{t.Second}.{labelMap[t.Second],8} - {t.First:N8}"));
Console.WriteLine();
Console.WriteLine($"So, the predicted action is {prediction.Action}.");
// Save the trained model to a .ZIP file
Console.WriteLine();
Console.WriteLine($"{DateTime.Now} Saving the trained model into {PendPredictor.PathToSaveModel} file...");
SaveModel(mlModel, allData.Schema, PendPredictor.PathToSaveModel);
Console.WriteLine($"{DateTime.Now} Saving the trained model into {PendPredictor.PathToSaveModel2} file...");
SaveModel(mlModel, allData.Schema, PendPredictor.PathToSaveModel2);
Console.WriteLine("Rebuild the API and Tester projects in order for the new model to be in effect!");
Console.WriteLine();
Console.WriteLine($"{DateTime.Now} ALL DONE!");
if (Debugger.IsAttached)
{
Console.Write("Press any key to continue . . . ");
Console.ReadKey();
}
}
