This repository was archived by the owner on Nov 19, 2020. It is now read-only.
-
Notifications
You must be signed in to change notification settings - Fork 2k
This repository was archived by the owner on Nov 19, 2020. It is now read-only.
GeneralConfusionMatrix's precision and recall numbers seem to be flipped #1187
Open
Description
What would you like to submit? (put an 'x' inside the bracket that applies)
- question
- [ x ] bug report
- feature request
Issue description
-
Precision and Recall numbers from GeneralConfusionMatrix seem to be flipped. See below for how I computed precision/recall numbers and compared them against GeneralConfusionMatrix's precision/recall numbers
-
Code:
var cv = CrossValidation.Create, NaiveBayesLearning, double[], int>(
k: 3, // number of folds
learner: (p) => new NaiveBayesLearning(), // Naive Bayes Classifier
loss: (actual, expected, p) => new ZeroOneLoss(expected).Loss(actual),
fit: (teacher, x, y, w) => teacher.Learn(x, y, w),
x: input,
y: output
);
// After the cross-validation object has been created,
// we can call its .Learn method with the input and
// output data that will be partitioned into the folds:
var result = cv.Learn(input, output);
// We can grab some information about the problem:
int numberOfSamples = result.NumberOfSamples;
int numberOfInputs = result.NumberOfInputs;
int numberOfOutputs = result.NumberOfOutputs;
double trainingError = result.Training.Mean;
double validationError = result.Validation.Mean;
var expectedOut = result.Models[0].Model.Decide(input);
var actualOut = targetVariables.ValuesAll.ToArray();
var correct = 0;
var correctClass1 = 0;
for(int i = 0; i < expectedOut.Length; i++)
{
if(expectedOut[i] == actualOut[i])
{
correct += 1;
if(expectedOut[i] > 0)
{
correctClass1 += 1;
}
}
}
Console.WriteLine("{0}", ((float)correct / (float)actualOut.Length));
Console.WriteLine("{0}", ((float)correctClass1 / (float)targetVariables.NumSum()));
// If desired, compute an aggregate confusion matrix for the validation sets:
GeneralConfusionMatrix gcm = result.ToConfusionMatrix(input, output);
Console.WriteLine("");
Console.Write("\t\tActual 0\t\tActual 1\n");
for (int i = 0; i < gcm.Matrix.GetLength(0); i++)
{
Console.Write("Pred {0} :\t", i);
for (int j = 0; j < gcm.Matrix.GetLength(1); j++)
{
Console.Write(gcm.Matrix[i, j] + "\t\t\t");
}
Console.WriteLine();
}
double accuracy = gcm.Accuracy;
double[] precision = gcm.Precision;
double[] recall = gcm.Recall;
Console.WriteLine("---- Using GeneralConfusionMatrix's Precision/Recall ----");
Console.WriteLine("# samples: {0}, # inputs: {1}, # outputs: {2}", numberOfSamples, numberOfInputs, numberOfOutputs);
//Console.WriteLine("training error: {0}", trainingError);
//Console.WriteLine("validation error: {0}", validationError);
Console.WriteLine("Precision: {0}, {1}", precision[0], precision[1]);
Console.WriteLine("Recall: {0}, {1}", recall[0], recall[1]);
//Console.WriteLine("Accuracy: {0}", accuracy);
Console.WriteLine("---- Manually Calculating Precision & Recall Numbers");
Console.WriteLine("Row total: {0}, {1}", gcm.RowTotals[0], gcm.RowTotals[1]);
Console.WriteLine("Col total: {0}, {1}", gcm.ColumnTotals[0], gcm.ColumnTotals[1]);
// True-Positive / (True-Positive + False-Positive)
Console.WriteLine("Precision: {0}", ((float)gcm.Matrix[1, 1]) / (gcm.Matrix[1, 0] + gcm.Matrix[1, 1]));
// True-Positive / (True-Positive + False-Negative)
Console.WriteLine("Recall: {0}", ((float)gcm.Matrix[1, 1]) / (gcm.Matrix[1, 1] + gcm.Matrix[0, 1]));
Metadata
Metadata
Assignees
Labels
No labels