public static Tuple <double, double> ObtainProbabilityOfZeroAndOne(List <int> testValues, Dictionary <string, BucketCount> trainingEmails, double totalProbabilityOne)
{
double probabilityZero = 0, probabilityOne = 0;
double totalProbabilityZero = 1 - totalProbabilityOne;
for (int index = 0; index < testValues.Count; index++)
{
var testValue = testValues[index];
string key = NaiveBayesDataTransform.BuildKey(index, testValue);
// Count all the times this word was spam
BucketCount bucketCount;
if (!trainingEmails.TryGetValue(key, out bucketCount))
{
// laplace smoothing
bucketCount = new BucketCount();
}
const double smoothingNum = 0.1; // Feel free to change this #
double probabilityOfExampleBeingZero = (1.0 * bucketCount.ZeroCount + smoothingNum) /
(bucketCount.ZeroCount + bucketCount.OneCount + smoothingNum);
double probabilityOfExampleBeingOne = (1.0 * bucketCount.OneCount + smoothingNum) /
(bucketCount.ZeroCount + bucketCount.OneCount + smoothingNum);
probabilityZero += Math.Log(probabilityOfExampleBeingZero);
probabilityOne += Math.Log(probabilityOfExampleBeingOne);
}
probabilityZero += Math.Log(totalProbabilityZero);
probabilityOne += Math.Log(totalProbabilityOne);
return(new Tuple <double, double>(probabilityZero, probabilityOne));
}
public static Dictionary <string, BucketCount> CountSamples(List <DataSetValue> rows)
{
var bucketCounts = new Dictionary <string, BucketCount>();
for (int rowIndex = 0; rowIndex < rows.Count; rowIndex++)
{
var row = rows[rowIndex];
bool output = row.Output;
for (int index = 0; index < row.Values.Count; index++)
{
var rowValue = row.Values[index];
string bucketKey = BuildKey(index, rowValue);
BucketCount bucketCount;
if (!bucketCounts.TryGetValue(bucketKey, out bucketCount))
{
bucketCount = new BucketCount();
bucketCounts.Add(bucketKey, bucketCount);
}
bucketCount.Add(output);
}
}
return(bucketCounts);
}
