static void Statistics(GisModel model, string testFilePath)
{
var results = new List <Tuple <string, bool, double> >();
// create detector from model
var invalidEmailDetector = new MaximumEntropyInvalidEmailDetector(model);
// read all test (email + invalid flag) lines
var allLines = File.ReadAllLines(testFilePath);
foreach (var line in allLines)
{
var parts = line.Split('\t');
var email = parts.First();
var isInvalid = parts.Last() == "1";
// store the result of the detection as well as the actual validity of the email
var invalidProbability = invalidEmailDetector.GetInvalidProbability(email);
results.Add(new Tuple <string, bool, double>(email, isInvalid, invalidProbability));
}
// pretend to send the emails, and stop when we sent to many emails that bounced
var nbOfEmailsSent = 0;
var nbOfEmailsSentWhichWouldBounce = 0;
var maxNbOfBouncePerDay = 25;
foreach (var result in results.OrderBy(tup => tup.Item3))
{
if (nbOfEmailsSentWhichWouldBounce < maxNbOfBouncePerDay)
{
nbOfEmailsSent++;
if (result.Item2)
{
nbOfEmailsSentWhichWouldBounce++;
}
}
Console.WriteLine("{0} ({1})", result.Item1, result.Item2 ? "INVALID": "OK");
}
Console.WriteLine("Email that could have been sent: {0}", nbOfEmailsSent);
Console.WriteLine("========");
// Quick summary of the detection
var probaOfInvalidity = 0.2497;
var nbOfSamples = results.Count;
Console.WriteLine("Nb of samples: {0}", nbOfSamples);
var nbOfCorrectResults = results.Count(tup => tup.Item2 == tup.Item3 > probaOfInvalidity);
Console.WriteLine("Nb of correct results: {0}", nbOfCorrectResults);
var nbOfNotDetected = results.Count(tup => tup.Item2 && !(tup.Item3 > probaOfInvalidity));
Console.WriteLine("Nb of not detected: {0}", nbOfNotDetected);
var nbOfFalsePositive = results.Count(tup => !tup.Item2 && (tup.Item3 > probaOfInvalidity));
Console.WriteLine("Nb of false positive: {0}", nbOfFalsePositive);
}
static void Main(string[] args)
{
// Training -------------------------------------
// train the model from a data set
var emailsAndValidities = new List <EmailAndValidity>()
{
new EmailAndValidity()
{
Email = "*****@*****.**", IsInvalid = false
},
new EmailAndValidity()
{
Email = "*****@*****.**", IsInvalid = false
},
new EmailAndValidity()
{
Email = "*****@*****.**", IsInvalid = false
},
new EmailAndValidity()
{
Email = "*****@*****.**", IsInvalid = false
},
new EmailAndValidity()
{
Email = "*****@*****.**", IsInvalid = true
},
new EmailAndValidity()
{
Email = "*****@*****.**", IsInvalid = true
},
new EmailAndValidity()
{
Email = "*****@*****.**", IsInvalid = true
},
new EmailAndValidity()
{
Email = "*****@*****.**", IsInvalid = true
},
};
var iterations = 100;
var cut = 5;
var model = MaximumEntropyInvalidEmailDetector.TrainModel(emailsAndValidities, iterations, cut);
// train the model from a formatted file
var inputFilePath = currentDirectory + "Input/invalidEmailDetection.train";
var model2 = MaximumEntropyInvalidEmailDetector.TrainModel(inputFilePath, iterations, cut);
// Detection -----------------------------------
// use trained model to build a detector
var invalidEmailDetector = new MaximumEntropyInvalidEmailDetector(model2);
// use the detector to compute the probability of invalidity of new email addresses
var newEmailAddresses = new List <string>()
{
"*****@*****.**", "[email protected]", "*****@*****.**"
};
foreach (var newEmailAddress in newEmailAddresses)
{
var probabilityOfInvalidity = invalidEmailDetector.GetInvalidProbability(newEmailAddress);
Console.WriteLine("{0} -> proba invalidity: {1}", newEmailAddress, probabilityOfInvalidity);
}
Console.WriteLine("----");
Console.ReadLine();
}
