public static double testTextModel(model textModel, textTrainDataProvider tdp)
{
List <double> percents = new List <double>();
int total = 0;
int correct = 0;
for (int i = 0; i < tdp.data.Count; i++)
{
List <string[]> cat = tdp.getCatagory(i);
string expected = tdp.getCatID(i);
for (int j = 0; j < cat.Count; j++)
{
List <object> list = cat[j].ToList <object>();
Dictionary <object, double> pred = textModel.predict(list, true);
if (same(pred.Values.ToList()))
{
continue;
}
object output = pred.Keys.ElementAt(max(pred));
if (expected.Equals(output))
{
correct++;
}
}
total += cat.Count;
}
return((double)correct / (double)total);
}
//This will distort data proportions.
//Not recommended big size, take long time
public static model trainTextToGoal(model tm, double pg, textTrainDataProvider tdp, bool learnMode, bool fineMode)
{
model textModel = tm;
double percentGoal = pg / 100;
bool mode = false;
model prev = textModel;
double prevScore = 0;
while (true)
{
Console.WriteLine("[ttg]: Calculating percent score");
double score = testTextModel(textModel, tdp);
Console.WriteLine("[ttg]: Current percent score: " + score.ToString("0." + new string('#', 339)));
if (score.CompareTo(percentGoal) >= 0)
{
break;
}
Console.WriteLine("[ttg]: compare " + score.CompareTo(prevScore));
if ((score.CompareTo(prevScore) < 0) && (!mode) && fineMode)
{
Console.WriteLine("[ttg]: Switching mode");
mode = true;
textModel = prev;
}
else if ((score.CompareTo(prevScore) < 0) && (mode || !fineMode))
{
return(prev);
}
prev = textModel;
prevScore = score;
Console.WriteLine("[ttg]: Starting Pass");
for (int i = 0; i < tdp.data.Count; i++)
{
List <string[]> cat = tdp.getCatagory(i);
string expected = tdp.getCatID(i);
for (int j = 0; j < cat.Count; j++)
{
List <object> list = cat[j].ToList <object>();
Dictionary <object, double> pred = textModel.predict(list, true);
if (same(pred.Values.ToList()))
{
continue;
}
object output = pred.Keys.ElementAt(max(pred));
if (!expected.Equals(output))
{
//Donald Trump: WRONG!
//Maybe addfeature and remove feature are swaped because of the word: "the"?
for (int k = 0; k < textModel.catagories.Count; k++)
{
object catid = textModel.catagories[k].id;
if (output.Equals(catid))
{
//Penalize wrong output
if (mode)
{
for (int n = 0; n < list.Count; n++)
{
List <object> objectTest = new List <object>();
objectTest.Add(list[n]);
if (!expected.Equals(textModel.predict(objectTest, true)))
{
textModel.removeFeature(catid, list[n], 1);
}
else
{
textModel.addFeature(expected, list[n], 1);
}
}
}
else
{
for (int n = 0; n < list.Count; n++)
{
textModel.addFeature(catid, list[n], 1);
}
}
}
else if (expected.Equals(catid))
{
//re-enforce correct output
if (mode)
{
for (int n = 0; n < list.Count; n++)
{
List <object> objectTest = new List <object>();
objectTest.Add(list[n]);
if (expected.Equals(textModel.predict(objectTest, true)))
{
textModel.addFeature(expected, list[n], 1);
}
else
{
textModel.removeFeature(output, list[n], 1);
}
}
}
else
{
for (int n = 0; n < list.Count; n++)
{
textModel.removeFeature(catid, list[n], 1);
}
}
}
else if (learnMode)
{
//Not wrong but not right
//Penalize output
for (int n = 0; n < list.Count; n++)
{
textModel.removeFeature(catid, list[n], 1);
}
}
}
}
}
}
Console.WriteLine("[ttg]: Pass completed");
}
return(textModel);
}