private TreeNode CreateNode(List <EvaluatedColumnPredicate> evaluatedParent, DrugDataset dataset)
{
var nodeCandidates = new Tuple <double, TreeNode>(double.MaxValue, null);
EvaluatedColumnPredicate evaluationCandidate = new EvaluatedColumnPredicate();
double entropieCandidate = -1;
_remainingColumns.ForEach(column =>
{
if (!_columnsInConsideration.Contains(column))
{
return;
}
var values = dataset.GetTrainingRows(column, _trainingRatio).Select(c => double.Parse(c)).ToList();
var valuePredicateRange = DeterminePredicateRange(values);
for (int i = 0; i < _resultPredicateRange.Item1; i++)
{
for (int j = 0; j < valuePredicateRange.Item1; j++)
{
var resultPredicate = new Predicate <double>(d => d >= i * _resultPredicateRange.Item2 && d < (i + 1) * _resultPredicateRange.Item2);
var minExp = 0.ToString();
var maxExp = (valuePredicateRange.Item1 * valuePredicateRange.Item2 - ((j + 1) * valuePredicateRange.Item2)).ToString(CultureInfo.InvariantCulture);
//var minExp = (j * valuePredicateRange.Item2).ToString();
//var maxExp = ((j + 1) * valuePredicateRange.Item2).ToString();
var reg = new Regex(@"^[\d]+\.[\d]+$");
if (!reg.IsMatch(minExp))
{
minExp += ".0";
}
if (!reg.IsMatch(maxExp))
{
maxExp += ".0";
}
double test = 0;
if (!double.TryParse(maxExp, out test))
{
continue;
}
var valuePredicate = new Predicate <double>(d => d >= double.Parse(minExp) && d < double.Parse(maxExp));
var trueFilterExpression = $"{column} >= {minExp} AND {column} < {maxExp}";
var falseFilterExpression = $"NOT ({column} >= {minExp} AND {column} < {maxExp})";
var evaluation = new EvaluatedColumnPredicate {
Column = column, PredicateDescription = trueFilterExpression
};
if (evaluatedParent.Contains(evaluation))
{
continue;
}
var results = dataset.GetTrainingRows("Nicotine", _trainingRatio).Select(c => double.Parse(c)).ToList();
var entropie = CalculateEntropie(results, values, valuePredicate, resultPredicate);
if (entropie < 0 || entropie > 1)
{
Logger.LogError($"Calculated entropie on column {column} with predicate {trueFilterExpression} is not valid => {entropie}");
}
if (entropie < nodeCandidates.Item1 && entropie > _stopEntropie)
{
evaluationCandidate = evaluation;
entropieCandidate = entropie;
nodeCandidates = new Tuple <double, TreeNode>(entropie, new TreeNode
{
ChildFalse = null,
ChildTrue = null,
Column = column,
Predicate = valuePredicate,
PredicateMaxExp = maxExp,
PredicateMinExp = minExp,
TrueFilterExpression = trueFilterExpression,
FalseFilterExpression = falseFilterExpression,
Result = i * _resultPredicateRange.Item2
});
}
if (_resultPredicateRange.Item2 < 0.16 || _resultPredicateRange.Item2 > 0.17)
{
Logger.LogError("Worng result");
}
}
}
});
evaluatedParent.Add(evaluationCandidate);
if (nodeCandidates.Item2 != null)
{
Logger.LogMessage($"Treenode created with column {nodeCandidates.Item2?.Column}, entropie = {entropieCandidate}, predicate = {nodeCandidates.Item2?.TrueFilterExpression}, result = {nodeCandidates.Item2?.Result}");
nodeCandidates.Item2.ChildFalse = CreateNode(evaluatedParent, CreateFilteredDataset(dataset, nodeCandidates.Item2.TrueFilterExpression));
nodeCandidates.Item2.ChildTrue = CreateNode(evaluatedParent, CreateFilteredDataset(dataset, nodeCandidates.Item2.FalseFilterExpression));
}
return(nodeCandidates.Item2);
}
private void InitResult()
{
var results = _dataset.GetTrainingRows("Nicotine", _trainingRatio).Select(c => double.Parse(c)).ToList();
_resultPredicateRange = DeterminePredicateRange(results);
}