public AnswerTreeNode(DatumSchema schema, string answer, Datum[] data, int level)
{
Level = level;
if (schema.AnswerValidator(answer))
{
/* for (int i = 0; i < Level; i++)
* Console.Write(" |");
* Console.WriteLine(" Answer: {0}", answer); */
Answer = answer;
for (int i = 0; i < data.Length; i++)
{
if (data[i].Answer != Answer)
{
throw new InvalidOperationException(String.Format(
"The datum {0} does not have the answer {1}.", data[i], Answer));
}
}
Data = data;
Working.Printf("Every element in the current data subset (\\{{{0}\\}}) now " +
"all share the same answer of \\texttt{{{1}}}. Hence, " +
"that is the answer that should be declared given this " +
"node of the decision tree has been reached.",
String.Join(", ", Data.Select(d => d.ToString())),
Answer);
}
else
{
throw new InvalidOperationException(String.Format(
"The answer {0} is not valid with the given schema.", answer));
}
}
private void Compute()
{
var tableAttributes = Schema.Attributes.Where(a => !a.IsQueryable).Concat(DecidableAttributes);
Working.Print("Current data subset:");
Working.Print("");
Working.Printf(@"\begin{{tabular}}{{{0}l}}", "l ".Repeat(tableAttributes.Count()));
Working.Printf(@" {0}{1} \\",
String.Join(" & ", tableAttributes
.Select(a => String.Format(@"\textbf{{{0}}}", a.Name))),
@" & \textbf{Answer}");
foreach (var datum in Data)
{
Working.Printf(@" {0} & \textbf{{{1}}} \\",
String.Join(" & ", tableAttributes
.Select(a => datum[a])),
datum.Answer);
}
Working.Print(@"\end{tabular}");
Working.Print("");
var entropyData = Data
.GroupBy(d => d.Answer)
.Select(g => new { Answer = g.Key, Count = (double)g.Count() });
Working.Printf("Entropy calculation: {0}.",
String.Join(", ",
entropyData
.Select(i => String.Format(@"\texttt{{{0}}} occurs $ {1} $ time{2}",
i.Answer,
i.Count,
i.Count == 1 ? "" : "s"))));
var entropy = Entropy(entropyData
.Select(ans => ans.Count));
Working.Printf("$$ {0}={1:0.######} $$",
EntropyWorking(entropyData.Select(e => (int)e.Count), (int)entropyData.Sum(d => d.Count)),
entropy);
Dictionary <Attribute, double> attributeGains = new Dictionary <Attribute, double>();
foreach (Attribute attribute in DecidableAttributes)
{
double remainder = Remainder(Data, attribute),
gain = entropy - remainder;
var remainderData = Data
.GroupBy(d => d[attribute])
.Select(g => new { Value = g.Key, Count = g.Count() });
Working.Printf(@"Remainder calculation for \texttt{{{1}}} as follows. " +
@"Number of occurrences for each value of \texttt{{{1}}}: {0}.",
String.Join(", ",
remainderData
.Select(i => String.Format(@"\texttt{{{0}}} occurs $ {1} $ time{2}",
i.Value,
i.Count,
i.Count == 1 ? "" : "s"))),
attribute.Name);
Working.Printf("$$ Remainder({2})={0}={1:0.######} $$",
String.Join("+", remainderData
.Select(g => String.Format(@"\frac{{{0}}}{{{1}}}\left({2}\right)", g.Count, Data.Count(),
EntropyWorking(
Data
.Where(d => d[attribute] == g.Value)
.GroupBy(d => d.Answer)
.Select(g2 => g2.Count()), g.Count)))),
remainder,
attribute.Name);
Working.Printf("Hence, $ Gain({0}) = H - Remainder({0}) = {1:0.######} $.",
attribute.Name,
gain);
Working.Print("");
attributeGains.Add(attribute, gain);
}
var questionAttributeGain = attributeGains
.OrderByDescending(kvp => kvp.Value)
.First();
Working.Printf(@"The information gain from \texttt{{{0}}} is the largest, at $ {1:0.######} $ bits - " +
@"therefore, this attribute should form the next decision.",
questionAttributeGain.Key,
questionAttributeGain.Value);
QuestionAttribute = questionAttributeGain.Key;
Children = new Dictionary <string, ITreeNode>();
var byBest = Data
.GroupBy(d => d[QuestionAttribute]);
Answers = byBest
.Select(g => g.Key)
.ToArray();
foreach (var group in byBest)
{
/* for (int i = 0; i < Level; i++)
* Console.Write(" |");
* Console.WriteLine(" If {0} = {1}:", QuestionAttribute, group.Key); */
Working.Print("");
Working.Printf(@"Assume \texttt{{{0}}} was chosen for the attribute \texttt{{{1}}}.",
group.Key,
QuestionAttribute);
Children.Add(group.Key,
group.Count() == 1 || group.AllEqual(v => v.Answer) ?
(ITreeNode)(new AnswerTreeNode(Schema, group.First().Answer, group, Level + 1)) :
(ITreeNode)(new QuestionTreeNode(Schema, group,
DecidableAttributes.Where(a => a != QuestionAttribute).ToArray(),
Level + 1,
KnownValues.Add(QuestionAttribute, group.Key))));
}
Working.Print("");
Working.Printf(@"This accounts for every possibility of the attribute \texttt{{{0}}} " +
"at this level of the decision tree.",
QuestionAttribute);
}
