public static void TrainFromCSV(Policy thePolicy, string labelFeaturName, string csvAddress, int readLimit)
{
// Read the file and display it line by line.
System.IO.StreamReader file = new System.IO.StreamReader(csvAddress);
//Read headers
string[] headers = file.ReadLine().Split(',');
double[] rewards = file.ReadLine().Split(',').Select(double.Parse).ToArray();
//Read data into feature vector
string line; int counter = 0;
while ((line = file.ReadLine()) != null)
{
//increment counter to use as import id
counter++;
//Read a line to a string array
string[] dataobjects = line.Split(',');
//Create a data vector from the headers and read data line
DataVectorTraining dataVector = new DataVectorTraining(headers, dataobjects, rewards, labelFeaturName);
//Submit to the reinforcement learner
thePolicy.Learn(dataVector);
//If limit reached, end early
if (counter == readLimit)
{
break;
}
}
file.Close();
}
public void Learn_xorOperation_SeeAssert(int iterations)
{
///Why does this sometimes fail? There are no bad examples.
string[] headers = new string[]
{
"bool1",
"bool2",
"xor"
};
double[] importance = new double[] {
0,
0,
0
};
Policy thePolicy = new Policy();
#region Training
Random rand = new Random();
for (int i = 0; i < iterations; i++)
{
bool bool1 = Convert.ToBoolean(rand.Next(0, 2));
bool bool2 = Convert.ToBoolean(rand.Next(0, 2));
object[] data = new object[] { bool1, bool2, (bool1 ^ bool2) };
DataVectorTraining dvt = new DataVectorTraining(headers, data, importance, "xor");
thePolicy.Learn(dvt);
}
#endregion
string htmlTree = thePolicy.ToDecisionTree(new DecisionTree.TreeSettings()
{
ShowSubScores = true, ShowBlanks = true
}).ToHtmlTree();
System.IO.File.WriteAllText("xor" + iterations + ".html", htmlTree);
#region Assert
var dataVectorTT = new DataVector(
new string[] { "bool1", "bool2" },
new object[] { true, true });
var dataVectorFF = new DataVector(
new string[] { "bool1", "bool2" },
new object[] { false, false });
var dataVectorTF = new DataVector(
new string[] { "bool1", "bool2" },
new object[] { true, false });
var dataVectorFT = new DataVector(
new string[] { "bool1", "bool2" },
new object[] { false, true });
Assert.False(Convert.ToBoolean(thePolicy.Classify_ByPolicy(dataVectorTT)));
Assert.False(Convert.ToBoolean(thePolicy.Classify_ByTree(dataVectorTT)));
Assert.False(Convert.ToBoolean(thePolicy.Classify_ByPolicy(dataVectorFF)));
Assert.False(Convert.ToBoolean(thePolicy.Classify_ByTree(dataVectorFF)));
Assert.True(Convert.ToBoolean(thePolicy.Classify_ByPolicy(dataVectorTF)));
Assert.True(Convert.ToBoolean(thePolicy.Classify_ByTree(dataVectorTF)));
Assert.True(Convert.ToBoolean(thePolicy.Classify_ByPolicy(dataVectorFT)));
Assert.True(Convert.ToBoolean(thePolicy.Classify_ByTree(dataVectorFT)));
#endregion
}
public void Constructor_DataVector_OnlyQueries()
{
DataVectorTraining dvt = new DataVectorTraining(
new string[] { "name1", "name2", "class" },
new object[] { 0.0, 0.5, "c" },
new double[] { 0.0, 0.0, 0.0 },
"class"
);
State s = new State(dvt);
}
public void Constructor_WrongClassName_Exception()
{
string[] featureNames = new string[] { "name1", "name2", "class" };
object[] featureValues = new object[] { 1.0, 2.0, 3.0 };
double[] importance = new double[] { -1.0, 0.0, 0.0 };
string labelFeatureName = "class2";
Assert.ThrowsAny <ArgumentException>(delegate {
DataVectorTraining dvt = new DataVectorTraining(featureNames, featureValues, importance, labelFeatureName);
});
}
public DataVectorTraining ReadLine(string labelFeatureName)
{
//Try to read a line
string line = file.ReadLine();
if (line == null)
{
return(null);
}
//Read a line to a string array
string[] dataobjects = line.Split(',');
lineCounter++;
//Build Datavector
DataVectorTraining dvt = new DataVectorTraining();
foreach (var fieldNameKVP in fieldNames)
{
//Get feaure name and value
string name = fieldNameKVP.Key;
string value = dataobjects[fieldNameKVP.Value];
//Get importance value, if available
double importance = 0;
if (fieldImportances.ContainsKey(name))
{
double.TryParse(dataobjects[fieldImportances[name]], out importance);
}
//Add to the datavector
dvt.AddFeature(name, value, importance);
//Track seen key-value pairs (expect the class label)
if (name != labelFeatureName)
{
if (!fieldUniqueValues.ContainsKey(name))
{
fieldUniqueValues.Add(name, new HashSet <object>());
}
fieldUniqueValues[name].Add(value);
}
}
//Set the label
dvt.SetLabel(labelFeatureName, dvt[labelFeatureName].Value);
return(dvt);
}
public void Constructor_With2Inputs_2()
{
string[] featureNames = new string[] { "name1", "name2", "class" };
object[] featureValues = new object[] { 1.0, 2.0, 3.0 };
double[] importance = new double[] { -1.0, 0.0, 0.0 };
string labelFeatureName = "class";
DataVectorTraining dvt = new DataVectorTraining(featureNames, featureValues, importance, labelFeatureName);
Assert.Equal(2, dvt.Features.Count);
Assert.Equal(1.0, dvt["name1"].Value);
Assert.Equal(-1.0, dvt["name1"].Importance);
Assert.Equal(2.0, dvt["name2"].Value);
Assert.Equal(0.0, dvt["name2"].Importance);
Assert.Equal("class", dvt.Label.Name);
Assert.Equal(3.0, dvt.Label.Value);
}
