//Instance call
public string Predict(DataRow dataInput)
{
var valuesForQuery = new Dictionary <string, string>();
for (int i = 0; i < Dataset.Columns.Count - 1; i++) // exclude resolution column
{
var input = dataInput[i].ToString();
var colName = Dataset.Columns[i].ColumnName;
if (CategoricalFactory.IsDescriptorExists(colName))
{
input = CategoricalFactory.DescriptNumericalValue(input, colName);
}
valuesForQuery.Add(Dataset.Columns[i].ToString(), input);
}
return(Predict(Root, valuesForQuery, ""));
}
public static DataTable CSVtoDataTable(string strFilePath, ref DataTable testSet, Random rnd)
{
DataTable dt = new DataTable();
var lines = File.ReadAllLines(strFilePath).ToList();
// Preserve coma inside quotes
for (int i = 0; i < lines.Count; i++)
{
var s = lines[i];
if (Regex.IsMatch(s, "\"[^\"]+\""))
{
foreach (Match m in Regex.Matches(s, "\"[^\"]+\""))
{
var ori = m.Value;
var replace = ori.Replace(",", "_COMA_").Replace("\"", "");
lines[i] = s.Replace(ori, replace);
}
}
}
// get headers and exclude header row
string[] headers = lines[0].Split(',');
lines.RemoveAt(0);
// Randomize data order, using Fisher-Yate algorithm
int n = lines.Count;
for (int i = 0; i < n; i++)
{
int cc = i + rnd.Next(n - i);
string t = lines[cc];
lines[cc] = lines[i];
lines[i] = t;
}
string[,] tableVals = new string[lines.Count, headers.Length];
string[] modes = new string[headers.Length];
foreach (string header in headers)
{
dt.Columns.Add(header);
}
int r = 0;
for (int q = 0; q < lines.Count; q++)
{
string[] rowVals = lines[q].Split(',');
for (int i = 0; i < headers.Length; i++)
{
tableVals[q, i] = rowVals[i].Replace("_COMA_", ",");
}
r++;
}
// Calculate mode for each column, for missing data subtitution
for (int i = 0; i < headers.Length; i++)
{
string[] vals = new string[tableVals.GetLength(0)];
for (int j = 0; j < tableVals.GetLength(0); j++)
{
vals[j] = tableVals[j, i];
}
modes[i] = vals.Where(a => a != "?").GroupBy(a => a).OrderByDescending(a => a.Count()).First().Key;
}
// Replace missing data with Mode
int l = tableVals.GetLength(0);
for (int i = 0; i < l; i++)
{
DataRow dr = dt.NewRow();
for (int j = 0; j < headers.Length; j++)
{
var s = tableVals[i, j];
if (IsMissingValue(s))
{
s = modes[j];
}
dr[j] = s;
}
dt.Rows.Add(dr);
}
// Take 30% as Test Set, return the rest
testSet = dt.Clone();
var count = (int)Math.Round(dt.Rows.Count * 0.3, 0);
for (int i = 0; i < count; i++)
{
testSet.Rows.Add(dt.Rows[i].ItemArray);
dt.Rows.RemoveAt(i);
}
// Create descriptor for numerical values
for (int i = 0; i < dt.Columns.Count; i++)
{
var distinctVals = Feature.GetDistinctValuesOfColumn(dt, i);
if (distinctVals.Count > 8 && distinctVals.All(a => double.TryParse(a, out double ou)))
{
// values are numerical, create descriptor
var descriptor = CategoricalFactory.GenerateEqualWidthBins(dt.Columns[i].ColumnName, distinctVals.Select(a => double.Parse(a)).ToArray());
for (int s = 0; s < dt.Rows.Count; s++)
{
dt.Rows[s][i] = descriptor.DescriptNumericalValue(dt.Rows[s][i].ToString());
}
}
}
return(dt);
}
