private static (DataFrame edf, float[] fValues) ScaleColumn(DataFrame dff, string colName, ColumnTransformer transformer, bool transformedColumnsOnly)
{
var newColName = colName + "_scaled";
var df = dff[dff.Columns.ToArray()];
var s = Series.FromDataFrame(df, colName);
float param1;
float param2;
if (transformer == ColumnTransformer.MinMax)
{
var minObj = s.Aggregate <float>(Aggregation.Min);
param1 = Convert.ToSingle(minObj);
var maxObj = s.Aggregate <float>(Aggregation.Max);
param2 = Convert.ToSingle(maxObj);
df.AddCalculatedColumn(newColName, (IDictionary <string, object> row, int i) =>
{
var val = Convert.ToDouble(row[colName]);
return((val - param1) / (param2 - param1));
});
}
else if (transformer == ColumnTransformer.MinMax ||
transformer == ColumnTransformer.Standardizer)
{
var avgObj = s.Aggregate <float>(Aggregation.Avg);
param1 = Convert.ToSingle(avgObj);
var stdObj = s.Aggregate <float>(Aggregation.Std);
param2 = Convert.ToSingle(stdObj);
df.AddCalculatedColumn(colName + "_scaled", (IDictionary <string, object> row, int i) =>
{
var val = Convert.ToDouble(row[colName]);
return((val - param1) / (param2));
});
}
else
{
throw new NotSupportedException("Column transformation is not supported.");
}
if (transformedColumnsOnly)
{
var ddf = df.Create((newColName, null));
return(ddf, new float[] { param1, param2 });
}
else
{
return(df, new float[] { param1, param2 });
}
}
public static (DataFrame df, float[] scaledValues, string[] labels) TransformColumn(this DataFrame df, string colName, ColumnTransformer transformer, bool transformedColumnsOnly = false)
{
switch (transformer)
{
case ColumnTransformer.None:
return(df, null, null);
case ColumnTransformer.Binary1:
(var edf, var cValues) = BinaryEncoding(df, colName, transformedColumnsOnly);
return(edf, null, cValues);
case ColumnTransformer.Binary2:
(edf, cValues) = BinaryEncoding2(df, colName, transformedColumnsOnly);
return(edf, null, cValues);
case ColumnTransformer.Ordinal:
(edf, cValues) = OrdinalEncoding(df, colName, transformedColumnsOnly);
return(edf, null, cValues);
case ColumnTransformer.OneHot:
(edf, cValues) = OneHotEncodeColumn(df, colName, transformedColumnsOnly);
return(edf, null, cValues);
case ColumnTransformer.Dummy:
(edf, cValues) = DummyEncodeColumn(df, colName, transformedColumnsOnly);
return(edf, null, cValues);
case ColumnTransformer.MinMax:
case ColumnTransformer.Standardizer:
(var tdf, float[] fValues) = ScaleColumn(df, colName, transformer, transformedColumnsOnly);
return(tdf, fValues, null);
default:
throw new NotSupportedException("Data normalization is not supported.");
}
throw new NotImplementedException();
}
