/*
* /// <summary>
* /// Joins this data view to another data view on the given column.
* /// </summary>
* /// <param name="other"></param>
* /// <param name="columnName"></param>
* /// <returns></returns>
* public DataFrame Join (DataFrame other, string columnName)
* {
* int thisColumnIndex = this.GetColumnIndex(columnName);
* Type thisType = this.columns[thisColumnIndex].StorageType;
* int otherColumnIndex = other.GetColumnIndex(columnName);
* Type otherType = other.columns[otherColumnIndex].StorageType;
*
* // Form a lookup from the other table
* Dictionary<object, int> hash = new Dictionary<object, int>();
* for (int otherRowIndex = 0; otherRowIndex < other.Rows.Count; otherRowIndex++)
* {
* hash[other.columns[otherColumnIndex].GetItem(other.map[otherRowIndex])] = otherRowIndex;
* }
*
* // Construct the joined columns
* List<DataList> joinedColumns = new List<DataList>();
* for (int i = 0; i < this.columns.Count; i++)
* {
* DataList joinedColumn = DataList.Create(this.columns[i].Name, this.columns[i].StorageType);
* joinedColumns.Add(joinedColumn);
* }
* for (int j = 0; j < other.columns.Count; j++)
* {
* DataList joinedColumn = DataList.Create(other.columns[j].Name, other.columns[j].StorageType);
* joinedColumns.Add(joinedColumn);
* }
*
* // Populate the joined columns
* for (int thisRowIndex = 0; thisRowIndex < this.map.Count; thisRowIndex++)
* {
* object thisValue = this.columns[thisColumnIndex].GetItem(this.map[thisRowIndex]);
* int otherRowIndex;
* if (hash.TryGetValue(thisValue, out otherRowIndex))
* {
* for (int i = 0; i < this.columns.Count; i++)
* {
* joinedColumns[i].AddItem(this.columns[i].GetItem(this.map[i]));
* }
* for (int j = 0; j < other.columns.Count; j++)
* {
* joinedColumns[this.columns.Count + j].AddItem(other.columns[j].GetItem(other.map[otherRowIndex]));
* }
* }
* }
*
* DataFrame result = new DataFrame(joinedColumns);
* return (result);
*
* }
*/
internal void AddColumn(NamedList column)
{
if (column == null)
{
throw new ArgumentNullException(nameof(column));
}
if (this.columns.Count == 0)
{
// This is the first column; create a row map.
for (int i = 0; i < column.Count; i++)
{
this.map.Add(i);
}
}
else
{
// This is not the first column; if it isn't computed, it's length must match the existing columns.
if (!column.IsComputed && column.Count != map.Count)
{
throw new DimensionMismatchException();
}
}
this.columnMap.Add(column.Name, this.columns.Count);
this.columns.Add(column);
}
public ConvertedFrameColumn(NamedList column, List <int> map)
{
Debug.Assert(column != null);
Debug.Assert(map != null);
this.column = column;
this.map = map;
}
/// <summary>
/// Groups the data by the values in the given column, and computes aggregate quantities for each group.
/// </summary>
/// <param name="groupByColumnName">The name of the column to group by.</param>
/// <param name="aggregator">A function that computes the aggregate quantities.</param>
/// <returns>A new data frame containing the aggregates for each group.</returns>
/// <remarks>
/// <para>The first column of the returned <see cref="FrameTable"/> has the same name as the
/// original <paramref name="groupByColumnName"/> and contains all the distinct
/// values of that column in the original view. There is an additional column for each
/// dictionary entry returned by <paramref name="aggregator"/>, whose name is the returned
/// key and whose values are values returned for each group.</para>
/// <para>The function that computes the aggregate receives a <see cref="FrameView"/> containing
/// all the rows in the group. To produce aggregate results, it can use values in any of
/// the columns. Each invocation of the <paramref name="aggregator"/> must return the same keys
/// and values for the same keys must be of the same type. (Values for different keys may be
/// of different types.) Aggregate column names are taken from the keys and storage types are
/// inferred from the returned values.</para>
/// <para>To produce just one aggregate value, you may find it simpler and more efficient
/// to use the <see cref="GroupBy(string, Func{FrameView, IReadOnlyDictionary{string, object}})"/>
/// overload.</para>
/// </remarks>
public FrameTable GroupBy(string groupByColumnName, Func <FrameView, IReadOnlyDictionary <string, object> > aggregator)
{
// Collect rows into groups.
int groupByColumnIndex = GetColumnIndex(groupByColumnName);
NamedList groupByColumn = columns[groupByColumnIndex];
NullableDictionary <object, List <int> > groups = FindGroups(groupByColumn);
// Create a column to hold the group values.
NamedList groupsColumn = NamedList.Create(groupByColumnName, groupByColumn.StorageType);
// Create an enumerator that feeds the groups into the aggregator and presents them as dictionaries.
IEnumerable <IReadOnlyDictionary <string, object> > aggregatesEnumerator = GetGroupEnumerator(groups, aggregator, groupsColumn);
// Column-ify and validate the presented dictionaries.
List <NamedList> aggregateColumns = DictionaryHelper.ReadDictionaries(aggregatesEnumerator);
// Collect the results into a frame table.
FrameTable result = new FrameTable();
// First column is the group values.
result.AddColumn(groupsColumn);
// Remaining columns are aggregate columns.
foreach (NamedList aggregateColumn in aggregateColumns)
{
result.AddColumn(aggregateColumn);
}
return(result);
}
private static void ReadCsvAsStrings(TextReader reader, out NamedList <string>[] columns, out DataAdaptor[] headers)
{
Debug.Assert(reader != null);
// Get the column names from the first line.
string firstline = reader.ReadLine();
if (firstline == null)
{
columns = null;
headers = null;
return;
}
List <string> names = CsvHelper.ReadCells(firstline);
int count = names.Count;
// Put the columns into lists of strings, and as we do so, maintain the collection of
// types it can be parsed into, and whether any entries are null.
columns = new NamedList <string> [names.Count];
headers = new DataAdaptor[names.Count];
for (int columnIndex = 0; columnIndex < columns.Length; columnIndex++)
{
columns[columnIndex] = new NamedList <string>(names[columnIndex]);
headers[columnIndex] = new DataAdaptor();
}
while (true)
{
string line = reader.ReadLine();
if (line == null)
{
break;
}
List <string> cells = CsvHelper.ReadCells(line);
if (cells.Count != count)
{
throw new FormatException();
}
for (int columnIndex = 0; columnIndex < count; columnIndex++)
{
string cell = cells[columnIndex];
if (String.IsNullOrEmpty(cell))
{
headers[columnIndex].IsNullable = true;
columns[columnIndex].Add(null);
}
else
{
DataAdaptor header = headers[columnIndex];
header.TryParse(cell);
columns[columnIndex].Add(cell);
}
}
}
}
/// <summary>
/// Creates a new frame table from a file of comma-separated values.
/// </summary>
/// <param name="reader">A reader positioned at the beginning of the file.</param>
/// <returns>A new data frame with data from the file.</returns>
/// <remarks>
/// <para>The column names are taken from the first line of the file.</para>
/// <para>The storage type of each column is inferred from the types of objects
/// encountered are the frame table is constructed.</para>
/// </remarks>
public static FrameTable FromCsv(TextReader reader)
{
if (reader == null)
{
throw new ArgumentNullException(nameof(reader));
}
NamedList <string>[] textColumns;
DataAdaptor[] headers;
ReadCsvAsStrings(reader, out textColumns, out headers);
NamedList[] columns = new NamedList[headers.Length];
for (int columnIndex = 0; columnIndex < columns.Length; columnIndex++)
{
DataAdaptor header = headers[columnIndex];
if (header.TypeCandidates.Count == 0)
{
columns[columnIndex] = textColumns[columnIndex];
}
else
{
TypeParser adaptor = header.TypeCandidates.First.Value;
NamedList column = adaptor.CreateStorage(textColumns[columnIndex].Name, header.IsNullable);
foreach (string textValue in textColumns[columnIndex])
{
if (textValue == null)
{
column.AddItem(null);
}
else
{
object value = adaptor.Parse(textValue);
column.AddItem(value);
}
}
columns[columnIndex] = column;
}
}
FrameTable frame = new FrameTable(columns);
return(frame);
}
// This method collects rows into groups by the group-defining column. It is shared by
// both GroupBy overloads.
private NullableDictionary <object, List <int> > FindGroups(NamedList groupByColumn)
{
NullableDictionary <object, List <int> > groups = new NullableDictionary <object, List <int> >();
for (int r = 0; r < this.map.Count; r++)
{
int index = this.map[r];
object value = groupByColumn.GetItem(index);
List <int> members;
if (!groups.TryGetValue(value, out members))
{
members = new List <int>();
groups.Add(value, members);
}
members.Add(index);
}
return(groups);
}
/// <summary>
/// Sort the rows by the values in the given column in the given direction.
/// </summary>
/// <param name="columnName">The name of the column to sort by.</param>
/// <param name="order">The direction of the ordering.</param>
/// <returns>A new view, with rows sorted by the values in the given column.</returns>
/// <remarks>
/// <para><see langword="null"/> values are supported and are ordered before all other values.</para>
/// <para>The type of data in the column must implement <see cref="IComparable"/>.</para>
/// </remarks>
/// <exception cref="ArgumentNullException"><paramref name="columnName"/> is <see langword="null"/>.</exception>
/// <exception cref="IndexOutOfRangeException"><paramref name="columnName"/> is not the name of a column in the view.</exception>
/// <exception cref="InvalidCastException">The type of data in the column is not <see cref="IComparable"/>.</exception>
public FrameView OrderBy(string columnName, SortOrder order)
{
if (columnName == null)
{
throw new ArgumentNullException(nameof(columnName));
}
int columnIndex = GetColumnIndex(columnName);
NamedList column = columns[columnIndex];
List <int> newMap = new List <int>(map);
if (order == SortOrder.Ascending)
{
newMap.Sort((i, j) => NullableComparer((IComparable)column.GetItem(i), (IComparable)column.GetItem(j)));
}
else
{
newMap.Sort((i, j) => NullableComparer((IComparable)column.GetItem(j), (IComparable)column.GetItem(i)));
}
return(new FrameView(this.columns, newMap));
}
/// <summary>
/// Groups the data by the values in the given column, and computes the given aggregate quantity for each group.
/// </summary>
/// <typeparam name="T">The type of the aggregate output.</typeparam>
/// <param name="groupByColumnName">The name of the column to group by.</param>
/// <param name="aggregateColumnName">The name of the column for the aggregate output.</param>
/// <param name="aggregator">A function that computes the aggregate quantity.</param>
/// <returns>A new data frame containing the requested aggregate values for each group.</returns>
/// <remarks>
/// <para>The function that computes the aggregate receives a <see cref="FrameView"/> containing
/// all the rows in the group. To produce an aggregate result, it can use values in any of
/// the columns.</para>
/// <para>To produce more than one aggregate value, use <see cref="GroupBy(string, Func{FrameView, IReadOnlyDictionary{string, object}})"/>.</para>
/// </remarks>
public FrameTable GroupBy <T>(string groupByColumnName, Func <FrameView, T> aggregator, string aggregateColumnName)
{
if (groupByColumnName == null)
{
throw new ArgumentNullException(nameof(groupByColumnName));
}
if (aggregator == null)
{
throw new ArgumentNullException(nameof(aggregator));
}
if (aggregateColumnName == null)
{
throw new ArgumentNullException(nameof(aggregateColumnName));
}
// Collect the rows into groups.
int groupByColumnIndex = GetColumnIndex(groupByColumnName);
NamedList groupByColumn = columns[groupByColumnIndex];
NullableDictionary <object, List <int> > groups = FindGroups(groupByColumn);
// Form destination columns based on group aggregates.
NamedList groupsColumn = NamedList.Create(groupByColumnName, groupByColumn.StorageType);
NamedList <T> aggregateColumn = new NamedList <T>(aggregateColumnName);
foreach (KeyValuePair <object, List <int> > group in groups)
{
FrameView values = new FrameView(this.columns, group.Value);
T aggregateValue = aggregator(values);
aggregateColumn.AddItem(aggregateValue);
object groupKey = group.Key;
groupsColumn.AddItem(groupKey);
}
FrameTable result = new FrameTable(groupsColumn, aggregateColumn);
return(result);
}
/// <summary>
/// Adds a new row of data to the data frame.
/// </summary>
/// <param name="values">A dictionary that maps the existing column names to the cell values for the new row.</param>
public void AddRow(IReadOnlyDictionary <string, object> values)
{
if (values == null)
{
throw new ArgumentNullException(nameof(values));
}
int rowCount = map.Count;
for (int columnIndex = 0; columnIndex < columns.Count; columnIndex++)
{
NamedList column = columns[columnIndex];
if (column.IsComputed)
{
continue;
}
object value = values[column.Name];
int rowIndex = column.AddItem(value);
if (rowIndex != rowCount)
{
throw new InvalidOperationException();
}
}
map.Add(rowCount);
}
public static List <NamedList> ReadDictionaries(IEnumerable <IReadOnlyDictionary <string, object> > dictionaries)
{
Debug.Assert(dictionaries != null);
// Iterate through the dictionaries, creating header objects that contain the un-cast values
// and some information about them.
List <DictionaryColumn> headers = null;
foreach (IReadOnlyDictionary <string, object> dictionary in dictionaries)
{
// From the first row, create the headers list based on key names.
if (headers == null)
{
headers = new List <DictionaryColumn>(dictionary.Count);
foreach (string key in dictionary.Keys)
{
DictionaryColumn header = new DictionaryColumn()
{
Name = key, IsNullable = false, Type = null, Data = new List <object>()
};
headers.Add(header);
}
}
if (dictionary.Count != headers.Count)
{
throw new InvalidOperationException();
}
// For all rows, check for null, record the type if we haven't found it yet, and store the value.
for (int i = 0; i < headers.Count; i++)
{
DictionaryColumn header = headers[i];
object value = dictionary[header.Name];
if (value == null)
{
header.IsNullable = true;
}
else
{
if (header.Type == null)
{
header.Type = value.GetType();
}
}
header.Data.Add(value);
}
}
// Arrange the columns into named lists of the appropriate type
List <NamedList> columns = new List <NamedList>(headers.Count);
foreach (DictionaryColumn header in headers)
{
NamedList column;
if (header.Type == null)
{
// If no non-null value was ever found, we can't infer a type, so just make an object-column.
column = new NamedList <object>(header.Name, header.Data);
}
else
{
// Based on null-ability and observed type, create the appropriate storage.
Type type = header.Type;
if (header.IsNullable && type.GetTypeInfo().IsValueType)
{
type = typeof(Nullable <>).MakeGenericType(type);
}
column = NamedList.Create(header.Name, type);
// Copy the objects into the storage, which will cast them to the storage type.
foreach (object value in header.Data)
{
column.AddItem(value);
}
}
columns.Add(column);
}
Debug.Assert(columns.Count == headers.Count);
return(columns);
}
// This method turns our group row dictionary into an iterator that can be fed into the dictionary parser.
// As it iterates to produce the views and aggregates in turn, it also adds the group values to the given
// groups column. This logic is so closely coupled to the internal logic of the GroupBy method that calls
// it that I would rather do this via lambda inside it, but lambdas that produce iterators are not allowed
// and I really want an iterator to feed into the shared dictionary parsing logic.
private IEnumerable <IReadOnlyDictionary <string, object> > GetGroupEnumerator(NullableDictionary <object, List <int> > groups, Func <FrameView, IReadOnlyDictionary <string, object> > aggregator, NamedList groupsColumn)
{
foreach (KeyValuePair <object, List <int> > group in groups)
{
FrameView view = new FrameView(this.columns, group.Value);
IReadOnlyDictionary <string, object> aggregate = aggregator(view);
yield return(aggregate);
groupsColumn.AddItem(group.Key);
}
}
