public DataFrame ElementwiseGreaterThanOrEqual <T>(IReadOnlyList <T> values)
where T : unmanaged
{
if (values.Count != Columns.Count)
{
throw new ArgumentException(Strings.MismatchedColumnLengths, nameof(values));
}
DataFrame retDataFrame = new DataFrame();
for (int i = 0; i < Columns.Count; i++)
{
DataFrameColumn baseColumn = _columnCollection[i];
DataFrameColumn newColumn = baseColumn.ElementwiseGreaterThanOrEqual(values[i]);
retDataFrame.Columns.Insert(i, newColumn);
}
return(retDataFrame);
}
/// <summary>
/// Performs a reversed element-wise boolean Xor on each column
/// </summary>
public DataFrame ReverseXor(bool value, bool inPlace = false)
{
DataFrame retDataFrame = inPlace ? this : new DataFrame();
for (int i = 0; i < Columns.Count; i++)
{
DataFrameColumn baseColumn = _columnCollection[i];
DataFrameColumn newColumn = baseColumn.ReverseXor(value, inPlace);
if (inPlace)
{
retDataFrame.Columns[i] = newColumn;
}
else
{
retDataFrame.Columns.Insert(i, newColumn);
}
}
return(retDataFrame);
}
/// <summary>
/// Performs an element-wise subtraction on each column
/// </summary>
public DataFrame Subtract <T>(T value, bool inPlace = false)
where T : unmanaged
{
DataFrame retDataFrame = inPlace ? this : new DataFrame();
for (int i = 0; i < Columns.Count; i++)
{
DataFrameColumn baseColumn = _columnCollection[i];
DataFrameColumn newColumn = baseColumn.Subtract(value, inPlace);
if (inPlace)
{
retDataFrame.Columns[i] = newColumn;
}
else
{
retDataFrame.Columns.Insert(i, newColumn);
}
}
return(retDataFrame);
}
public DataFrame Xor(IReadOnlyList <bool> values, bool inPlace = false)
{
if (values.Count != Columns.Count)
{
throw new ArgumentException(Strings.MismatchedColumnLengths, nameof(values));
}
DataFrame retDataFrame = inPlace ? this : new DataFrame();
for (int i = 0; i < Columns.Count; i++)
{
DataFrameColumn baseColumn = _columnCollection[i];
DataFrameColumn newColumn = baseColumn.Xor(values[i], inPlace);
if (inPlace)
{
retDataFrame.Columns[i] = newColumn;
}
else
{
retDataFrame.Columns.Insert(i, newColumn);
}
}
return(retDataFrame);
}
/// <summary>
/// Returns an <see cref="IEnumerable{RecordBatch}"/> without copying data
/// </summary>
public IEnumerable <RecordBatch> ToArrowRecordBatches()
{
Apache.Arrow.Schema.Builder schemaBuilder = new Apache.Arrow.Schema.Builder();
int columnCount = Columns.Count;
for (int i = 0; i < columnCount; i++)
{
DataFrameColumn column = Columns[i];
Field field = column.GetArrowField();
schemaBuilder.Field(field);
}
Schema schema = schemaBuilder.Build();
List <Apache.Arrow.Array> arrays = new List <Apache.Arrow.Array>();
int recordBatchLength = Int32.MaxValue;
int numberOfRowsInThisRecordBatch = (int)Math.Min(recordBatchLength, RowCount);
long numberOfRowsProcessed = 0;
// Sometimes .NET for Spark passes in DataFrames with no rows. In those cases, we just return a RecordBatch with the right Schema and no rows
do
{
for (int i = 0; i < columnCount; i++)
{
DataFrameColumn column = Columns[i];
numberOfRowsInThisRecordBatch = (int)Math.Min(numberOfRowsInThisRecordBatch, column.GetMaxRecordBatchLength(numberOfRowsProcessed));
}
for (int i = 0; i < columnCount; i++)
{
DataFrameColumn column = Columns[i];
arrays.Add(column.ToArrowArray(numberOfRowsProcessed, numberOfRowsInThisRecordBatch));
}
numberOfRowsProcessed += numberOfRowsInThisRecordBatch;
yield return(new RecordBatch(schema, arrays, numberOfRowsInThisRecordBatch));
} while (numberOfRowsProcessed < RowCount);
}
private static Dictionary <long, ICollection <long> > GetOccurences(DataFrame retainedDataFrame, DataFrame supplementaryDataFrame,
string[] retainedJoinColumnNames, string[] supplemetaryJoinColumnNames, out HashSet <long> supplementaryJoinColumnsNullIndices)
{
supplementaryJoinColumnsNullIndices = new HashSet <long>();
// Get occurrences of values in columns used for join in the retained and supplementary dataframes
Dictionary <long, ICollection <long> > occurrences = null;
Dictionary <long, long> retainedIndicesReverseMapping = null;
for (int colNameIndex = 0; colNameIndex < retainedJoinColumnNames.Length; colNameIndex++)
{
DataFrameColumn shrinkedRetainedColumn = retainedDataFrame.Columns[retainedJoinColumnNames[colNameIndex]];
// Shrink retained column by row occurrences from previous step
if (occurrences != null)
{
// Only rows with occurences from previose step should go for futher processing
var shrinkedRetainedIndices = occurrences.Keys.ToArray();
// Create reverse mapping of index of the row in the shrinked column to the index of this row in the original dataframe (new index -> original index)
var newRetainedIndicesReverseMapping = new Dictionary <long, long>(shrinkedRetainedIndices.Length);
for (int i = 0; i < shrinkedRetainedIndices.Length; i++)
{
// Store reverse mapping to restore original dataframe indices from indices in shrinked row
var originalIndex = shrinkedRetainedIndices[i];
newRetainedIndicesReverseMapping.Add(i, originalIndex);
}
retainedIndicesReverseMapping = newRetainedIndicesReverseMapping;
var indices = new Int64DataFrameColumn("Indices", shrinkedRetainedIndices);
shrinkedRetainedColumn = shrinkedRetainedColumn.Clone(indices);
}
DataFrameColumn supplementaryColumn = supplementaryDataFrame.Columns[supplemetaryJoinColumnNames[colNameIndex]];
// Find occurrenses on current step (join column)
var newOccurrences = shrinkedRetainedColumn.GetGroupedOccurrences(supplementaryColumn, out HashSet <long> supplementaryColumnNullIndices);
// Convert indices from in key from local (shrinked row) to indices in original dataframe
if (retainedIndicesReverseMapping != null)
{
newOccurrences = newOccurrences.ToDictionary(kvp => retainedIndicesReverseMapping[kvp.Key], kvp => kvp.Value);
}
supplementaryJoinColumnsNullIndices.UnionWith(supplementaryColumnNullIndices);
// Shrink join result on current column by previous join columns (if any)
// (we have to remove occurrences that doesn't exist in previous columns, because JOIN happens only if ALL left and right columns in JOIN are matched)
if (occurrences != null)
{
newOccurrences = GetShrinkedOccurences(occurrences, newOccurrences);
}
occurrences = newOccurrences;
}
return(occurrences);
}
public override PrimitiveDataFrameColumn <bool> ElementwiseEquals(DataFrameColumn column)
{
return(ElementwiseEqualsImplementation(this, column));
}
public virtual PrimitiveDataFrameColumn <bool> ElementwiseLessThan(DataFrameColumn column)
{
throw new NotImplementedException();
}
private static void AppendDataFrameColumnFromArrowArray(Field field, IArrowArray arrowArray, DataFrame ret, string fieldNamePrefix = "")
{
IArrowType fieldType = field.DataType;
DataFrameColumn dataFrameColumn = null;
string fieldName = fieldNamePrefix + field.Name;
switch (fieldType.TypeId)
{
case ArrowTypeId.Boolean:
BooleanArray arrowBooleanArray = (BooleanArray)arrowArray;
ReadOnlyMemory <byte> valueBuffer = arrowBooleanArray.ValueBuffer.Memory;
ReadOnlyMemory <byte> nullBitMapBuffer = arrowBooleanArray.NullBitmapBuffer.Memory;
dataFrameColumn = new BooleanDataFrameColumn(fieldName, valueBuffer, nullBitMapBuffer, arrowArray.Length, arrowArray.NullCount);
break;
case ArrowTypeId.Double:
PrimitiveArray <double> arrowDoubleArray = (PrimitiveArray <double>)arrowArray;
ReadOnlyMemory <byte> doubleValueBuffer = arrowDoubleArray.ValueBuffer.Memory;
ReadOnlyMemory <byte> doubleNullBitMapBuffer = arrowDoubleArray.NullBitmapBuffer.Memory;
dataFrameColumn = new DoubleDataFrameColumn(fieldName, doubleValueBuffer, doubleNullBitMapBuffer, arrowArray.Length, arrowArray.NullCount);
break;
case ArrowTypeId.Float:
PrimitiveArray <float> arrowFloatArray = (PrimitiveArray <float>)arrowArray;
ReadOnlyMemory <byte> floatValueBuffer = arrowFloatArray.ValueBuffer.Memory;
ReadOnlyMemory <byte> floatNullBitMapBuffer = arrowFloatArray.NullBitmapBuffer.Memory;
dataFrameColumn = new SingleDataFrameColumn(fieldName, floatValueBuffer, floatNullBitMapBuffer, arrowArray.Length, arrowArray.NullCount);
break;
case ArrowTypeId.Int8:
PrimitiveArray <sbyte> arrowsbyteArray = (PrimitiveArray <sbyte>)arrowArray;
ReadOnlyMemory <byte> sbyteValueBuffer = arrowsbyteArray.ValueBuffer.Memory;
ReadOnlyMemory <byte> sbyteNullBitMapBuffer = arrowsbyteArray.NullBitmapBuffer.Memory;
dataFrameColumn = new SByteDataFrameColumn(fieldName, sbyteValueBuffer, sbyteNullBitMapBuffer, arrowArray.Length, arrowArray.NullCount);
break;
case ArrowTypeId.Int16:
PrimitiveArray <short> arrowshortArray = (PrimitiveArray <short>)arrowArray;
ReadOnlyMemory <byte> shortValueBuffer = arrowshortArray.ValueBuffer.Memory;
ReadOnlyMemory <byte> shortNullBitMapBuffer = arrowshortArray.NullBitmapBuffer.Memory;
dataFrameColumn = new Int16DataFrameColumn(fieldName, shortValueBuffer, shortNullBitMapBuffer, arrowArray.Length, arrowArray.NullCount);
break;
case ArrowTypeId.Int32:
PrimitiveArray <int> arrowIntArray = (PrimitiveArray <int>)arrowArray;
ReadOnlyMemory <byte> intValueBuffer = arrowIntArray.ValueBuffer.Memory;
ReadOnlyMemory <byte> intNullBitMapBuffer = arrowIntArray.NullBitmapBuffer.Memory;
dataFrameColumn = new Int32DataFrameColumn(fieldName, intValueBuffer, intNullBitMapBuffer, arrowArray.Length, arrowArray.NullCount);
break;
case ArrowTypeId.Int64:
PrimitiveArray <long> arrowLongArray = (PrimitiveArray <long>)arrowArray;
ReadOnlyMemory <byte> longValueBuffer = arrowLongArray.ValueBuffer.Memory;
ReadOnlyMemory <byte> longNullBitMapBuffer = arrowLongArray.NullBitmapBuffer.Memory;
dataFrameColumn = new Int64DataFrameColumn(fieldName, longValueBuffer, longNullBitMapBuffer, arrowArray.Length, arrowArray.NullCount);
break;
case ArrowTypeId.String:
StringArray stringArray = (StringArray)arrowArray;
ReadOnlyMemory <byte> dataMemory = stringArray.ValueBuffer.Memory;
ReadOnlyMemory <byte> offsetsMemory = stringArray.ValueOffsetsBuffer.Memory;
ReadOnlyMemory <byte> nullMemory = stringArray.NullBitmapBuffer.Memory;
dataFrameColumn = new ArrowStringDataFrameColumn(fieldName, dataMemory, offsetsMemory, nullMemory, stringArray.Length, stringArray.NullCount);
break;
case ArrowTypeId.UInt8:
PrimitiveArray <byte> arrowbyteArray = (PrimitiveArray <byte>)arrowArray;
ReadOnlyMemory <byte> byteValueBuffer = arrowbyteArray.ValueBuffer.Memory;
ReadOnlyMemory <byte> byteNullBitMapBuffer = arrowbyteArray.NullBitmapBuffer.Memory;
dataFrameColumn = new ByteDataFrameColumn(fieldName, byteValueBuffer, byteNullBitMapBuffer, arrowArray.Length, arrowArray.NullCount);
break;
case ArrowTypeId.UInt16:
PrimitiveArray <ushort> arrowUshortArray = (PrimitiveArray <ushort>)arrowArray;
ReadOnlyMemory <byte> ushortValueBuffer = arrowUshortArray.ValueBuffer.Memory;
ReadOnlyMemory <byte> ushortNullBitMapBuffer = arrowUshortArray.NullBitmapBuffer.Memory;
dataFrameColumn = new UInt16DataFrameColumn(fieldName, ushortValueBuffer, ushortNullBitMapBuffer, arrowArray.Length, arrowArray.NullCount);
break;
case ArrowTypeId.UInt32:
PrimitiveArray <uint> arrowUintArray = (PrimitiveArray <uint>)arrowArray;
ReadOnlyMemory <byte> uintValueBuffer = arrowUintArray.ValueBuffer.Memory;
ReadOnlyMemory <byte> uintNullBitMapBuffer = arrowUintArray.NullBitmapBuffer.Memory;
dataFrameColumn = new UInt32DataFrameColumn(fieldName, uintValueBuffer, uintNullBitMapBuffer, arrowArray.Length, arrowArray.NullCount);
break;
case ArrowTypeId.UInt64:
PrimitiveArray <ulong> arrowUlongArray = (PrimitiveArray <ulong>)arrowArray;
ReadOnlyMemory <byte> ulongValueBuffer = arrowUlongArray.ValueBuffer.Memory;
ReadOnlyMemory <byte> ulongNullBitMapBuffer = arrowUlongArray.NullBitmapBuffer.Memory;
dataFrameColumn = new UInt64DataFrameColumn(fieldName, ulongValueBuffer, ulongNullBitMapBuffer, arrowArray.Length, arrowArray.NullCount);
break;
case ArrowTypeId.Struct:
StructArray structArray = (StructArray)arrowArray;
StructType structType = (StructType)field.DataType;
IEnumerator <Field> fieldsEnumerator = structType.Fields.GetEnumerator();
IEnumerator <IArrowArray> structArrayEnumerator = structArray.Fields.GetEnumerator();
while (fieldsEnumerator.MoveNext() && structArrayEnumerator.MoveNext())
{
AppendDataFrameColumnFromArrowArray(fieldsEnumerator.Current, structArrayEnumerator.Current, ret, field.Name + "_");
}
break;
case ArrowTypeId.Decimal:
case ArrowTypeId.Binary:
case ArrowTypeId.Date32:
case ArrowTypeId.Date64:
case ArrowTypeId.Dictionary:
case ArrowTypeId.FixedSizedBinary:
case ArrowTypeId.HalfFloat:
case ArrowTypeId.Interval:
case ArrowTypeId.List:
case ArrowTypeId.Map:
case ArrowTypeId.Null:
case ArrowTypeId.Time32:
case ArrowTypeId.Time64:
default:
throw new NotImplementedException(nameof(fieldType.Name));
}
if (dataFrameColumn != null)
{
ret.Columns.Insert(ret.Columns.Count, dataFrameColumn);
}
}
internal static PrimitiveDataFrameColumn <bool> ElementwiseEqualsImplementation(DataFrameColumn left, DataFrameColumn right)
{
if (left.Length != right.Length)
{
throw new ArgumentException(Strings.MismatchedColumnLengths, nameof(right));
}
PrimitiveDataFrameColumn <bool> ret = new PrimitiveDataFrameColumn <bool>(left.Name, left.Length);
for (long i = 0; i < left.Length; i++)
{
ret[i] = (string)left[i] == right[i]?.ToString();
}
return(ret);
}
public DataFrame Merge(DataFrame other, string[] leftJoinColumns, string[] rightJoinColumns, string leftSuffix = "_left", string rightSuffix = "_right", JoinAlgorithm joinAlgorithm = JoinAlgorithm.Left)
{
if (other == null)
{
throw new ArgumentNullException(nameof(other));
}
//In Outer join the joined dataframe retains each row — even if no other matching row exists in supplementary dataframe.
//Outer joins subdivide further into left outer joins (left dataframe is retained), right outer joins (rightdataframe is retained), in full outer both are retained
PrimitiveDataFrameColumn <long> retainedRowIndices;
PrimitiveDataFrameColumn <long> supplementaryRowIndices;
DataFrame supplementaryDataFrame;
DataFrame retainedDataFrame;
bool isLeftDataFrameRetained;
if (joinAlgorithm == JoinAlgorithm.Left || joinAlgorithm == JoinAlgorithm.Right)
{
isLeftDataFrameRetained = (joinAlgorithm == JoinAlgorithm.Left);
supplementaryDataFrame = isLeftDataFrameRetained ? other : this;
var supplementaryJoinColumns = isLeftDataFrameRetained ? rightJoinColumns : leftJoinColumns;
retainedDataFrame = isLeftDataFrameRetained ? this : other;
var retainedJoinColumns = isLeftDataFrameRetained ? leftJoinColumns : rightJoinColumns;
Merge(retainedDataFrame, supplementaryDataFrame, retainedJoinColumns, supplementaryJoinColumns, out retainedRowIndices, out supplementaryRowIndices);
}
else if (joinAlgorithm == JoinAlgorithm.Inner)
{
// use as supplementary (for Hashing) the dataframe with the smaller RowCount
isLeftDataFrameRetained = (Rows.Count > other.Rows.Count);
supplementaryDataFrame = isLeftDataFrameRetained ? other : this;
var supplementaryJoinColumns = isLeftDataFrameRetained ? rightJoinColumns : leftJoinColumns;
retainedDataFrame = isLeftDataFrameRetained ? this : other;
var retainedJoinColumns = isLeftDataFrameRetained ? leftJoinColumns : rightJoinColumns;
Merge(retainedDataFrame, supplementaryDataFrame, retainedJoinColumns, supplementaryJoinColumns, out retainedRowIndices, out supplementaryRowIndices, true);
}
else if (joinAlgorithm == JoinAlgorithm.FullOuter)
{
//In full outer join we would like to retain data from both side, so we do it into 2 steps: one first we do LEFT JOIN and then add lost data from the RIGHT side
//Step 1
//Do LEFT JOIN
isLeftDataFrameRetained = true;
supplementaryDataFrame = isLeftDataFrameRetained ? other : this;
var supplementaryJoinColumns = isLeftDataFrameRetained ? rightJoinColumns : leftJoinColumns;
retainedDataFrame = isLeftDataFrameRetained ? this : other;
var retainedJoinColumns = isLeftDataFrameRetained ? leftJoinColumns : rightJoinColumns;
var intersection = Merge(retainedDataFrame, supplementaryDataFrame, retainedJoinColumns, supplementaryJoinColumns, out retainedRowIndices, out supplementaryRowIndices, calculateIntersection: true);
//Step 2
//Do RIGHT JOIN to retain all data from supplementary DataFrame too (take into account data intersection from the first step to avoid duplicates)
for (long i = 0; i < supplementaryDataFrame.Columns.RowCount; i++)
{
var columns = supplementaryJoinColumns.Select(name => supplementaryDataFrame.Columns[name]).ToArray();
if (!IsAnyNullValueInColumns(columns, i))
{
if (!intersection.Contains(i))
{
retainedRowIndices.Append(null);
supplementaryRowIndices.Append(i);
}
}
}
}
else
{
throw new NotImplementedException(nameof(joinAlgorithm));
}
DataFrame ret = new DataFrame();
//insert columns from left dataframe (this)
for (int i = 0; i < this.Columns.Count; i++)
{
ret.Columns.Insert(i, this.Columns[i].Clone(isLeftDataFrameRetained ? retainedRowIndices : supplementaryRowIndices));
}
//insert columns from right dataframe (other)
for (int i = 0; i < other.Columns.Count; i++)
{
DataFrameColumn column = other.Columns[i].Clone(isLeftDataFrameRetained ? supplementaryRowIndices : retainedRowIndices);
SetSuffixForDuplicatedColumnNames(ret, column, leftSuffix, rightSuffix);
ret.Columns.Insert(ret.Columns.Count, column);
}
return(ret);
}
private static HashSet <long> Merge(DataFrame retainedDataFrame, DataFrame supplementaryDataFrame, string[] retainedJoinColumnNames, string[] supplemetaryJoinColumnNames, out PrimitiveDataFrameColumn <long> retainedRowIndices, out PrimitiveDataFrameColumn <long> supplementaryRowIndices, bool isInner = false, bool calculateIntersection = false)
{
if (retainedJoinColumnNames == null)
{
throw new ArgumentNullException(nameof(retainedJoinColumnNames));
}
if (supplemetaryJoinColumnNames == null)
{
throw new ArgumentNullException(nameof(supplemetaryJoinColumnNames));
}
if (retainedJoinColumnNames.Length != supplemetaryJoinColumnNames.Length)
{
throw new ArgumentException(Strings.MismatchedArrayLengths, nameof(retainedJoinColumnNames));
}
HashSet <long> intersection = calculateIntersection ? new HashSet <long>() : null;
// Get occurrences of values in columns used for join in the retained and supplementary dataframes
Dictionary <long, ICollection <long> > occurrences = null;
Dictionary <long, long> retainedIndicesReverseMapping = null;
HashSet <long> supplementaryJoinColumnsNullIndices = new HashSet <long>();
for (int colNameIndex = 0; colNameIndex < retainedJoinColumnNames.Length; colNameIndex++)
{
DataFrameColumn shrinkedRetainedColumn = retainedDataFrame.Columns[retainedJoinColumnNames[colNameIndex]];
//shrink retained column by row occurrences from previous step
if (occurrences != null)
{
//only rows with occurences from previose step should go for futher processing
var shrinkedRetainedIndices = occurrences.Keys.ToArray();
//create reverse mapping of index of the row in the shrinked column to the index of this row in the original dataframe (new index -> original index)
var newRetainedIndicesReverseMapping = new Dictionary <long, long>(shrinkedRetainedIndices.Length);
for (int i = 0; i < shrinkedRetainedIndices.Length; i++)
{
//store reverse mapping to restore original dataframe indices from indices in shrinked row
var originalIndex = shrinkedRetainedIndices[i];
newRetainedIndicesReverseMapping.Add(i, originalIndex);
}
retainedIndicesReverseMapping = newRetainedIndicesReverseMapping;
shrinkedRetainedColumn = shrinkedRetainedColumn.Clone(new Int64DataFrameColumn("Indices", shrinkedRetainedIndices));
}
DataFrameColumn supplementaryColumn = supplementaryDataFrame.Columns[supplemetaryJoinColumnNames[colNameIndex]];
//Find occurrenses on current step (join column)
var newOccurrences = shrinkedRetainedColumn.GetGroupedOccurrences(supplementaryColumn, out HashSet <long> supplementaryColumnNullIndices);
//Convert indices from in key from local (shrinked row) to indices in original dataframe
if (retainedIndicesReverseMapping != null)
{
newOccurrences = newOccurrences.ToDictionary(kvp => retainedIndicesReverseMapping[kvp.Key], kvp => kvp.Value);
}
supplementaryJoinColumnsNullIndices.UnionWith(supplementaryColumnNullIndices);
// shrink join result on current column by previous join columns (if any)
// (we have to remove occurrences that doesn't exist in previous columns, because JOIN happens only if ALL left and right columns in JOIN are matched)
if (occurrences != null)
{
var shrinkedOccurences = new Dictionary <long, ICollection <long> >();
foreach (var kvp in newOccurrences)
{
var newValue = kvp.Value.Where(i => occurrences[kvp.Key].Contains(i)).ToArray();
if (newValue.Any())
{
shrinkedOccurences.Add(kvp.Key, newValue);
}
}
newOccurrences = shrinkedOccurences;
}
occurrences = newOccurrences;
}
retainedRowIndices = new Int64DataFrameColumn("RetainedIndices");
supplementaryRowIndices = new Int64DataFrameColumn("SupplementaryIndices");
//Perform Merging
var retainJoinColumns = retainedJoinColumnNames.Select(name => retainedDataFrame.Columns[name]).ToArray();
for (long i = 0; i < retainedDataFrame.Columns.RowCount; i++)
{
if (!IsAnyNullValueInColumns(retainJoinColumns, i))
{
//Get all row indexes from supplementary dataframe that sutisfy JOIN condition
if (occurrences.TryGetValue(i, out ICollection <long> rowIndices))
{
foreach (long supplementaryRowIndex in rowIndices)
{
retainedRowIndices.Append(i);
supplementaryRowIndices.Append(supplementaryRowIndex);
//store intersection if required
if (calculateIntersection)
{
if (!intersection.Contains(supplementaryRowIndex))
{
intersection.Add(supplementaryRowIndex);
}
}
}
}
else
{
if (isInner)
{
continue;
}
retainedRowIndices.Append(i);
supplementaryRowIndices.Append(null);
}
}
else
{
foreach (long row in supplementaryJoinColumnsNullIndices)
{
retainedRowIndices.Append(i);
supplementaryRowIndices.Append(row);
}
}
}
return(intersection);
}
private Delegate CreateGetterDelegate(int col)
{
DataFrameColumn column = _dataFrame.Columns[col];
return(column.GetDataViewGetter(this));
}
/// <summary>
/// Wraps a <see cref="DataFrame"/> around an Arrow <see cref="RecordBatch"/> without copying data
/// </summary>
/// <param name="recordBatch"></param>
/// <returns><see cref="DataFrame"/></returns>
public static DataFrame FromArrowRecordBatch(RecordBatch recordBatch)
{
DataFrame ret = new DataFrame();
Apache.Arrow.Schema arrowSchema = recordBatch.Schema;
int fieldIndex = 0;
IEnumerable <IArrowArray> arrowArrays = recordBatch.Arrays;
foreach (IArrowArray arrowArray in arrowArrays)
{
Field field = arrowSchema.GetFieldByIndex(fieldIndex);
IArrowType fieldType = field.DataType;
DataFrameColumn dataFrameColumn = null;
switch (fieldType.TypeId)
{
case ArrowTypeId.Boolean:
BooleanArray arrowBooleanArray = (BooleanArray)arrowArray;
ReadOnlyMemory <byte> valueBuffer = arrowBooleanArray.ValueBuffer.Memory;
ReadOnlyMemory <byte> nullBitMapBuffer = arrowBooleanArray.NullBitmapBuffer.Memory;
dataFrameColumn = new PrimitiveDataFrameColumn <bool>(field.Name, valueBuffer, nullBitMapBuffer, arrowArray.Length, arrowArray.NullCount);
break;
case ArrowTypeId.Double:
PrimitiveArray <double> arrowDoubleArray = (PrimitiveArray <double>)arrowArray;
ReadOnlyMemory <byte> doubleValueBuffer = arrowDoubleArray.ValueBuffer.Memory;
ReadOnlyMemory <byte> doubleNullBitMapBuffer = arrowDoubleArray.NullBitmapBuffer.Memory;
dataFrameColumn = new PrimitiveDataFrameColumn <double>(field.Name, doubleValueBuffer, doubleNullBitMapBuffer, arrowArray.Length, arrowArray.NullCount);
break;
case ArrowTypeId.Float:
PrimitiveArray <float> arrowFloatArray = (PrimitiveArray <float>)arrowArray;
ReadOnlyMemory <byte> floatValueBuffer = arrowFloatArray.ValueBuffer.Memory;
ReadOnlyMemory <byte> floatNullBitMapBuffer = arrowFloatArray.NullBitmapBuffer.Memory;
dataFrameColumn = new PrimitiveDataFrameColumn <float>(field.Name, floatValueBuffer, floatNullBitMapBuffer, arrowArray.Length, arrowArray.NullCount);
break;
case ArrowTypeId.Int8:
PrimitiveArray <sbyte> arrowsbyteArray = (PrimitiveArray <sbyte>)arrowArray;
ReadOnlyMemory <byte> sbyteValueBuffer = arrowsbyteArray.ValueBuffer.Memory;
ReadOnlyMemory <byte> sbyteNullBitMapBuffer = arrowsbyteArray.NullBitmapBuffer.Memory;
dataFrameColumn = new PrimitiveDataFrameColumn <sbyte>(field.Name, sbyteValueBuffer, sbyteNullBitMapBuffer, arrowArray.Length, arrowArray.NullCount);
break;
case ArrowTypeId.Int16:
PrimitiveArray <short> arrowshortArray = (PrimitiveArray <short>)arrowArray;
ReadOnlyMemory <byte> shortValueBuffer = arrowshortArray.ValueBuffer.Memory;
ReadOnlyMemory <byte> shortNullBitMapBuffer = arrowshortArray.NullBitmapBuffer.Memory;
dataFrameColumn = new PrimitiveDataFrameColumn <short>(field.Name, shortValueBuffer, shortNullBitMapBuffer, arrowArray.Length, arrowArray.NullCount);
break;
case ArrowTypeId.Int32:
PrimitiveArray <int> arrowIntArray = (PrimitiveArray <int>)arrowArray;
ReadOnlyMemory <byte> intValueBuffer = arrowIntArray.ValueBuffer.Memory;
ReadOnlyMemory <byte> intNullBitMapBuffer = arrowIntArray.NullBitmapBuffer.Memory;
dataFrameColumn = new PrimitiveDataFrameColumn <int>(field.Name, intValueBuffer, intNullBitMapBuffer, arrowArray.Length, arrowArray.NullCount);
break;
case ArrowTypeId.Int64:
PrimitiveArray <long> arrowLongArray = (PrimitiveArray <long>)arrowArray;
ReadOnlyMemory <byte> longValueBuffer = arrowLongArray.ValueBuffer.Memory;
ReadOnlyMemory <byte> longNullBitMapBuffer = arrowLongArray.NullBitmapBuffer.Memory;
dataFrameColumn = new PrimitiveDataFrameColumn <long>(field.Name, longValueBuffer, longNullBitMapBuffer, arrowArray.Length, arrowArray.NullCount);
break;
case ArrowTypeId.String:
StringArray stringArray = (StringArray)arrowArray;
ReadOnlyMemory <byte> dataMemory = stringArray.ValueBuffer.Memory;
ReadOnlyMemory <byte> offsetsMemory = stringArray.ValueOffsetsBuffer.Memory;
ReadOnlyMemory <byte> nullMemory = stringArray.NullBitmapBuffer.Memory;
dataFrameColumn = new ArrowStringDataFrameColumn(field.Name, dataMemory, offsetsMemory, nullMemory, stringArray.Length, stringArray.NullCount);
break;
case ArrowTypeId.UInt8:
PrimitiveArray <byte> arrowbyteArray = (PrimitiveArray <byte>)arrowArray;
ReadOnlyMemory <byte> byteValueBuffer = arrowbyteArray.ValueBuffer.Memory;
ReadOnlyMemory <byte> byteNullBitMapBuffer = arrowbyteArray.NullBitmapBuffer.Memory;
dataFrameColumn = new PrimitiveDataFrameColumn <byte>(field.Name, byteValueBuffer, byteNullBitMapBuffer, arrowArray.Length, arrowArray.NullCount);
break;
case ArrowTypeId.UInt16:
PrimitiveArray <ushort> arrowUshortArray = (PrimitiveArray <ushort>)arrowArray;
ReadOnlyMemory <byte> ushortValueBuffer = arrowUshortArray.ValueBuffer.Memory;
ReadOnlyMemory <byte> ushortNullBitMapBuffer = arrowUshortArray.NullBitmapBuffer.Memory;
dataFrameColumn = new PrimitiveDataFrameColumn <ushort>(field.Name, ushortValueBuffer, ushortNullBitMapBuffer, arrowArray.Length, arrowArray.NullCount);
break;
case ArrowTypeId.UInt32:
PrimitiveArray <uint> arrowUintArray = (PrimitiveArray <uint>)arrowArray;
ReadOnlyMemory <byte> uintValueBuffer = arrowUintArray.ValueBuffer.Memory;
ReadOnlyMemory <byte> uintNullBitMapBuffer = arrowUintArray.NullBitmapBuffer.Memory;
dataFrameColumn = new PrimitiveDataFrameColumn <uint>(field.Name, uintValueBuffer, uintNullBitMapBuffer, arrowArray.Length, arrowArray.NullCount);
break;
case ArrowTypeId.UInt64:
PrimitiveArray <ulong> arrowUlongArray = (PrimitiveArray <ulong>)arrowArray;
ReadOnlyMemory <byte> ulongValueBuffer = arrowUlongArray.ValueBuffer.Memory;
ReadOnlyMemory <byte> ulongNullBitMapBuffer = arrowUlongArray.NullBitmapBuffer.Memory;
dataFrameColumn = new PrimitiveDataFrameColumn <ulong>(field.Name, ulongValueBuffer, ulongNullBitMapBuffer, arrowArray.Length, arrowArray.NullCount);
break;
case ArrowTypeId.Decimal:
case ArrowTypeId.Binary:
case ArrowTypeId.Date32:
case ArrowTypeId.Date64:
case ArrowTypeId.Dictionary:
case ArrowTypeId.FixedSizedBinary:
case ArrowTypeId.HalfFloat:
case ArrowTypeId.Interval:
case ArrowTypeId.List:
case ArrowTypeId.Map:
case ArrowTypeId.Null:
case ArrowTypeId.Struct:
case ArrowTypeId.Time32:
case ArrowTypeId.Time64:
default:
throw new NotImplementedException(nameof(fieldType.Name));
}
ret.Columns.Insert(ret.Columns.Count, dataFrameColumn);
fieldIndex++;
}
return(ret);
}
// TODO: Merge API with an "On" parameter that merges on a column common to 2 dataframes
/// <summary>
/// Merge DataFrames with a database style join
/// </summary>
/// <param name="other"></param>
/// <param name="leftJoinColumn"></param>
/// <param name="rightJoinColumn"></param>
/// <param name="leftSuffix"></param>
/// <param name="rightSuffix"></param>
/// <param name="joinAlgorithm"></param>
/// <returns></returns>
public DataFrame Merge <TKey>(DataFrame other, string leftJoinColumn, string rightJoinColumn, string leftSuffix = "_left", string rightSuffix = "_right", JoinAlgorithm joinAlgorithm = JoinAlgorithm.Left)
{
// A simple hash join
DataFrame ret = new DataFrame();
DataFrame leftDataFrame = this;
DataFrame rightDataFrame = other;
// The final table size is not known until runtime
long rowNumber = 0;
PrimitiveDataFrameColumn <long> leftRowIndices = new PrimitiveDataFrameColumn <long>("LeftIndices");
PrimitiveDataFrameColumn <long> rightRowIndices = new PrimitiveDataFrameColumn <long>("RightIndices");
if (joinAlgorithm == JoinAlgorithm.Left)
{
// First hash other dataframe on the rightJoinColumn
DataFrameColumn otherColumn = other[rightJoinColumn];
Dictionary <TKey, ICollection <long> > multimap = otherColumn.GroupColumnValues <TKey>();
// Go over the records in this dataframe and match with the dictionary
DataFrameColumn thisColumn = this[leftJoinColumn];
for (long i = 0; i < thisColumn.Length; i++)
{
var thisColumnValue = thisColumn[i];
TKey thisColumnValueOrDefault = (TKey)(thisColumnValue == null ? default(TKey) : thisColumnValue);
if (multimap.TryGetValue(thisColumnValueOrDefault, out ICollection <long> rowNumbers))
{
foreach (long row in rowNumbers)
{
if (thisColumnValue == null)
{
// Match only with nulls in otherColumn
if (otherColumn[row] == null)
{
leftRowIndices.Append(i);
rightRowIndices.Append(row);
}
}
else
{
// Cannot match nulls in otherColumn
if (otherColumn[row] != null)
{
leftRowIndices.Append(i);
rightRowIndices.Append(row);
}
}
}
}
else
{
leftRowIndices.Append(i);
rightRowIndices.Append(null);
}
}
}
else if (joinAlgorithm == JoinAlgorithm.Right)
{
DataFrameColumn thisColumn = this[leftJoinColumn];
Dictionary <TKey, ICollection <long> > multimap = thisColumn.GroupColumnValues <TKey>();
DataFrameColumn otherColumn = other[rightJoinColumn];
for (long i = 0; i < otherColumn.Length; i++)
{
var otherColumnValue = otherColumn[i];
TKey otherColumnValueOrDefault = (TKey)(otherColumnValue == null ? default(TKey) : otherColumnValue);
if (multimap.TryGetValue(otherColumnValueOrDefault, out ICollection <long> rowNumbers))
{
foreach (long row in rowNumbers)
{
if (otherColumnValue == null)
{
if (thisColumn[row] == null)
{
leftRowIndices.Append(row);
rightRowIndices.Append(i);
}
}
else
{
if (thisColumn[row] != null)
{
leftRowIndices.Append(row);
rightRowIndices.Append(i);
}
}
}
}
else
{
leftRowIndices.Append(null);
rightRowIndices.Append(i);
}
}
}
else if (joinAlgorithm == JoinAlgorithm.Inner)
{
// Hash the column with the smaller RowCount
long leftRowCount = Rows.Count;
long rightRowCount = other.Rows.Count;
DataFrame longerDataFrame = leftRowCount <= rightRowCount ? other : this;
DataFrame shorterDataFrame = ReferenceEquals(longerDataFrame, this) ? other : this;
DataFrameColumn hashColumn = (leftRowCount <= rightRowCount) ? this[leftJoinColumn] : other[rightJoinColumn];
DataFrameColumn otherColumn = ReferenceEquals(hashColumn, this[leftJoinColumn]) ? other[rightJoinColumn] : this[leftJoinColumn];
Dictionary <TKey, ICollection <long> > multimap = hashColumn.GroupColumnValues <TKey>();
for (long i = 0; i < otherColumn.Length; i++)
{
var otherColumnValue = otherColumn[i];
TKey otherColumnValueOrDefault = (TKey)(otherColumnValue == null ? default(TKey) : otherColumnValue);
if (multimap.TryGetValue(otherColumnValueOrDefault, out ICollection <long> rowNumbers))
{
foreach (long row in rowNumbers)
{
if (otherColumnValue == null)
{
if (hashColumn[row] == null)
{
leftRowIndices.Append(row);
rightRowIndices.Append(i);
}
}
else
{
if (hashColumn[row] != null)
{
leftRowIndices.Append(row);
rightRowIndices.Append(i);
}
}
}
}
}
leftDataFrame = shorterDataFrame;
rightDataFrame = longerDataFrame;
}
else if (joinAlgorithm == JoinAlgorithm.FullOuter)
{
DataFrameColumn otherColumn = other[rightJoinColumn];
Dictionary <TKey, ICollection <long> > multimap = otherColumn.GroupColumnValues <TKey>();
Dictionary <TKey, long> intersection = new Dictionary <TKey, long>(EqualityComparer <TKey> .Default);
// Go over the records in this dataframe and match with the dictionary
DataFrameColumn thisColumn = this[leftJoinColumn];
for (long i = 0; i < thisColumn.Length; i++)
{
var thisColumnValue = thisColumn[i];
TKey thisColumnValueOrDefault = (TKey)(thisColumnValue == null ? default(TKey) : thisColumnValue);
if (multimap.TryGetValue(thisColumnValueOrDefault, out ICollection <long> rowNumbers))
{
foreach (long row in rowNumbers)
{
if (thisColumnValue == null)
{
// Has to match only with nulls in otherColumn
if (otherColumn[row] == null)
{
leftRowIndices.Append(i);
rightRowIndices.Append(row);
if (!intersection.ContainsKey(thisColumnValueOrDefault))
{
intersection.Add(thisColumnValueOrDefault, rowNumber);
}
}
}
else
{
// Cannot match to nulls in otherColumn
if (otherColumn[row] != null)
{
leftRowIndices.Append(i);
rightRowIndices.Append(row);
if (!intersection.ContainsKey(thisColumnValueOrDefault))
{
intersection.Add(thisColumnValueOrDefault, rowNumber);
}
}
}
}
}
else
{
leftRowIndices.Append(i);
rightRowIndices.Append(null);
}
}
for (long i = 0; i < otherColumn.Length; i++)
{
TKey value = (TKey)(otherColumn[i] ?? default(TKey));
if (!intersection.ContainsKey(value))
{
leftRowIndices.Append(null);
rightRowIndices.Append(i);
}
}
}
else
{
throw new NotImplementedException(nameof(joinAlgorithm));
}
for (int i = 0; i < leftDataFrame.Columns.Count; i++)
{
ret.Columns.Insert(i, leftDataFrame.Columns[i].Clone(leftRowIndices));
}
for (int i = 0; i < rightDataFrame.Columns.Count; i++)
{
DataFrameColumn column = rightDataFrame.Columns[i].Clone(rightRowIndices);
SetSuffixForDuplicatedColumnNames(ret, column, leftSuffix, rightSuffix);
ret.Columns.Insert(ret.Columns.Count, column);
}
return(ret);
}
public virtual PrimitiveDataFrameColumn <bool> ElementwiseGreaterThanOrEqual(DataFrameColumn column)
{
throw new NotImplementedException();
}
public DataFrame Join(DataFrame other, string leftSuffix = "_left", string rightSuffix = "_right", JoinAlgorithm joinAlgorithm = JoinAlgorithm.Left)
{
DataFrame ret = new DataFrame();
if (joinAlgorithm == JoinAlgorithm.Left)
{
for (int i = 0; i < Columns.Count; i++)
{
DataFrameColumn newColumn = Columns[i].Clone();
ret.Columns.Insert(ret.Columns.Count, newColumn);
}
long minLength = Math.Min(Rows.Count, other.Rows.Count);
PrimitiveDataFrameColumn <long> mapIndices = new PrimitiveDataFrameColumn <long>("mapIndices", minLength);
for (long i = 0; i < minLength; i++)
{
mapIndices[i] = i;
}
for (int i = 0; i < other.Columns.Count; i++)
{
DataFrameColumn newColumn;
if (other.Rows.Count < Rows.Count)
{
newColumn = other.Columns[i].Clone(numberOfNullsToAppend: Rows.Count - other.Rows.Count);
}
else
{
newColumn = other.Columns[i].Clone(mapIndices);
}
SetSuffixForDuplicatedColumnNames(ret, newColumn, leftSuffix, rightSuffix);
ret.Columns.Insert(ret.Columns.Count, newColumn);
}
}
else if (joinAlgorithm == JoinAlgorithm.Right)
{
long minLength = Math.Min(Rows.Count, other.Rows.Count);
PrimitiveDataFrameColumn <long> mapIndices = new PrimitiveDataFrameColumn <long>("mapIndices", minLength);
for (long i = 0; i < minLength; i++)
{
mapIndices[i] = i;
}
for (int i = 0; i < Columns.Count; i++)
{
DataFrameColumn newColumn;
if (Rows.Count < other.Rows.Count)
{
newColumn = Columns[i].Clone(numberOfNullsToAppend: other.Rows.Count - Rows.Count);
}
else
{
newColumn = Columns[i].Clone(mapIndices);
}
ret.Columns.Insert(ret.Columns.Count, newColumn);
}
for (int i = 0; i < other.Columns.Count; i++)
{
DataFrameColumn newColumn = other.Columns[i].Clone();
SetSuffixForDuplicatedColumnNames(ret, newColumn, leftSuffix, rightSuffix);
ret.Columns.Insert(ret.Columns.Count, newColumn);
}
}
else if (joinAlgorithm == JoinAlgorithm.FullOuter)
{
long newRowCount = Math.Max(Rows.Count, other.Rows.Count);
long numberOfNulls = newRowCount - Rows.Count;
for (int i = 0; i < Columns.Count; i++)
{
DataFrameColumn newColumn = Columns[i].Clone(numberOfNullsToAppend: numberOfNulls);
ret.Columns.Insert(ret.Columns.Count, newColumn);
}
numberOfNulls = newRowCount - other.Rows.Count;
for (int i = 0; i < other.Columns.Count; i++)
{
DataFrameColumn newColumn = other.Columns[i].Clone(numberOfNullsToAppend: numberOfNulls);
SetSuffixForDuplicatedColumnNames(ret, newColumn, leftSuffix, rightSuffix);
ret.Columns.Insert(ret.Columns.Count, newColumn);
}
}
else if (joinAlgorithm == JoinAlgorithm.Inner)
{
long newRowCount = Math.Min(Rows.Count, other.Rows.Count);
PrimitiveDataFrameColumn <long> mapIndices = new PrimitiveDataFrameColumn <long>("mapIndices", newRowCount);
for (long i = 0; i < newRowCount; i++)
{
mapIndices[i] = i;
}
for (int i = 0; i < Columns.Count; i++)
{
DataFrameColumn newColumn = Columns[i].Clone(mapIndices);
ret.Columns.Insert(ret.Columns.Count, newColumn);
}
for (int i = 0; i < other.Columns.Count; i++)
{
DataFrameColumn newColumn = other.Columns[i].Clone(mapIndices);
SetSuffixForDuplicatedColumnNames(ret, newColumn, leftSuffix, rightSuffix);
ret.Columns.Insert(ret.Columns.Count, newColumn);
}
}
return(ret);
}
public virtual DataFrameColumn Modulo(DataFrameColumn column, bool inPlace = false)
{
throw new NotImplementedException();
}
// TODO: Merge API with an "On" parameter that merges on a column common to 2 dataframes
/// <summary>
/// Merge DataFrames with a database style join
/// </summary>
/// <param name="other"></param>
/// <param name="leftJoinColumn"></param>
/// <param name="rightJoinColumn"></param>
/// <param name="leftSuffix"></param>
/// <param name="rightSuffix"></param>
/// <param name="joinAlgorithm"></param>
/// <returns></returns>
public DataFrame Merge <TKey>(DataFrame other, string leftJoinColumn, string rightJoinColumn, string leftSuffix = "_left", string rightSuffix = "_right", JoinAlgorithm joinAlgorithm = JoinAlgorithm.Left)
{
// A simple hash join
DataFrame ret = new DataFrame();
DataFrame leftDataFrame = this;
DataFrame rightDataFrame = other;
// The final table size is not known until runtime
long rowNumber = 0;
PrimitiveDataFrameColumn <long> leftRowIndices = new PrimitiveDataFrameColumn <long>("LeftIndices");
PrimitiveDataFrameColumn <long> rightRowIndices = new PrimitiveDataFrameColumn <long>("RightIndices");
if (joinAlgorithm == JoinAlgorithm.Left)
{
// First hash other dataframe on the rightJoinColumn
DataFrameColumn otherColumn = other.Columns[rightJoinColumn];
Dictionary <TKey, ICollection <long> > multimap = otherColumn.GroupColumnValues <TKey>(out HashSet <long> otherColumnNullIndices);
// Go over the records in this dataframe and match with the dictionary
DataFrameColumn thisColumn = Columns[leftJoinColumn];
for (long i = 0; i < thisColumn.Length; i++)
{
var thisColumnValue = thisColumn[i];
if (thisColumnValue != null)
{
if (multimap.TryGetValue((TKey)thisColumnValue, out ICollection <long> rowNumbers))
{
foreach (long row in rowNumbers)
{
leftRowIndices.Append(i);
rightRowIndices.Append(row);
}
}
else
{
leftRowIndices.Append(i);
rightRowIndices.Append(null);
}
}
else
{
foreach (long row in otherColumnNullIndices)
{
leftRowIndices.Append(i);
rightRowIndices.Append(row);
}
}
}
}
else if (joinAlgorithm == JoinAlgorithm.Right)
{
DataFrameColumn thisColumn = Columns[leftJoinColumn];
Dictionary <TKey, ICollection <long> > multimap = thisColumn.GroupColumnValues <TKey>(out HashSet <long> thisColumnNullIndices);
DataFrameColumn otherColumn = other.Columns[rightJoinColumn];
for (long i = 0; i < otherColumn.Length; i++)
{
var otherColumnValue = otherColumn[i];
if (otherColumnValue != null)
{
if (multimap.TryGetValue((TKey)otherColumnValue, out ICollection <long> rowNumbers))
{
foreach (long row in rowNumbers)
{
leftRowIndices.Append(row);
rightRowIndices.Append(i);
}
}
else
{
leftRowIndices.Append(null);
rightRowIndices.Append(i);
}
}
else
{
foreach (long thisColumnNullIndex in thisColumnNullIndices)
{
leftRowIndices.Append(thisColumnNullIndex);
rightRowIndices.Append(i);
}
}
}
}
else if (joinAlgorithm == JoinAlgorithm.Inner)
{
// Hash the column with the smaller RowCount
long leftRowCount = Rows.Count;
long rightRowCount = other.Rows.Count;
bool leftColumnIsSmaller = leftRowCount <= rightRowCount;
DataFrameColumn hashColumn = leftColumnIsSmaller ? Columns[leftJoinColumn] : other.Columns[rightJoinColumn];
DataFrameColumn otherColumn = ReferenceEquals(hashColumn, Columns[leftJoinColumn]) ? other.Columns[rightJoinColumn] : Columns[leftJoinColumn];
Dictionary <TKey, ICollection <long> > multimap = hashColumn.GroupColumnValues <TKey>(out HashSet <long> smallerDataFrameColumnNullIndices);
for (long i = 0; i < otherColumn.Length; i++)
{
var otherColumnValue = otherColumn[i];
if (otherColumnValue != null)
{
if (multimap.TryGetValue((TKey)otherColumnValue, out ICollection <long> rowNumbers))
{
foreach (long row in rowNumbers)
{
leftRowIndices.Append(leftColumnIsSmaller ? row : i);
rightRowIndices.Append(leftColumnIsSmaller ? i : row);
}
}
}
else
{
foreach (long nullIndex in smallerDataFrameColumnNullIndices)
{
leftRowIndices.Append(leftColumnIsSmaller ? nullIndex : i);
rightRowIndices.Append(leftColumnIsSmaller ? i : nullIndex);
}
}
}
}
else if (joinAlgorithm == JoinAlgorithm.FullOuter)
{
DataFrameColumn otherColumn = other.Columns[rightJoinColumn];
Dictionary <TKey, ICollection <long> > multimap = otherColumn.GroupColumnValues <TKey>(out HashSet <long> otherColumnNullIndices);
Dictionary <TKey, long> intersection = new Dictionary <TKey, long>(EqualityComparer <TKey> .Default);
// Go over the records in this dataframe and match with the dictionary
DataFrameColumn thisColumn = Columns[leftJoinColumn];
Int64DataFrameColumn thisColumnNullIndices = new Int64DataFrameColumn("ThisColumnNullIndices");
for (long i = 0; i < thisColumn.Length; i++)
{
var thisColumnValue = thisColumn[i];
if (thisColumnValue != null)
{
if (multimap.TryGetValue((TKey)thisColumnValue, out ICollection <long> rowNumbers))
{
foreach (long row in rowNumbers)
{
leftRowIndices.Append(i);
rightRowIndices.Append(row);
if (!intersection.ContainsKey((TKey)thisColumnValue))
{
intersection.Add((TKey)thisColumnValue, rowNumber);
}
}
}
else
{
leftRowIndices.Append(i);
rightRowIndices.Append(null);
}
}
else
{
thisColumnNullIndices.Append(i);
}
}
for (long i = 0; i < otherColumn.Length; i++)
{
var value = otherColumn[i];
if (value != null)
{
if (!intersection.ContainsKey((TKey)value))
{
leftRowIndices.Append(null);
rightRowIndices.Append(i);
}
}
}
// Now handle the null rows
foreach (long?thisColumnNullIndex in thisColumnNullIndices)
{
foreach (long otherColumnNullIndex in otherColumnNullIndices)
{
leftRowIndices.Append(thisColumnNullIndex.Value);
rightRowIndices.Append(otherColumnNullIndex);
}
if (otherColumnNullIndices.Count == 0)
{
leftRowIndices.Append(thisColumnNullIndex.Value);
rightRowIndices.Append(null);
}
}
if (thisColumnNullIndices.Length == 0)
{
foreach (long otherColumnNullIndex in otherColumnNullIndices)
{
leftRowIndices.Append(null);
rightRowIndices.Append(otherColumnNullIndex);
}
}
}
else
{
throw new NotImplementedException(nameof(joinAlgorithm));
}
for (int i = 0; i < leftDataFrame.Columns.Count; i++)
{
ret.Columns.Insert(i, leftDataFrame.Columns[i].Clone(leftRowIndices));
}
for (int i = 0; i < rightDataFrame.Columns.Count; i++)
{
DataFrameColumn column = rightDataFrame.Columns[i].Clone(rightRowIndices);
SetSuffixForDuplicatedColumnNames(ret, column, leftSuffix, rightSuffix);
ret.Columns.Insert(ret.Columns.Count, column);
}
return(ret);
}