/// <summary>
/// Creates a CardinalityEstimator with the given <paramref name="state" />
/// </summary>
internal CardinalityEstimator(CardinalityEstimatorState state)
{
this.bitsPerIndex = state.BitsPerIndex;
this.bitsForHll = 64 - this.bitsPerIndex;
this.m = (int)Math.Pow(2, this.bitsPerIndex);
this.alphaM = GetAlphaM(this.m);
this.subAlgorithmSelectionThreshold = GetSubAlgorithmSelectionThreshold(this.bitsPerIndex);
// Init the hash function
this.hashFunctionId = state.HashFunctionId;
this.hashFunction = HashFunctionFactory.GetHashFunction(this.hashFunctionId);
// Init the direct count
this.directCount = state.DirectCount != null ? new HashSet <ulong>(state.DirectCount) : null;
// Init the sparse representation
this.isSparse = state.IsSparse;
this.lookupSparse = state.LookupSparse != null ? new Dictionary <ushort, byte>(state.LookupSparse) : null;
this.lookupDense = state.LookupDense;
this.CountAdditions = state.CountAdditions;
// Each element in the sparse representation takes 15 bytes, and there is some constant overhead
this.sparseMaxElements = Math.Max(0, this.m / 15 - 10);
// If necessary, switch to the dense representation
if (this.sparseMaxElements <= 0)
{
SwitchToDenseRepresentation();
}
// if DirectCount is not null, populate the HLL lookup with its elements. This allows serialization to include only directCount
if (this.directCount != null)
{
foreach (ulong element in this.directCount)
{
AddElementHash(element);
}
}
else
{
this.directCount = null;
}
}
/// <summary>
/// Serialize the given <paramref name="cardinalityEstimator" /> to <paramref name="stream" />
/// </summary>
public void Serialize(Stream stream, CardinalityEstimator cardinalityEstimator)
{
using (var bw = new BinaryWriter(stream))
{
bw.Write(DataFormatMajorVersion);
bw.Write(DataFormatMinorVersion);
CardinalityEstimatorState data = cardinalityEstimator.GetState();
bw.Write((byte)data.HashFunctionId);
bw.Write(data.BitsPerIndex);
bw.Write((byte)(((data.IsSparse ? 1 : 0) << 1) + (data.DirectCount != null ? 1 : 0)));
if (data.DirectCount != null)
{
bw.Write(data.DirectCount.Count);
foreach (ulong element in data.DirectCount)
{
bw.Write(element);
}
}
else if (data.IsSparse)
{
bw.Write(data.LookupSparse.Count);
foreach (KeyValuePair <ushort, byte> element in data.LookupSparse)
{
bw.Write(element.Key);
bw.Write(element.Value);
}
}
else
{
bw.Write(data.LookupDense.Length);
foreach (byte element in data.LookupDense)
{
bw.Write(element);
}
}
bw.Write(data.CountAdditions);
bw.Flush();
}
}
/// <summary>
/// Deserialize a <see cref="CardinalityEstimator" /> from the given <paramref name="stream" />
/// </summary>
public CardinalityEstimator Deserialize(Stream stream)
{
using (var br = new BinaryReader(stream))
{
int dataFormatMajorVersion = br.ReadUInt16();
int dataFormatMinorVersion = br.ReadUInt16();
AssertDataVersionCanBeRead(dataFormatMajorVersion, dataFormatMinorVersion);
HashFunctionId hashFunctionId;
if (dataFormatMajorVersion >= 2)
{
// Starting with version 2.0, the serializer writes the hash function ID
hashFunctionId = (HashFunctionId)br.ReadByte();
}
else
{
// Versions before 2.0 all used FNV-1a
hashFunctionId = HashFunctionId.Fnv1A;
}
int bitsPerIndex = br.ReadInt32();
byte flags = br.ReadByte();
bool isSparse = ((flags & 2) == 2);
bool isDirectCount = ((flags & 1) == 1);
HashSet <ulong> directCount = null;
IDictionary <ushort, byte> lookupSparse = isSparse ? new Dictionary <ushort, byte>() : null;
byte[] lookupDense = null;
if (isDirectCount)
{
int count = br.ReadInt32();
directCount = new HashSet <ulong>();
for (var i = 0; i < count; i++)
{
ulong element = br.ReadUInt64();
directCount.Add(element);
}
}
else if (isSparse)
{
int count = br.ReadInt32();
for (var i = 0; i < count; i++)
{
ushort elementKey = br.ReadUInt16();
byte elementValue = br.ReadByte();
lookupSparse.Add(elementKey, elementValue);
}
}
else
{
int count = br.ReadInt32();
lookupDense = br.ReadBytes(count);
}
// Starting with version 2.1, the serializer writes CountAdditions
ulong countAdditions = 0UL;
if (dataFormatMajorVersion >= 2 && dataFormatMinorVersion >= 1)
{
countAdditions = br.ReadUInt64();
}
var data = new CardinalityEstimatorState
{
HashFunctionId = hashFunctionId,
BitsPerIndex = bitsPerIndex,
DirectCount = directCount,
IsSparse = isSparse,
LookupDense = lookupDense,
LookupSparse = lookupSparse,
CountAdditions = countAdditions,
};
var result = new CardinalityEstimator(data);
return(result);
}
}
/// <summary>
/// Deserialize a <see cref="CardinalityEstimator" /> from the given <paramref name="stream" />
/// </summary>
public CardinalityEstimator Deserialize(Stream stream)
{
using (var br = new BinaryReader(stream))
{
int dataFormatMajorVersion = br.ReadUInt16();
int dataFormatMinorVersion = br.ReadUInt16();
AssertDataVersionCanBeRead(dataFormatMajorVersion, dataFormatMinorVersion);
HashFunctionId hashFunctionId;
if (dataFormatMajorVersion >= 2)
{
// Starting with version 2.0, the serializer writes the hash function ID
hashFunctionId = (HashFunctionId)br.ReadByte();
}
else
{
// Versions before 2.0 all used FNV-1a
hashFunctionId = HashFunctionId.Fnv1A;
}
int bitsPerIndex = br.ReadInt32();
byte flags = br.ReadByte();
bool isSparse = ((flags & 2) == 2);
bool isDirectCount = ((flags & 1) == 1);
HashSet<ulong> directCount = null;
IDictionary<ushort, byte> lookupSparse = isSparse ? new Dictionary<ushort, byte>() : null;
byte[] lookupDense = null;
if (isDirectCount)
{
int count = br.ReadInt32();
directCount = new HashSet<ulong>();
for (var i = 0; i < count; i++)
{
ulong element = br.ReadUInt64();
directCount.Add(element);
}
}
else if (isSparse)
{
int count = br.ReadInt32();
for (var i = 0; i < count; i++)
{
ushort elementKey = br.ReadUInt16();
byte elementValue = br.ReadByte();
lookupSparse.Add(elementKey, elementValue);
}
}
else
{
int count = br.ReadInt32();
lookupDense = br.ReadBytes(count);
}
// Starting with version 2.1, the serializer writes CountAdditions
ulong countAdditions = 0UL;
if (dataFormatMajorVersion >= 2 && dataFormatMinorVersion >= 1)
{
countAdditions = br.ReadUInt64();
}
var data = new CardinalityEstimatorState
{
HashFunctionId = hashFunctionId,
BitsPerIndex = bitsPerIndex,
DirectCount = directCount,
IsSparse = isSparse,
LookupDense = lookupDense,
LookupSparse = lookupSparse,
CountAdditions = countAdditions,
};
var result = new CardinalityEstimator(data);
return result;
}
}
