public override RecordSet Initialize(DataSet Data, Predicate Where, FNodeSet Fields, int Clusters)
{
// Get the min of each field //
AggregateSet set1 = new AggregateSet();
for (int i = 0; i < Fields.Count; i++)
{
set1.Add(new AggregateMin(Fields[i].CloneOfMe()), Fields.Alias(i));
}
// Get the max of each field //
AggregateSet set2 = new AggregateSet();
for (int i = 0; i < Fields.Count; i++)
{
set2.Add(new AggregateMax(Fields[i].CloneOfMe()), Fields.Alias(i));
}
// Render the min and max //
RecordSet rs1 = AggregatePlan.Render(Data, Where, new FNodeSet(), set1);
RecordSet rs2 = AggregatePlan.Render(Data, Where, new FNodeSet(), set2);
// Create the output means table //
RecordSet rs = new RecordSet(Schema.Join(new Schema("key int, count double"), rs1.Columns));
// Fill in the gaps //
for (int i = 0; i < Clusters; i++)
{
if (i == 0)
{
RecordBuilder rb = new RecordBuilder();
rb.Add(0);
rb.Add(0D);
rb.Add(rs1[0]);
rs.Add(rb.ToRecord());
}
else if (i == Clusters - 1)
{
RecordBuilder rb = new RecordBuilder();
rb.Add(Clusters - 1);
rb.Add(0D);
rb.Add(rs2[0]);
rs.Add(rb.ToRecord());
}
else
{
RecordBuilder rb = new RecordBuilder();
rb.Add(i);
rb.Add(0D);
for (int j = 0; j < rs1.Columns.Count; j++)
{
double clus = (double)Clusters;
double jay = (double)j;
rb.Add(rs1[0][j].DOUBLE + (rs2[0][j].DOUBLE - rs1[0][j].DOUBLE) / clus * jay);
}
rs.Add(rb.ToRecord());
}
}
return rs;
}
private bool ItterateOnce()
{
// Create the cluster mapping FNode; this node does the nearest neighbor test //
FNodeSet keys = new FNodeSet();
FNode n = new FNodeResult(null, new RowClusterCellFunction(this._rule, this._means));
foreach (FNode t in this._fields.Nodes)
{
n.AddChildNode(t.CloneOfMe());
}
keys.Add("CLUSTER_ID", n);
// Create the aggregate//
AggregateSet set = new AggregateSet();
// Add a counter to the aggregate //
set.Add(new AggregateSum(FNodeFactory.Value(1D)), "CLUSTER_ELEMENT_COUNT");
// load the aggregate with the mean aggregates //
for (int i = 0; i < this._fields.Count; i++)
{
set.Add(new AggregateAverage(this._fields[i].CloneOfMe()), this._fields.Alias(i));
}
// Load the aggregate with the variance aggregates //
for (int i = 0; i < this._fields.Count; i++)
{
set.Add(new AggregateVarianceP(this._fields[i].CloneOfMe()), "VAR_" + this._fields.Alias(i));
}
// Run the aggregate; this is basically a horse aggregate step with the cluster node mapping as the key, and averaging as the value
RecordSet rs = AggregatePlan.Render(this._data, this._where, keys, set);
// Need to chop up the recordset we just created //
Key mean_keeper = Key.Build(this._means.Columns.Count);
RecordSet means = FastReadPlan.Render(rs, Predicate.TrueForAll, mean_keeper, long.MaxValue);
Key stat_keeper = new Key(0,1); // keep the id and the count
for (int i = mean_keeper.Count; i < rs.Columns.Count; i++)
{
stat_keeper.Add(i);
}
this._stats = FastReadPlan.Render(rs, Predicate.TrueForAll, stat_keeper, long.MaxValue);
// Check for cluster misses; cluster misses occur when no node maps to a cluster correctly //
if (means.Count != this._means.Count)
{
this.HandleNullCluster(means);
}
// Compare the changes between itterations
double change = this.CompareChanges(this._means, means);
// Set the means to the newly calculated means //
this._means = means;
// Return a boolean indicating if we failed or not
return change < this._exit_condition;
}
public override RecordSet Initialize(DataSet Data, Predicate Where, FNodeSet Fields, int Clusters)
{
AggregateSet set = new AggregateSet();
set.Add(new AggregateSum(FNodeFactory.Value(1D)), "CLUSTER_ELEMENT_COUNT");
for (int i = 0; i < Fields.Count; i++)
{
set.Add(new AggregateAverage(Fields[i].CloneOfMe()), Fields.Alias(i));
}
FNode rnd = new FNodeResult(null, new CellRandomInt());
rnd.AddChildNode(new FNodeValue(rnd, new Cell(this.Seed)));
rnd.AddChildNode(new FNodeValue(rnd, new Cell(0)));
rnd.AddChildNode(new FNodeValue(rnd, new Cell(Clusters)));
FNodeSet keys = new FNodeSet();
keys.Add(rnd);
RecordSet rs = AggregatePlan.Render(Data, Where, keys, set);
return rs;
}
public AggregateSet CloneOfMe()
{
AggregateSet agg = new AggregateSet();
for (int i = 0; i < this.Count; i++)
{
agg.Add(this._cache[i].CloneOfMe(), this._alias[i]);
}
return agg;
}