public static bool Test2()
{
var i1 = new InputNode();
var i2 = new InputNode();
var h1 = new Perceptron(new Heaviside(1.0));
var h2 = new Perceptron(new Heaviside(2.0));
var h3 = new Perceptron(new Heaviside(1.0));
LinearWeight.Connect(i1, h1, 1.0);
LinearWeight.Connect(i1, h2, 1.0);
LinearWeight.Connect(i2, h2, 1.0);
LinearWeight.Connect(i2, h3, 1.0);
var o = new Perceptron(new Linear());
LinearWeight.Connect(h1, o, 1.0);
LinearWeight.Connect(h2, o, -2.0);
LinearWeight.Connect(h3, o, 1.0);
var network = new GeneralNetwork(new List <InputNode> {
i1, i2
}, new List <INode> {
o
});
network.Forward(new[] { 0.0, 0.0 });
network.Forward(new[] { 1.0, 0.0 });
network.Forward(new[] { 0.0, 1.0 });
network.Forward(new[] { 1.0, 1.0 });
return(true);
}
public LinearWeight[] GetLinearWeigths(GeoCellTuple cell, Nodes nodes)
{
int N = nodes.Lons.Length;
LinearWeight[] weights = new LinearWeight[N];
for (int i = 0; i < N; i++)
{
weights[i] = new LinearWeight(N - i - 1, cell.LatMax * (i + 1));
}
return(weights);
}
public async Task LinearCombinationAggregatorTest()
{
LinearCombinationAggregator lca = new LinearCombinationAggregator();
var nodes = new RealValueNodes(new double[3], new double[3], new double[] { 2.0, 3.0, 5.0 });
var weights = new LinearWeight[] { new LinearWeight(1, 7.0), new LinearWeight(0, 11.0), new LinearWeight(2, 13.0) };
LinearCombinationContext lcc = new LinearCombinationContext(new Tuple <ICellRequest, RealValueNodes, IEnumerable <LinearWeight> >[] { new Tuple <ICellRequest, RealValueNodes, IEnumerable <LinearWeight> >(new RequestStubs(), nodes, (IEnumerable <LinearWeight>)weights) });
double[] result = await lca.AggregateCellsBatchAsync(lcc, new ICellRequest[] { new RequestStubs() });
Assert.AreEqual(1, result.Length);
Assert.AreEqual(108, result[0]);
}
public async Task <LinearWeight[]> GetLinearWeigthsAsync(INodes nodes, ICellRequest cell)
{
int N = nodes.Lons.Length;
LinearWeight[] weights = new LinearWeight[N];
for (int i = 0; i < N; i++)
{
weights[i] = new LinearWeight(N - i - 1, cell.LatMax * (i + 1));
}
return(weights);
}
/// <summary>
/// Returns a set of linear weights along with timeSegmint for with dataIndeces in linear weights are valid
/// </summary>
/// <param name="context"></param>
/// <param name="computationalContext"></param>
/// <param name="cells"></param>
/// <returns></returns>
protected IEnumerable <Tuple <ITimeSegment, LinearWeight[]> > CalcLinearWeights(ComputationalContext computationalContext, IEnumerable <GeoCellTuple> cells)
{
if (!computationalContext.ContainsKey("observations"))
{
throw new InvalidOperationException("Call SaveObservationsAndDalanayDiagsForCells prior calling CompleteAggregateCellsBatch");
}
ISpatPointsLinearInterpolator2D spli2d = (ISpatPointsLinearInterpolator2D)spatialIntegrator;
Dictionary <ITimeSegment, IObservationsInformation> observations = (Dictionary <ITimeSegment, IObservationsInformation>)computationalContext["observations"];
Dictionary <ITimeSegment, object> delanays = (Dictionary <ITimeSegment, object>)computationalContext["dalanays"];
var sw1 = System.Diagnostics.Stopwatch.StartNew();
TraceVerbose("Computing field values");
//WARNING: can't be paralleled as dalanays are not thread safe
foreach (var cell in cells)
{
var timeSegment = cell.Time;
var observation = observations[timeSegment].Observations;
object delanay = delanays[timeSegment];
double latmax = cell.LatMax, latmin = cell.LatMin, lonmax = cell.LonMax, lonmin = cell.LonMin;
if (latmax == latmin && lonmax == lonmin)
{
yield return(Tuple.Create(timeSegment, spli2d.GetLinearWeigths(latmin, lonmin, delanay)));
}
else
{
var sizeSqrt = (int)Math.Sqrt(cellAveragingGridSize);
LinearWeight[][] toFlatten = new LinearWeight[sizeSqrt * sizeSqrt][];
double latStep = (latmax - latmin) / (sizeSqrt - 1);
double lonStep = (lonmax - lonmin) / (sizeSqrt - 1);
for (int i = 0; i < sizeSqrt; i++)
{
for (int j = 0; j < sizeSqrt; j++)
{
var w = spli2d.GetLinearWeigths(latmin + latStep * i, lonmin + lonStep * j, delanay);
toFlatten[sizeSqrt * i + j] = w;
}
}
var devisor = sizeSqrt * sizeSqrt;
var flattenedWeights = toFlatten.SelectMany(weights => weights).GroupBy(w => w.DataIndex).Select(g => new LinearWeight(g.Key, g.Select(g1 => g1.Weight).Sum() / devisor)).ToArray();
yield return(Tuple.Create(timeSegment, flattenedWeights));
}
}
TraceVerbose("Computing field values finished in {0}", sw1.Elapsed);
}
public static bool Test1()
{
var i1 = new InputNode();
var i2 = new InputNode();
var h1 = new Perceptron(new Linear());
var h2 = new Perceptron(new Linear());
var h3 = new Perceptron(new Linear());
LinearWeight.Connect(i1, h1, 1.0);
LinearWeight.Connect(i1, h2, 1.0);
LinearWeight.Connect(i2, h2, 1.0);
LinearWeight.Connect(i2, h3, 1.0);
var ib = new BiasNode();
LinearWeight.Connect(ib, h1, 1.0);
LinearWeight.Connect(ib, h2, 2.0);
LinearWeight.Connect(ib, h3, 1.0);
var o = new Perceptron(new Linear());
LinearWeight.Connect(h1, o, 1.0);
LinearWeight.Connect(h2, o, -2.0);
LinearWeight.Connect(h3, o, 1.0);
var hb = new BiasNode();
var whbo = new LinearWeight(hb, o);
whbo[0] = 1.0;
var network = new GeneralNetwork(new List <InputNode> {
i1, i2
}, new List <INode> {
o
});
network.Forward(new[] { 0.0, 0.0 });
network.Forward(new[] { 1.0, 0.0 });
network.Forward(new[] { 0.0, 1.0 });
network.Forward(new[] { 1.0, 1.0 });
return(true);
}
