private static VBuffer <Float>[] GaussianMatrix(int k, int d, int seed)
{
var rv = Zeros(k, d);
var rng = new SysRandom(seed);
// REVIEW: use a faster Gaussian random matrix generator
//MKL has a fast vectorized random number generation.
for (var i = 0; i < k; ++i)
{
for (var j = 0; j < d; ++j)
{
rv[i].Values[j] = (Float)Stats.SampleFromGaussian(rng); // not fast for large matrix generation
}
}
return(rv);
}
/// <summary>
/// Draws n random integers in [0, N-1].
/// They can be distinct or not.
/// The function is not efficient if n is close to N and distinct is true.
/// </summary>
public static int[] RandomIntegers(int n, int N, bool distinct = false, IRandom rand = null)
{
var res = new int[n];
if (rand == null)
{
rand = new SysRandom();
}
if (distinct)
{
if (n > N)
{
throw new DataValueError($"Cannot draw more than {N} distinct values.");
}
var hash = new HashSet <int>();
int nb = 0;
int i;
while (nb < n)
{
i = rand.Next(N);
if (hash.Contains(i))
{
continue;
}
hash.Add(i);
res[nb] = i;
++nb;
}
}
else
{
for (int i = 0; i < n; ++i)
{
res[i] = rand.Next(N);
}
}
return(res);
}
/// <summary>
/// Sample child configuration from configuration centered at parent, using fitness proportional mutation.
/// </summary>
/// <param name="parent">Starting parent configuration (used as mean in multivariate Gaussian).</param>
/// <param name="fitness">Numeric value indicating how good a configuration parent is.</param>
/// <param name="n">Count of how many items currently in history.</param>
/// <param name="previousRuns">Run history.</param>
/// <param name="rMean">Mean metric value of previous random runs.</param>
/// <param name="rVar">Metric value empirical variance of previous random runs.</param>
/// <param name="isMetricMaximizing">Flag for if we are minimizing or maximizing values.</param>
/// <returns>A mutated version of parent (i.e., point sampled near parent).</returns>
private ParameterSet SampleChild(ParameterSet parent, double fitness, int n, IRunResult[] previousRuns, double rMean, double rVar, bool isMetricMaximizing)
{
Float[] child = SweeperProbabilityUtils.ParameterSetAsFloatArray(_host, _sweepParameters, parent, false);
List <int> numericParamIndices = new List <int>();
List <double> numericParamValues = new List <double>();
int loopCount = 0;
// Interleave uniform random samples, according to proportion defined.
if (_spu.SampleUniform() <= _args.ProportionRandom)
{
ParameterSet ps = _randomSweeper.ProposeSweeps(1)[0];
_randomParamSets.Add(ps);
return(ps);
}
do
{
for (int i = 0; i < _sweepParameters.Length; i++)
{
// This allows us to query possible values of this parameter.
var sweepParam = _sweepParameters[i];
if (sweepParam is DiscreteValueGenerator parameterDiscrete)
{
// Sample categorical parameter.
double[] categoryWeights = _args.LegacyDpBehavior
? CategoriesToWeightsOld(parameterDiscrete, previousRuns)
: CategoriesToWeights(parameterDiscrete, previousRuns);
child[i] = SampleCategoricalDist(1, categoryWeights)[0];
}
else
{
var parameterNumeric = sweepParam as INumericValueGenerator;
_host.Check(parameterNumeric != null, "KDO sweeper can only sweep over discrete and numeric parameters");
numericParamIndices.Add(i);
numericParamValues.Add(child[i]);
}
}
if (numericParamIndices.Count > 0)
{
if (!_args.Beta)
{
// Sample point from multivariate Gaussian, centered on parent values, with mutation proportional to fitness.
double[] mu = numericParamValues.ToArray();
double correctedVal = isMetricMaximizing
? 1.0 - SweeperProbabilityUtils.NormalCdf(fitness, rMean, rVar)
: 1.0 - SweeperProbabilityUtils.NormalCdf(2 * rMean - fitness, rMean, rVar);
double bandwidthScale = Math.Max(_args.MinimumMutationSpread, correctedVal);
double[] stddevs = Enumerable.Repeat(_args.Simple ? 0.2 : bandwidthScale, mu.Length).ToArray();
double[][] bandwidthMatrix = BuildBandwidthMatrix(n, stddevs);
double[] sampledPoint = SampleDiagonalCovMultivariateGaussian(1, mu, bandwidthMatrix)[0];
for (int j = 0; j < sampledPoint.Length; j++)
{
child[numericParamIndices[j]] = (Float)Corral(sampledPoint[j]);
}
}
else
{
// If Beta flag set, sample from independent Beta distributions instead.
SysRandom rng = new SysRandom();
double alpha = 1 + 15 * fitness;
foreach (int index in numericParamIndices)
{
const double epsCutoff = 1e-10;
double eps = Math.Min(Math.Max(child[index], epsCutoff), 1 - epsCutoff);
double beta = alpha / eps - alpha;
child[index] = (Float)Stats.SampleFromBeta(rng, alpha, beta);
}
}
}
// Don't get stuck at local point.
loopCount++;
if (loopCount > 10)
{
return(_randomSweeper.ProposeSweeps(1, null)[0]);
}
} while (_alreadySeenConfigs.Contains(child));
_alreadySeenConfigs.Add(child);
return(SweeperProbabilityUtils.FloatArrayAsParameterSet(_host, _sweepParameters, child, false));
}
public void FactoryExampleTest()
{
var dataset = TestDatasets.adultText;
string dataFilename = GetDataPath(dataset.trainFilename);
string testDataFilename = GetDataPath(dataset.testFilename);
///********* Training a model *******//
string modelFilename = Path.GetTempFileName();
TLCArguments cmd = new TLCArguments();
Assert.True(CmdParser.ParseArguments(dataset.extraSettings, cmd));
cmd.command = Command.Train;
cmd.modelfile = modelFilename;
cmd.datafile = dataFilename;
cmd.instancesSettings = dataset.settings;
cmd.classifierName = TestLearners.linearSVM.Trainer;
RunExperiments.Run(cmd);
// Load and make predictions with a previously saved model.
IDataModel dataModel;
IDataStats dataStats;
var predictor = (IDistributionPredictor <Instance, Float, Float>)PredictorUtils.LoadPredictor(
out dataModel, out dataStats, modelFilename);
var instanceFactory = ReflectionUtilsOld.CreateInstanceOld <IInstanceFactory, SignatureInstances>(
cmd.instancesClass, cmd.instancesSettings, null, dataModel);
bool headerSkip = true;
List <Float> outputs = new List <Float>();
List <Float> probabilities = new List <Float>();
using (StreamReader reader = new StreamReader(testDataFilename))
{
List <string> features = new List <string>();
string text;
long line = 0;
while ((text = reader.ReadLine()) != null)
{
++line;
if (string.IsNullOrWhiteSpace(text))
{
continue;
}
string[] cols = text.Split(',');
Assert.True(cols.Length == 15);
if (headerSkip)
{
// skip header line
headerSkip = false;
continue;
}
features.Clear();
// Add in the "max dimensionality"
features.Add("15");
for (int col = 0; col < cols.Length; col++)
{
string s = cols[col].Trim();
switch (col)
{
case 0:
case 2:
case 4:
case 10:
case 11:
case 12:
case 14:
// numeric feature or label -- add if non-zero
Float val = InstancesUtils.FloatParse(s);
if (val == 0) // Beware of NaNs - they should be recorded!
{
continue;
}
break;
}
features.Add(col + ":" + s);
}
Instance instance = instanceFactory.ProduceInstance(line, features.ToArray());
Float rawOutput, probability;
probability = predictor.PredictDistribution(instance, out rawOutput);
outputs.Add(rawOutput);
probabilities.Add(probability);
}
}
List <Float> originalOutputs = new List <Float>();
List <Float> originalProbabilities = new List <Float>();
var env = new LocalEnvironment(SysRandom.Wrap(RunExperiments.GetRandom(cmd)));
Instances instances = RunExperiments.CreateTestData(cmd, testDataFilename, dataModel, null, env);
foreach (Instance instance in instances)
{
Float rawOutput, probability;
probability = predictor.PredictDistribution(instance, out rawOutput);
originalOutputs.Add(rawOutput);
originalProbabilities.Add(probability);
}
CollectionAssert.Equal(outputs, originalOutputs);
CollectionAssert.Equal(probabilities, originalProbabilities);
File.Delete(modelFilename);
Done();
}
