public static IPredictorProducing <Float> Create(IHostEnvironment env, ModelLoadContext ctx)
{
Contracts.CheckValue(env, nameof(env));
env.CheckValue(ctx, nameof(ctx));
ctx.CheckAtModel(GetVersionInfo());
var predictor = new LinearBinaryPredictor(env, ctx);
ICalibrator calibrator;
ctx.LoadModelOrNull <ICalibrator, SignatureLoadModel>(env, out calibrator, @"Calibrator");
if (calibrator == null)
{
return(predictor);
}
if (calibrator is IParameterMixer)
{
return(new ParameterMixingCalibratedPredictor(env, predictor, calibrator));
}
return(new SchemaBindableCalibratedPredictor(env, predictor, calibrator));
}
public void SaveText(TextWriter writer, LinearBinaryPredictor parent, RoleMappedSchema schema, int paramCountCap)
{
Contracts.AssertValue(_env);
_env.CheckValue(writer, nameof(writer));
_env.AssertValueOrNull(parent);
_env.AssertValueOrNull(schema);
writer.WriteLine();
writer.WriteLine("*** MODEL STATISTICS SUMMARY *** ");
writer.WriteLine("Count of training examples:\t{0}", _trainingExampleCount);
writer.WriteLine("Residual Deviance: \t{0}", _deviance);
writer.WriteLine("Null Deviance: \t{0}", _nullDeviance);
writer.WriteLine("AIC: \t{0}", 2 * _paramCount + _deviance);
if (parent == null)
{
return;
}
var coeffStats = GetCoefficientStatistics(parent, schema, paramCountCap);
if (coeffStats == null)
{
return;
}
writer.WriteLine();
writer.WriteLine("Coefficients statistics:");
writer.WriteLine("Coefficient \tEstimate\tStd. Error\tz value \tPr(>|z|)");
foreach (var coeffStat in coeffStats)
{
writer.WriteLine("{0,-15}\t{1,-10:G7}\t{2,-10:G7}\t{3,-10:G7}\t{4}",
coeffStat.Name,
coeffStat.Estimate,
coeffStat.StandardError,
coeffStat.ZScore,
DecorateProbabilityString(coeffStat.PValue));
}
writer.WriteLine("---");
writer.WriteLine("Significance codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1");
}
public void AddStatsColumns(List <IColumn> list, LinearBinaryPredictor parent, RoleMappedSchema schema, ref VBuffer <ReadOnlyMemory <char> > names)
{
_env.AssertValue(list);
_env.AssertValueOrNull(parent);
_env.AssertValue(schema);
long count = _trainingExampleCount;
list.Add(RowColumnUtils.GetColumn("Count of training examples", NumberType.I8, ref count));
var dev = _deviance;
list.Add(RowColumnUtils.GetColumn("Residual Deviance", NumberType.R4, ref dev));
var nullDev = _nullDeviance;
list.Add(RowColumnUtils.GetColumn("Null Deviance", NumberType.R4, ref nullDev));
var aic = 2 * _paramCount + _deviance;
list.Add(RowColumnUtils.GetColumn("AIC", NumberType.R4, ref aic));
if (parent == null)
{
return;
}
Single biasStdErr;
Single biasZScore;
Single biasPValue;
if (!TryGetBiasStatistics(parent.Statistics, parent.Bias, out biasStdErr, out biasZScore, out biasPValue))
{
return;
}
var biasEstimate = parent.Bias;
list.Add(RowColumnUtils.GetColumn("BiasEstimate", NumberType.R4, ref biasEstimate));
list.Add(RowColumnUtils.GetColumn("BiasStandardError", NumberType.R4, ref biasStdErr));
list.Add(RowColumnUtils.GetColumn("BiasZScore", NumberType.R4, ref biasZScore));
list.Add(RowColumnUtils.GetColumn("BiasPValue", NumberType.R4, ref biasPValue));
var weights = default(VBuffer <Single>);
parent.GetFeatureWeights(ref weights);
var estimate = default(VBuffer <Single>);
var stdErr = default(VBuffer <Single>);
var zScore = default(VBuffer <Single>);
var pValue = default(VBuffer <Single>);
ValueGetter <VBuffer <ReadOnlyMemory <char> > > getSlotNames;
GetUnorderedCoefficientStatistics(parent.Statistics, ref weights, ref names, ref estimate, ref stdErr, ref zScore, ref pValue, out getSlotNames);
var slotNamesCol = RowColumnUtils.GetColumn(MetadataUtils.Kinds.SlotNames,
new VectorType(TextType.Instance, stdErr.Length), getSlotNames);
var slotNamesRow = RowColumnUtils.GetRow(null, slotNamesCol);
var colType = new VectorType(NumberType.R4, stdErr.Length);
list.Add(RowColumnUtils.GetColumn("Estimate", colType, ref estimate, slotNamesRow));
list.Add(RowColumnUtils.GetColumn("StandardError", colType, ref stdErr, slotNamesRow));
list.Add(RowColumnUtils.GetColumn("ZScore", colType, ref zScore, slotNamesRow));
list.Add(RowColumnUtils.GetColumn("PValue", colType, ref pValue, slotNamesRow));
}
private IEnumerable <CoefficientStatistics> GetUnorderedCoefficientStatistics(LinearBinaryPredictor parent, RoleMappedSchema schema)
{
Contracts.AssertValue(_env);
_env.CheckValue(parent, nameof(parent));
if (!_coeffStdError.HasValue)
{
yield break;
}
var weights = parent.Weights2 as IReadOnlyList <Single>;
_env.Assert(_paramCount == 1 || weights != null);
_env.Assert(_coeffStdError.Value.Length == weights.Count + 1);
var names = default(VBuffer <ReadOnlyMemory <char> >);
MetadataUtils.GetSlotNames(schema, RoleMappedSchema.ColumnRole.Feature, weights.Count, ref names);
Single[] stdErrorValues = _coeffStdError.Value.Values;
const Double sqrt2 = 1.41421356237; // Math.Sqrt(2);
bool denseStdError = _coeffStdError.Value.IsDense;
int[] stdErrorIndices = _coeffStdError.Value.Indices;
Single[] zScores = new Single[_paramCount - 1];
for (int i = 1; i < _paramCount; i++)
{
int wi = denseStdError ? i - 1 : stdErrorIndices[i] - 1;
_env.Assert(0 <= wi && wi < weights.Count);
var name = names.GetItemOrDefault(wi).ToString();
if (string.IsNullOrEmpty(name))
{
name = $"f{wi}";
}
var weight = weights[wi];
var stdError = stdErrorValues[i];
var zScore = zScores[i - 1] = weight / stdError;
var pValue = 1 - (Single)ProbabilityFunctions.Erf(Math.Abs(zScore / sqrt2));
yield return(new CoefficientStatistics(name, weight, stdError, zScore, pValue));
}
}