internal void PreCalcFeatureValue(SampleCollection posSamples, SampleCollection negSamples)
{
// foreach (WeakClassifier weak in _weakClassifiers)
// {
// weak.PreCalcFeatureValue(posSamples, negSamples);
// }
Parallel.ForEach(_weakClassifiers,
weak =>
{
weak.PreCalcFeatureValue(posSamples, negSamples);
});
}
private PredictResult EvaluateErrorRate(SampleCollection validateSamples, double minHitRate, double maxFalsePositiveRate)
{
const MyFloat FLT_EPSILON = 1.192092896e-07F; /* smallest such that 1.0+FLT_EPSILON != 1.0 */
int numPos = validateSamples.PosCount, numNeg = validateSamples.NegCount;
int count = validateSamples.Count;
int i, numFalse = 0, numPosTrue = 0;
List <MyFloat> values = new List <MyFloat>(numPos);
//统计正样本的特征值
foreach (ISample sample in validateSamples)
{
if (sample.IsPositive)
{
values.Add(this.PredictGetSum(sample));
}
}
values.Sort();
int thresholdIdx = (int)((1.0 - minHitRate) * numPos);
_threshold = values[thresholdIdx];
numPosTrue = numPos - thresholdIdx;
for (i = thresholdIdx - 1; i >= 0; i--)
{
if (Math.Abs(values[i] - _threshold) < FLT_EPSILON)
{
numPosTrue++;
}
}
double hitRate = ((double)numPosTrue) / ((double)numPos);
//统计负样本的分类误差
foreach (ISample sample in validateSamples)
{
if (!sample.IsPositive && this.Predict(sample))
{
numFalse++;
}
}
double falseAlarm = ((double)numFalse) / ((double)numNeg);
PredictResult result = new PredictResult();
result.Count = count;
result.PosCount = numPos;
result.NegCount = numNeg;
result.FalsePositiveRate = falseAlarm;
result.HitRate = hitRate;
return(result);
}
public void PreCalcFeatureValue(SampleCollection posSamples, SampleCollection negSamples)
{
int numPos = posSamples.Count, numNeg = negSamples.Count;
int count = numNeg + numPos;
_featureValues = new MyFloat[count];
int i, j;
MyFloat value;
for (i = 0; i < numPos; i++)
{
value = _feature.GetValue(posSamples[i]);
_featureValues[i] = value;
}
j = numPos;
for (i = 0; i < numNeg; i++)
{
value = _feature.GetValue(negSamples[i]);
_featureValues[j] = value;
j++;
}
}
/// <summary>
/// 使用指定的样本集和样本权值训练所有的弱分类器
/// </summary>
/// <param name="posSamples"></param>
/// <param name="negSamples"></param>
/// <param name="sampleWeight"></param>
internal void Train(SampleCollection posSamples, SampleCollection negSamples, MyFloat[] sampleWeight)
{
// List<WeakClassifier.FeatureValueWithPosFlag> trainTmp = new List<WeakClassifier.FeatureValueWithPosFlag>(sampleWeight.Length);
// foreach (WeakClassifier weak in _weakClassifiers)
// {
// weak.Train(posSamples, negSamples, sampleWeight, trainTmp);
// }
ConcurrentStack <List <WeakClassifier.FeatureValueWithPosFlag> > stack = new ConcurrentStack <List <WeakClassifier.FeatureValueWithPosFlag> >();
Parallel.ForEach(_weakClassifiers,
() =>
{
List <WeakClassifier.FeatureValueWithPosFlag> trainTmp = null;
if (stack.Count > 0)
{
if (stack.TryPop(out trainTmp) == false)
{
trainTmp = null;
}
}
if (trainTmp == null)
{
trainTmp = new List <WeakClassifier.FeatureValueWithPosFlag>(sampleWeight.Length);
}
return(trainTmp);
},
(weak, state, trainTmp) =>
{
weak.Train(posSamples, negSamples, sampleWeight, trainTmp);
return(trainTmp);
},
(trainTmp) =>
{
stack.Push(trainTmp);
}
);
}
/// <summary>
/// 训练级联分类器中一层的强分类器
/// </summary>
/// <param name="posSamples"></param>
/// <param name="negSamples"></param>
/// <param name="maxFalsePositiveRate"></param>
/// <param name="minHitRate"></param>
/// <returns>训练的统计结果</returns>
public PredictResult Train(SampleCollection posSamples,
SampleCollection negSamples,
SampleCollection validateSamples,
double maxFalsePositiveRate,
double minHitRate)
{
List <WeakClassifier> weakClassifiers = new List <WeakClassifier>(10);
PredictResult result = new PredictResult();
MyFloat[] sampleWeight = InitWeight(posSamples.Count, negSamples.Count);
WeakClassifierManager allWeak = WeakClassifierManager.Instance;
Stopwatch watch = new Stopwatch();
watch.Start();
allWeak.PreCalcFeatureValue(posSamples, negSamples);
watch.Stop();
if (DebugMsg.Debug)
{
string msg = string.Format("所有弱分类器特征值预计算完成，用时：{0}\r\n",
watch.Elapsed.ToString());
DebugMsg.AddMessage(msg, 0);
}
int trainTime = 0;
do
{
if (++trainTime != 1)
{
NormalizeWeight(sampleWeight);
}
if (DebugMsg.Debug)
{
string msg = string.Format("开始训练第{0}个弱分类器\r\n",
trainTime);
DebugMsg.AddMessage(msg, 0);
}
watch.Reset();
watch.Start();
allWeak.Train(posSamples, negSamples, sampleWeight);
WeakClassifier newBestClassifier = AdaBoost(posSamples.Count, negSamples.Count, sampleWeight);
//UpdateWeights(newBestClassifier, posSamples.Count, negSamples.Count, sampleWeight);
weakClassifiers.Add(newBestClassifier);
_classifiers = weakClassifiers.ToArray();
result = EvaluateErrorRate(validateSamples, minHitRate, maxFalsePositiveRate);
watch.Stop();
if (DebugMsg.Debug)
{
string msg = string.Format("训练完成，花费时间{0}\r\n检测率：\t{1:P5}\t误检率：\t{2:P5}\r\n",
watch.Elapsed.ToString(),
result.HitRate,
result.FalsePositiveRate);
DebugMsg.AddMessage(msg, 1);
}
} while (result.FalsePositiveRate > maxFalsePositiveRate);
allWeak.ReleaseTrainData();
foreach (WeakClassifier weak in _classifiers)
{
weak.ReleaseTrainData();
}
return(result);
}
private bool Init(SampleCollection posSamples,
SampleCollection negSamples)
{
throw new NotImplementedException();
}