/// <summary>
/// Enumerate sample batches
/// </summary>
public IEnumerable <ClassifiedSample <TObj> > Batch(int size)
{
if (size <= 0)
{
throw new MLException("Batch size must be positive");
}
var bucket = new ClassifiedSample <TObj>();
var count = 0;
foreach (var item in this)
{
bucket[item.Key] = item.Value;
count++;
if (count != size)
{
continue;
}
yield return(bucket);
bucket.Clear();
count = 0;
}
if (bucket.Count == 0)
{
yield break;
}
yield return(bucket);
}
protected AlgorithmBase(ClassifiedSample <TObj> trainingSample)
{
if (trainingSample == null || !trainingSample.Any())
{
throw new MLException("AlrogithmBase.ctor(trainingSample=null|empty)");
}
m_TrainingSample = new ClassifiedSample <TObj>(trainingSample);
m_Classes = m_TrainingSample.Classes.ToDictionary(c => c.Name);
}
/// <summary>
/// Returns all errors of the given algorithm on some initially classified sample
/// </summary>
public virtual IEnumerable <ErrorInfo> GetErrors(ClassifiedSample <TObj> classifiedSample)
{
var errors = new List <ErrorInfo>();
Parallel.ForEach(classifiedSample, pdata =>
{
var res = this.Classify(pdata.Key);
if (res != pdata.Value)
{
lock (errors)
{
errors.Add(new ErrorInfo(pdata.Key, pdata.Value, res));
}
}
});
return(errors);
}
/// <summary>
/// Retrieves subset of the sample
public ClassifiedSample <TObj> Subset(int skip, int take)
{
if (skip < 0)
{
throw new MLException("Skip value must be non-negative");
}
if (take <= 0)
{
throw new MLException("Take value must be positive");
}
var result = new ClassifiedSample <TObj>();
foreach (var item in this.Skip(skip).Take(take))
{
result[item.Key] = item.Value;
}
return(result);
}
public ClassifiedSample(ClassifiedSample <TObj> other) : base(other)
{
}
