private static void AddData(BinaryReader lblReader, BinaryReader dataReader, bool isTraining)
{
// Skip magic:
lblReader.ReadInt32();
dataReader.ReadInt32();
int count = lblReader.ReadInt32();
if (dataReader.ReadInt32() != count) throw new InvalidOperationException("Counts not match!");
int rows = dataReader.ReadInt32();
int cols = dataReader.ReadInt32();
if (rows != 28 || cols != 28) throw new InvalidOperationException("Invalid image size!");
Console.WriteLine("Reading: " + count + (isTraining ? " Training" : " Validation") + " samples.");
int packSize = 1000;
for (int idx = 0; idx < count; idx+= packSize)
{
Console.WriteLine(idx);
var pack = ReadSamples(lblReader, dataReader, isTraining, packSize, idx, count);
using (var ctx = new LeCunDataEntities())
{
foreach (var sample in pack)
{
ctx.Samples.AddObject(sample);
}
ctx.SaveChanges();
}
}
Console.WriteLine("Done.");
}
protected override UnorderedNeuralVectorFlowProvider.InitializationData Initialize()
{
using (var ctx = new LeCunDataEntities())
{
int count = ctx.Samples.Count(s => s.IsTraining == IsTraining);
indexMap = ctx.Samples.Where(s => s.IsTraining == IsTraining).OrderBy(s => s.ID).Select(s => s.ID).ToArray();
return new InitializationData(count, LeCunDefinitions.RowCount * LeCunDefinitions.ColCount, LeCunDefinitions.ResultCount);
}
}
protected override IEnumerable<NeuralVectorFlow> DoGetNext(IndexSet indexes)
{
using (var ctx = new LeCunDataEntities())
{
var selectIndexes = indexes.ToDictionary(i => indexMap[i]);
foreach (var sample in (from s in ctx.Samples
where s.IsTraining == IsTraining && selectIndexes.Keys.Contains(s.ID)
select s))
{
yield return ToFlow(selectIndexes[sample.ID], sample);
}
}
}
internal IEnumerable<NeuralVectorFlow> GetAll()
{
using (var ctx = new LeCunDataEntities())
{
int index = 0;
foreach (var sample in ctx.Samples.Where(s => s.IsTraining == IsTraining).OrderBy(s => s.ID))
{
yield return ToFlow(index++, sample);
}
}
}
