public async Task <BaseObject> Write(BaseObject item)
{
if (item == null)
{
throw new ArgumentNullException(nameof(item));
}
else if (string.IsNullOrWhiteSpace(item.id))
{
throw new ArgumentNullException("item.id");
}
item.published = item.published ?? DateTime.UtcNow;
var storePath = Localize(filter.ToPath(item.PublicId));
if (!File.Exists(storePath))
{
using (var storeWriter = new ObjectStreamWriter(File.Open(storePath, FileMode.CreateNew, FileAccess.Write, FileShare.Read), _serializer, ctx))
{
await storeWriter.Write(item);
}
foreach (var writer in Writers)
{
await writer.Write(item);
}
}
return(item);
}
public void ProcessPointsTest()
{
CloudBlobContainer container = AzureHelper.StorageAccount.CreateCloudBlobClient().GetContainerReference("test");
container.CreateIfNotExist();
CloudBlob points = container.GetBlobReference(Guid.NewGuid().ToString());
CloudBlob centroids = container.GetBlobReference(Guid.NewGuid().ToString());
const int NumPoints = 100, NumCentroids = 10;
using (ObjectStreamWriter<ClusterPoint> pointStream = new ObjectStreamWriter<ClusterPoint>(points, point => point.ToByteArray(), ClusterPoint.Size))
{
for (int i = 0; i < NumPoints; i++)
{
pointStream.Write(new ClusterPoint(1, 2, Guid.Empty));
}
}
Guid centroidID = Guid.NewGuid();
using (ObjectStreamWriter<Centroid> stream = new ObjectStreamWriter<Centroid>(centroids, point => point.ToByteArray(), Centroid.Size))
{
stream.Write(new Centroid(centroidID, 3, 4));
for (int i = 0; i < NumCentroids - 1; i++)
{
stream.Write(new Centroid(Guid.NewGuid(), 1000, 1000));
}
}
KMeansTaskProcessor_Accessor target = new KMeansTaskProcessor_Accessor(new KMeansTaskData(Guid.NewGuid(), Guid.NewGuid(), NumPoints, points.Uri, NumCentroids, 1, 0, 0, centroids.Uri, DateTime.UtcNow, DateTime.UtcNow, 0, null));
System.Diagnostics.Trace.WriteLine("Entering InitializeCentroids");
target.InitializeCentroids();
System.Diagnostics.Trace.WriteLine("Entering ProcessPoints");
System.Diagnostics.Trace.WriteLine("ProcessPoints took " + AzureHelper.Time(() =>
{
target.ProcessPoints();
}).TotalSeconds + " seconds");
// Commit the blocks
CloudBlockBlob newPointsBlob = AzureHelper.GetBlob(target.TaskResult.Points);
using (Stream stream = AzureHelper.GetBlob(target.TaskResult.PointsBlockListBlob).OpenRead())
{
BinaryFormatter bf = new BinaryFormatter();
List<string> pointsBlockList = bf.Deserialize(stream) as List<string>;
newPointsBlob.PutBlockList(pointsBlockList);
}
using (ObjectStreamReader<ClusterPoint> stream = new ObjectStreamReader<ClusterPoint>(newPointsBlob, ClusterPoint.FromByteArray, ClusterPoint.Size))
{
foreach (ClusterPoint p in stream)
{
Assert.AreEqual(centroidID, p.CentroidID);
}
}
Assert.AreEqual(NumPoints, target.TaskResult.NumPointsChanged);
Assert.IsTrue(target.TaskResult.PointsProcessedDataByCentroid.ContainsKey(centroidID));
Assert.AreEqual(NumPoints, target.TaskResult.PointsProcessedDataByCentroid[centroidID].NumPointsProcessed);
const double Epsilon = 0.0001;
Assert.IsTrue(Math.Abs((1 * NumPoints) - target.TaskResult.PointsProcessedDataByCentroid[centroidID].PartialPointSum.X) < Epsilon);
Assert.IsTrue(Math.Abs((2 * NumPoints) - target.TaskResult.PointsProcessedDataByCentroid[centroidID].PartialPointSum.Y) < Epsilon);
}
private void RecalculateCentroids()
{
AzureHelper.LogPerformance(() =>
{
// Initialize the output blob
CloudBlob writeBlob = AzureHelper.CreateBlob(jobData.JobID.ToString(), Guid.NewGuid().ToString());
// Do the mapping and write the new blob
using (ObjectStreamReader<Centroid> stream = new ObjectStreamReader<Centroid>(Centroids, Centroid.FromByteArray, Centroid.Size))
{
var newCentroids = stream.Select(c =>
{
Point newCentroidPoint;
if (totalPointsProcessedDataByCentroid.ContainsKey(c.ID) && totalPointsProcessedDataByCentroid[c.ID].NumPointsProcessed != 0)
{
newCentroidPoint = totalPointsProcessedDataByCentroid[c.ID].PartialPointSum
/ (double)totalPointsProcessedDataByCentroid[c.ID].NumPointsProcessed;
}
else
{
newCentroidPoint = new Point();
}
c.X = newCentroidPoint.X;
c.Y = newCentroidPoint.Y;
return c;
});
using (ObjectStreamWriter<Centroid> writeStream = new ObjectStreamWriter<Centroid>(writeBlob, point => point.ToByteArray(), Centroid.Size))
{
foreach (Centroid c in newCentroids)
{
writeStream.Write(c);
}
}
}
// Copy the contents of the new blob back into the old blob
Centroids.CopyFromBlob(writeBlob);
System.Diagnostics.Trace.TraceInformation("[ServerRole] Finished RecalculateCentroids(). Total points changed: {0}", TotalNumPointsChanged);
ResetPointChangedCounts();
}, jobData.JobID.ToString(), methodName: "RecalculateCentroids", iterationCount: IterationCount, points: Points.Uri.ToString(), centroids: Centroids.Uri.ToString(), machineID: MachineID);
}
/// <summary>
/// Sets up the Azure storage (Points and Centroids) for the first k-means iteration.
/// </summary>
public void InitializeStorage()
{
AzureHelper.LogPerformance(() =>
{
Random random = new Random();
if (jobData.Points == null)
{
// Initialize the points blob with N random ClusterPoints
Points = AzureHelper.CreateBlob(jobData.JobID.ToString(), AzureHelper.PointsBlob);
using (ObjectStreamWriter<ClusterPoint> stream = new ObjectStreamWriter<ClusterPoint>(Points, point => point.ToByteArray(), ClusterPoint.Size))
{
for (int i = 0; i < jobData.N; i++)
{
stream.Write(new ClusterPoint(
random.NextDouble() * 100 - 50,
random.NextDouble() * 100 - 50,
Guid.Empty));
}
}
}
else
{
// Use the given points blob
Points = AzureHelper.GetBlob(jobData.Points);
// Initialize N based on that
using (ObjectStreamReader<ClusterPoint> stream = new ObjectStreamReader<ClusterPoint>(Points, ClusterPoint.FromByteArray, ClusterPoint.Size))
{
jobData.N = (int)stream.Length;
}
}
// Initialize the centroids blob with K random Centroids
Centroids = AzureHelper.CreateBlob(jobData.JobID.ToString(), AzureHelper.CentroidsBlob);
using (ObjectStreamWriter<Centroid> stream = new ObjectStreamWriter<Centroid>(Centroids, point => point.ToByteArray(), Centroid.Size))
{
for (int i = 0; i < jobData.K; i++)
{
stream.Write(new Centroid(
Guid.NewGuid(),
random.Next(-PointRange, PointRange),
random.Next(-PointRange, PointRange)));
}
}
}, jobID: jobData.JobID.ToString(), methodName: "InitializeStorage", iterationCount: IterationCount, points: new Lazy<string>(() => Points.Uri.ToString()), centroids: new Lazy<string>(() => Centroids.Uri.ToString()), machineID: MachineID);
}
public void ProcessPointsTest()
{
CloudBlobContainer container = AzureHelper.StorageAccount.CreateCloudBlobClient().GetContainerReference("test");
container.CreateIfNotExist();
CloudBlob points = container.GetBlobReference(Guid.NewGuid().ToString());
CloudBlob centroids = container.GetBlobReference(Guid.NewGuid().ToString());
const int NumPoints = 100, NumCentroids = 10;
using (ObjectStreamWriter <ClusterPoint> pointStream = new ObjectStreamWriter <ClusterPoint>(points, point => point.ToByteArray(), ClusterPoint.Size))
{
for (int i = 0; i < NumPoints; i++)
{
pointStream.Write(new ClusterPoint(1, 2, Guid.Empty));
}
}
Guid centroidID = Guid.NewGuid();
using (ObjectStreamWriter <Centroid> stream = new ObjectStreamWriter <Centroid>(centroids, point => point.ToByteArray(), Centroid.Size))
{
stream.Write(new Centroid(centroidID, 3, 4));
for (int i = 0; i < NumCentroids - 1; i++)
{
stream.Write(new Centroid(Guid.NewGuid(), 1000, 1000));
}
}
KMeansTaskProcessor_Accessor target = new KMeansTaskProcessor_Accessor(new KMeansTaskData(Guid.NewGuid(), Guid.NewGuid(), NumPoints, points.Uri, NumCentroids, 1, 0, 0, centroids.Uri, DateTime.UtcNow, DateTime.UtcNow, 0, null));
System.Diagnostics.Trace.WriteLine("Entering InitializeCentroids");
target.InitializeCentroids();
System.Diagnostics.Trace.WriteLine("Entering ProcessPoints");
System.Diagnostics.Trace.WriteLine("ProcessPoints took " + AzureHelper.Time(() =>
{
target.ProcessPoints();
}).TotalSeconds + " seconds");
// Commit the blocks
CloudBlockBlob newPointsBlob = AzureHelper.GetBlob(target.TaskResult.Points);
using (Stream stream = AzureHelper.GetBlob(target.TaskResult.PointsBlockListBlob).OpenRead())
{
BinaryFormatter bf = new BinaryFormatter();
List <string> pointsBlockList = bf.Deserialize(stream) as List <string>;
newPointsBlob.PutBlockList(pointsBlockList);
}
using (ObjectStreamReader <ClusterPoint> stream = new ObjectStreamReader <ClusterPoint>(newPointsBlob, ClusterPoint.FromByteArray, ClusterPoint.Size))
{
foreach (ClusterPoint p in stream)
{
Assert.AreEqual(centroidID, p.CentroidID);
}
}
Assert.AreEqual(NumPoints, target.TaskResult.NumPointsChanged);
Assert.IsTrue(target.TaskResult.PointsProcessedDataByCentroid.ContainsKey(centroidID));
Assert.AreEqual(NumPoints, target.TaskResult.PointsProcessedDataByCentroid[centroidID].NumPointsProcessed);
const double Epsilon = 0.0001;
Assert.IsTrue(Math.Abs((1 * NumPoints) - target.TaskResult.PointsProcessedDataByCentroid[centroidID].PartialPointSum.X) < Epsilon);
Assert.IsTrue(Math.Abs((2 * NumPoints) - target.TaskResult.PointsProcessedDataByCentroid[centroidID].PartialPointSum.Y) < Epsilon);
}