如何保存Accord.Net K-Means的聚类结果以供重用？

Question

I am trying to save the clustering results of Accord.Net K-Means so that I don't have to recompute every time I run my program.

I also want to assign permanent human-readable labels to each cluster. This is not possible if I have to compute the clusters every time.

Answer 1

In Accord.NET, every object can be saved using standard .NET binary serialization. This means that you can save your K-Means using:

KMeans kmeans = ...

using (FileStream fs = new FileStream(path, FileMode.Create))
{
    new BinaryFormatter().Serialize(fs, kmeans);
}    

and load it back using:

KMeans kmeans = null;

using (FileStream fs = new FileStream(path, FileMode.Open))
{
    kmeans = new BinaryFormatter().Deserialize(fs) as KMeans;
}    

Hope it helps!

Answer 2

Accord.Net has a serializing class. In the package Accord.IO there exists the class Serializer . You can use it like

KMeans means;
Accord.IO.Serializer.Save(means, filename);

or

means = Accord.IO.Serializer.Load<KMeans>(filename);

Answer 3

Ok after messing around in Accord.Net and doing some research, I figured out how to save the Clustering, so I'm going to add the solution here in case anyone else bumps into the same problem.

Basically all you have to do is save the Cluster Centroids and then use them to initialize KMeans next time before running Compute:

// Saving the Centroids
int[] clusterIds = kmeans.Compute(observations);
double[][] centroids = kmeans.Clusters.Centroids;

...

// Initializing with Centroids
kmeans = new KMeans(numOfClusters);
kmeans.Clusters.Centroids = centroids;
int[] clusterIds = kmeans.Compute(observations);