public static void Test_Simple()
{
int ITERATIONS = 50;
int M = 10;
int N = 50;
int K = 50;
double SPARSITY = 0.5;
float[,] tW = new float[M, K];
float[,] tH = new float[K, N];
for (int m = 0; m < M; ++m)
{
for (int k = 0; k < K; ++k)
{
tW[m, k] = (float)RandomAugmented.Instance.NextDouble();
}
}
for (int n = 0; n < N; ++n)
{
for (int k = 0; k < K; ++k)
{
tH[k, n] = (float)RandomAugmented.Instance.NextDouble();
}
}
// Then create their product
Matrix mW = new Matrix(tW);
Matrix mH = new Matrix(tH);
Matrix mV = mW.multiply(mH);
for (int m = 0; m < M; ++m)
{
for (int n = 0; n < N; ++n)
{
if (RandomAugmented.Instance.NextDouble() < SPARSITY)
{
mV[m, n] = SMALL_NUMBER;
}
}
}
float[,] V = mV.ToFloatMatrix();
// Now factorise
float[,] W = new float[M, K];
float[,] H = new float[K, N];
//NMFMultiplicativeEuclidean.Factorise(K, M, N, V, W, H, ITERATIONS);
NMFMultiplicativeDivergence.Factorise(K, M, N, V, W, H, ITERATIONS);
// Then test the resulting product
Matrix mWW = new Matrix(W);
Matrix mHH = new Matrix(H);
Matrix mVV = mWW.multiply(mHH);
// Now compare...
Logging.Info("Before:\n" + PrintMatrix(mV.ToFloatMatrix()));
Logging.Info("After:\n" + PrintMatrix(mVV.ToFloatMatrix()));
}
public static void Test_ACL()
{
ACLPaperAttributeWordCounts word_counts = ACLPaperAttributeWordCountsGenerator.WordCounts;
int M = word_counts.AttributeCount;
int N = word_counts.PaperCount;
// Create the initial matrix of word counts
float[,] V = new float[M, N];
for (int n = 0; n < N; ++n)
{
for (int i = 0; i < word_counts.attributes_in_paper[n].Length; ++i)
{
int m = word_counts.attributes_in_paper[n][i];
++V[m, n];
}
}
// Check for zero rows or columns
Logging.Info("+Checking for empty rows and columns");
int[] m_count = new int[M];
int[] n_count = new int[N];
for (int m = 0; m < M; ++m)
{
for (int n = 0; n < N; ++n)
{
if (V[m, n] > 0)
{
++m_count[m];
++n_count[n];
}
}
}
for (int m = 0; m < M; ++m)
{
if (0 == m_count[m])
{
Logging.Info("Row {0} is all zero, corresponding to term {1}", m, word_counts.attributes[m].text);
V[m % M, m % N] = 1;
}
}
for (int n = 0; n < N; ++n)
{
if (0 == n_count[n])
{
Logging.Info("Column {0} is all zero, corresponding to doc {1}", n, word_counts.acl_ids[n]);
V[n % M, n % N] = 1;
}
}
Logging.Info("-Checking for empty rows and columns");
// Normalise the columns (i.e. each document gets a "unit" vector of words)
Tools.NormaliseColumns(M, N, V);
// Now factorize the dude
int ITERATIONS = 100;
int K = 200;
float[,] W = new float[M, K];
float[,] H = new float[N, K];
//Factorise(K, M, N, D, A, P);
//NMFMultiplicativeEuclidean.Factorise(K, M, N, D, A, P, ITERATIONS);
NMFMultiplicativeDivergence.Factorise(K, M, N, V, W, H, ITERATIONS);
// Let's save these matrices so that we can do some analysis with them...
Utilities.Files.SerializeFile.Save(String.Format(@"C:\temp\nmf_{0}_W.dat", K), W);
Utilities.Files.SerializeFile.Save(String.Format(@"C:\temp\nmf_{0}_H.dat", K), H);
}