Predicting Best Matching for new data

[theSage21]

Hi,

Noob question here. Once I've got my map, how do I predict the best matching units for new data?

[xgdgsc]

http://chem-eng.utoronto.ca/~datamining/Presentations/SOM.pdf

[theSage21]

@xgdgsc Let me rephrase.
How do I use somoclu to classify new data given an already existing set of .wts and .ux?
Essentially what is the equivalent of predict given that fit is done via something like
somoclu -x 10 -y 10 input_file output_file?

[ghost]

In python prediction is relatively straight forward:

W = som.codebook.reshape((som.codebook.shape[0] * som.codebook.shape[1], som.codebook.shape[2]))
X = data

D = -2*np.dot(W, X.T) + (W**2).sum(1)[:, None] + (X**2).sum(1)[:, None].T
BMU = (D==D.min(0)[None,:]).astype("float32").T
NBMU =  BMU.reshape((X.shape[0], som.codebook.shape[0], som.codebook.shape[1]))
new_bmus = np.vstack(NBMU.nonzero()[1:][::-1]).T

X can be a single instance or as many as your RAM would allow.

[theSage21]

@Dringite thank you.