pyACA.ToolSimpleKmeans Namespace Reference

Classes
class	CKMeansState

Functions
	ToolSimpleKmeans (V, K, numMaxIter=1000, prevState=None)
	helper function: kmeans clustering
	assignClusterLabels_I (V, state)
	computeClusterMeans_I (V, clusterIdx, K)
	reinitState_I (state, clusterIdx, K, range_V)

Function Documentation

assignClusterLabels_I()

assignClusterLabels_I	(		V,
			state )

Definition at line 51 of file ToolSimpleKmeans.py.

def assignClusterLabels_I(V, state):
 
    # number of clusters
    K = state.mu.shape[1]
 
    D = np.zeros([K, V.shape[1]])
 
    for k in range(K):
        D[k, :] = np.sqrt(np.sum((np.tile(state.mu[:, [k]], (1, V.shape[1])) - V)**2, axis=0, keepdims=True))
 
    clusterIdx = np.argmin(D, axis=0).astype(int)
    
    return clusterIdx
 
 

Here is the caller graph for this function:

computeClusterMeans_I()

computeClusterMeans_I	(		V,
			clusterIdx,
			K )

Definition at line 66 of file ToolSimpleKmeans.py.

def computeClusterMeans_I(V, clusterIdx, K):
 
    # init
    mu = np.zeros([V.shape[0], K])
 
    for k in range(K):
        if np.count_nonzero(clusterIdx == k) != 0:
            mu[:, k] = np.sum(V[:, clusterIdx == k], axis=1) / np.count_nonzero(clusterIdx == k)
     
    return CKMeansState(mu)
 
 

Here is the caller graph for this function:

reinitState_I()

reinitState_I	(		state,
			clusterIdx,
			K,
			range_V )

Definition at line 78 of file ToolSimpleKmeans.py.

def reinitState_I(state, clusterIdx, K, range_V):
 
    for k in range(K):
        if np.count_nonzero(clusterIdx == k) == 0:
            state.mu[:, k] = np.random.rand(state.mu.shape[0], 1) * (range_V[:, 1]-range_V[:, 0]) + range_V[:, 0]
 
    return state

Here is the caller graph for this function:

ToolSimpleKmeans()

ToolSimpleKmeans	(		V,
			K,
			numMaxIter = 1000,
			prevState = None )

helper function: kmeans clustering

Parameters

V	features for all train observations (dimension iNumFeatures x iNumObservations)
K	number of clusters
numMaxIter	maximum number of iterations (stop if not converged before, default: 1000)
prevState	internal state that can be stored to continue clustering later

Returns

clusterIdx: cluster index of each observation (iNumObservations)

state: result containing internal state (if needed)

Definition at line 20 of file ToolSimpleKmeans.py.

def ToolSimpleKmeans(V, K, numMaxIter=1000, prevState=None):
 
    # init
    if prevState is None:
        state = CKMeansState(V[:, np.round(np.random.rand(K) * (V.shape[1]-1)).astype(int)])
    else:
        state = CKMeansState(prevState.mu.copy())
    range_V = np.array([np.min(V, axis=1), np.max(V, axis=1)])
 
    # assign observations to clusters
    clusterIdx = assignClusterLabels_I(V, state)
 
    for j in range(numMaxIter):
        prevState = CKMeansState(state.mu.copy())
 
        # update clusters
        state = computeClusterMeans_I(V, clusterIdx, K)
 
        # reinit empty clusters
        state = reinitState_I(state, clusterIdx, K, range_V)
 
        # assign observations to clusters
        clusterIdx = assignClusterLabels_I(V, state)
   
        # if converged, break
        if np.max(np.sum(np.abs(state.mu-prevState.mu))) == 0:
            break
 
    return clusterIdx, state.mu
 
 

Here is the call graph for this function: