added adjustable distances for kmeans and fuzzy

Author: K3rn1n4tor

clustering_api.py

@@ -15,18 +15,19 @@
 
 ########################################################################################################################
 
-@app.route('/kmeans/<k>/<initMethod>/<datasetID>')
-def kmeansClustering(k, initMethod, datasetID):
+@app.route('/kmeans/<k>/<initMethod>/<distance>/<datasetID>')
+def kmeansClustering(k, initMethod, distance, datasetID):
     """
     Access k-means clustering plugin.
     :param k: number of clusters
     :param initMethod:  initialization method for initial clusters
+    :param distance: distance measurement
     :param datasetID:  identifier of data set
     :return: jsonified output
     """
     try:
         data = loadData(datasetID)
-        response = runKMeans(data, int(k), initMethod)
+        response = runKMeans(data, int(k), initMethod, distance)
         return flask.jsonify(response)
     except:
         return flask.jsonify({})
@@ -72,18 +73,19 @@ def affinityPropagationClustering(damping, factor, preference, distance, dataset
 
 ########################################################################################################################
 
-@app.route('/fuzzy/<numClusters>/<m>/<threshold>/<datasetID>')
-def fuzzyClustering(numClusters, m, threshold, datasetID):
+@app.route('/fuzzy/<numClusters>/<m>/<threshold>/<distance>/<datasetID>')
+def fuzzyClustering(numClusters, m, threshold, distance, datasetID):
     """
     :param numClusters:
     :param m:
     :param threshold:
+    :param distance:
     :param datasetID:
     :return:
     """
     try:
         data = loadData(datasetID)
-        response = runFuzzy(data, int(numClusters), float(m), float(threshold))
+        response = runFuzzy(data, int(numClusters), float(m), float(threshold), distance)
         return flask.jsonify(response)
     except:
         return flask.jsonify({})

clustering_fuzzy.py

@@ -205,12 +205,12 @@ def _plugin_initialize():
 
 # ----------------------------------------------------------------------------------------------------------------------
 
-def create(data, numCluster, m, threshold):
+def create(data, numCluster, m, threshold, distance):
     """
     by convention contain a factory called create returning the extension implementation
     :return:
     """
-    return Fuzzy(data, numCluster, m, threshold)
+    return Fuzzy(data, numCluster, m, threshold, distance)
 
 ########################################################################################################################
 

clustering_kmeans.py

@@ -28,14 +28,15 @@ class KMeans:
     Implementation detail: <https://en.wikipedia.org/wiki/K-means_clustering>
     """
 
-    def __init__(self, obs, k, initMode='kmeans++', iters=1000, compare='sqeuclidean'):
+    def __init__(self, obs, k, initMode='kmeans++', distance='sqeuclidean', iters=1000):
         """
         Initializes the algorithm with observation, number of k clusters, the initial method and
         the maximum number of iterations.
         Initialization method of random cluster choice can be: forgy, uniform, random, plusplus
         :param obs: genomic data / matrix
         :param k: number of clusters
         :param initMode: initialization method
+        :param distance: distance measurement
         :param iters: number of maximum iterations
         :return:
         """
@@ -59,7 +60,7 @@ def __init__(self, obs, k, initMode='kmeans++', iters=1000, compare='sqeuclidean
         # initialization method
         self.__initMode = initMode
         # compare function
-        self.__compare = compare
+        self.__distance = distance
 
     # ------------------------------------------------------------------------------------------------------------------
 
@@ -150,7 +151,7 @@ def __plusplusMethod(self):
             probs.fill(maxValue)
             # compute new probabilities, choose min of all distances
             for j in range(0, i):
-                dists = similarityMeasurement(self.__obs, self.__clusterMeans[j], self.__compare)
+                dists = similarityMeasurement(self.__obs, self.__clusterMeans[j], self.__distance)
                 # collect minimum squared distances to cluster centroids
                 probs = np.minimum(probs, dists)
 
@@ -210,7 +211,7 @@ def __assignment(self):
             value = self.__obs[i]
 
             # compute squared distances to each mean
-            dists = similarityMeasurement(self.__clusterMeans, value, self.__compare)
+            dists = similarityMeasurement(self.__clusterMeans, value, self.__distance)
             # nearest cluster
             nearestID = np.argmin(dists)
 
@@ -347,12 +348,12 @@ def _plugin_initialize():
 
 # ----------------------------------------------------------------------------------------------------------------------
 
-def create(data, k, initMethod):
+def create(data, k, initMethod, distance):
   """
   by convention contain a factory called create returning the extension implementation
   :return:
   """
-  return KMeans(data, k, initMethod)
+  return KMeans(data, k, initMethod, distance)
 
 ########################################################################################################################
 
@@ -377,7 +378,7 @@ def create(data, k, initMethod):
 
     for i in range(10):
         s1 = timer()
-        kMeansPlus = KMeans(data, k, 'kmeans++', 10)
+        kMeansPlus = KMeans(data, k, 'kmeans++', 'sqeuclidean', 10)
         result1 = kMeansPlus.run()
         #print(result)
         e1 = timer()

clustering_service.py

@@ -48,7 +48,7 @@ def loadPlugin(pluginID, *args, **kwargs):
 
 ########################################################################################################################
 
-def runKMeans(data, k, initMethod):
+def runKMeans(data, k, initMethod, distance):
     """
     Runs the k-Means clustering algorithm given the loaded data set, the number of clusters k and the initialization
     method.
@@ -57,7 +57,7 @@ def runKMeans(data, k, initMethod):
     :param initMethod: number of clusters
     :return: result of k-means
     """
-    KMeans = loadPlugin('caleydo-clustering-kmeans', data, k, initMethod)
+    KMeans = loadPlugin('caleydo-clustering-kmeans', data, k, initMethod, distance)
     # and run the kmeans extension
     centroids, labels, clusterLabels = KMeans()
     # clusterLabels, clusterDists = KMeans.getDistsPerCentroid()
@@ -108,8 +108,8 @@ def runAffinityPropagation(data, damping, factor, preference, distance):
 
 ########################################################################################################################
 
-def runFuzzy(data, numClusters, m, threshold):
-    Fuzzy = loadPlugin('caleydo-clustering-fuzzy', data, numClusters, m, threshold)
+def runFuzzy(data, numClusters, m, threshold, distance):
+    Fuzzy = loadPlugin('caleydo-clustering-fuzzy', data, numClusters, m, threshold, distance)
 
     centroids, clusterLabels, partitionMatrix, maxProb = Fuzzy()
 

clustering_util.py

@@ -247,7 +247,7 @@ def similarityMeasurement(matrix, vector, method='euclidean'):
     spatialMethods = ['cityblock', 'chebyshev', 'canberra', 'correlation', 'hamming', 'mahalanobis',]
 
     if method in spatialMethods:
-        return cdist(matrix, np.atleast_2d(vector), method).flatten()
+        return np.nan_to_num(cdist(matrix, np.atleast_2d(vector), method).flatten())
 
     corrMethods = ['spearman', 'pearson', 'kendall']
 
@@ -379,7 +379,7 @@ def similarityMeasurementMatrix(matrix, method):
     spatialMethods = ['cityblock', 'chebyshev', 'canberra', 'correlation', 'hamming', 'mahalanobis']
 
     if method in spatialMethods:
-        return squareform(pdist(matrix, method))
+        return squareform(np.nan_to_num(pdist(matrix, method)))
 
     corrMethods = ['spearman', 'pearson', 'kendall']