matrix = matrix_pce
#hierarchical clustering part starts here
linkage = dendro.compute_linkage(matrix)
#methods = ['single', 'complete', 'average', 'weighted', 'centroid', 'median', 'ward']
#for method in methods:
# linkage = sch.linkage(matrix, method=method)
threshold = 0.7*linkage[:,2].max() # default threshold used in sch.dendogram is 0.7*linkage[:,2].max()
#dendrogram = dendro.compute_dendrogram(linkage)
dendrogram = dendro.plot_dendrogram_and_matrix(linkage, matrix, color_threshold=threshold)
#compute flat clustering in the exact same way as sch.dendogram colors the clusters
cluster = numpy.array(sch.fcluster(linkage, threshold, criterion='distance'), dtype=numpy.int)
numpy.set_printoptions(threshold=numpy.nan) # make numpy print the full array
print "flat clustering:\n", cluster
#get the actual clustering
filelist = numpy.loadtxt(directory + "/filelist.txt", dtype=numpy.string_)
true_clustering = ["_".join(s.split("_")[:-1]) for s in filelist]
true_clusters = sorted(set(true_clustering))
true_labels = numpy.array([true_clusters.index(id) for id in true_clustering], dtype=numpy.int)
print "true clustering:\n", true_labels
def get_ground_truth():
filelist = numpy.loadtxt(directory + "/filelist.txt", dtype=numpy.string_)
numfiles = filelist.size
matrix_ans = numpy.zeros([numfiles,numfiles], dtype=numpy.float)
for i in range(numfiles):
for j in range(numfiles):
cam1 = "_".join(filelist[i].split("_")[:-1])
cam2 = "_".join(filelist[j].split("_")[:-1])
if cam1 == cam2:
matrix_ans[i][j] = 1.0
else:
matrix_ans[i][j] = 100.0
if i == j:
matrix_ans[i][j] = 0.0
return matrix_ans
#if __name__ == "__main__":
# import sys
# if len(sys.argv) != 2:
# print("Usage: ./camera_identification <name-of-dataset>")
# exit()
# import os
# if not os.path.isdir(directory + "/" + sys.argv[1]):
# print("incorrect dataset name, cannot find " + directory + "/" + sys.argv[1])
# exit()
dataset ="pentax" #sys.argv[1]
directory = directory + "/" + dataset
#load the distance matrixes from files
matrix_pce = numpy.fromfile(directory + "/matrix-" + dataset + "-pce.dat", dtype='>d')
matrix_pce0 = numpy.fromfile(directory + "/matrix-" + dataset + "-pce0.dat", dtype='>d')
matrix_ncc = numpy.fromfile(directory + "/matrix-" + dataset + "-ncc.dat", dtype='>d')
matrix_pce, matrix_ncc = map_ncc_scores_to_pce_domain(matrix_pce, matrix_ncc)
matrix_ncc = convert_similarity_to_distance(matrix_ncc)
matrix_pce = convert_similarity_to_distance(matrix_pce)
matrix_pce0 = convert_similarity_to_distance(matrix_pce0)
matrix_ans = get_ground_truth()
plot_distance_matrices(matrix_ncc, matrix_pce, matrix_pce0, matrix_ans)
#set metric to use for clustering
matrix = matrix_pce
#hierarchical clustering part starts here
linkage = dendro.compute_linkage(matrix)
#methods = ['single', 'complete', 'average', 'weighted', 'centroid', 'median', 'ward']
#for method in methods:
# linkage = sch.linkage(matrix, method=method)
threshold = 0.7*linkage[:,2].max() # default threshold used in sch.dendogram is 0.7*linkage[:,2].max()
#dendrogram = dendro.compute_dendrogram(linkage)
dendrogram = dendro.plot_dendrogram_and_matrix(linkage, matrix, color_threshold=threshold)
#compute flat clustering in the exact same way as sch.dendogram colors the clusters
cluster = numpy.array(sch.fcluster(linkage, threshold, criterion='distance'), dtype=numpy.int)
numpy.set_printoptions(threshold=numpy.nan) # make numpy print the full array
print("flat clustering:\n", cluster)
#get the actual clustering
filelist = numpy.loadtxt(directory + "/filelist.txt", dtype=numpy.string_)
true_clustering = ["_".join(s.split("_")[:-1]) for s in filelist]
true_clusters = sorted(set(true_clustering))
true_labels = numpy.array([true_clusters.index(id) for id in true_clustering], dtype=numpy.int)
print("true clustering:\n", true_labels)
print_metrics(true_labels, cluster)