def template_clustering(file, map_size, radius, sync_order = 0.999, show_dyn = False, show_layer1 = False, show_layer2 = False, show_clusters = True):
# Read sample
sample = read_sample(file);
# Create network
network = syncsom(sample, map_size[0], map_size[1], radius);
# Run processing
(ticks, (dyn_time, dyn_phase)) = timedcall(network.process, show_dyn, sync_order);
print("Sample: ", file, "\t\tExecution time: ", ticks, "\n");
# Show dynamic of the last layer.
if (show_dyn == True):
draw_dynamics(dyn_time, dyn_phase, x_title = "Time", y_title = "Phase", y_lim = [0, 3.14]);
if (show_clusters == True):
clusters = network.get_som_clusters();
visualizer = cluster_visualizer();
visualizer.append_clusters(clusters, network.som_layer.weights);
visualizer.show();
# Show network stuff.
if (show_layer1 == True):
network.show_som_layer();
if (show_layer2 == True):
network.show_sync_layer();
if (show_clusters == True):
clusters = network.get_clusters();
visualizer = cluster_visualizer();
visualizer.append_clusters(clusters, sample);
visualizer.show();
def templateLengthProcessData(self, file, som_map_size, avg_num_conn, eps, expected_cluster_length):
result_testing = False;
# If phases crosses each other because of random part of the network then we should try again.
for attempt in range(0, 5, 1):
sample = read_sample(file);
network = syncsom(sample, som_map_size[0], som_map_size[1]);
network.process(avg_num_conn, collect_dynamic = False, order = eps);
clusters = network.get_clusters();
obtained_cluster_sizes = [len(cluster) for cluster in clusters];
if (len(sample) != sum(obtained_cluster_sizes)):
continue;
obtained_cluster_sizes.sort();
expected_cluster_length.sort();
#print(obtained_cluster_sizes, expected_cluster_length);
if (obtained_cluster_sizes != expected_cluster_length):
continue;
# Unit-test is passed
result_testing = True;
break;
assert result_testing;
def templateLengthProcessData(self, file, som_map_size, radius, eps,
expected_cluster_length):
result_testing = False
# If phases crosses each other because of random part of the network then we should try again.
for _ in range(0, 5, 1):
sample = read_sample(file)
network = syncsom(sample, som_map_size[0], som_map_size[1], radius)
network.process(collect_dynamic=False, order=eps)
clusters = network.get_clusters()
obtained_cluster_sizes = [len(cluster) for cluster in clusters]
if (len(sample) != sum(obtained_cluster_sizes)):
continue
obtained_cluster_sizes.sort()
expected_cluster_length.sort()
#print(obtained_cluster_sizes, expected_cluster_length);
if (obtained_cluster_sizes != expected_cluster_length):
continue
# Unit-test is passed
result_testing = True
break
assert result_testing
def template_clustering(file, map_size, trust_order, sync_order = 0.999, show_dyn = False, show_layer1 = False, show_layer2 = False, show_clusters = True):
# Read sample
sample = read_sample(file);
# Create network
network = syncsom(sample, map_size[0], map_size[1]);
# Run processing
(ticks, (dyn_time, dyn_phase)) = timedcall(network.process, trust_order, show_dyn, sync_order);
print("Sample: ", file, "\t\tExecution time: ", ticks, "\n");
# Show dynamic of the last layer.
if (show_dyn == True):
draw_dynamics(dyn_time, dyn_phase, x_title = "Time", y_title = "Phase", y_lim = [0, 2 * 3.14]);
if (show_clusters == True):
clusters = network.get_som_clusters();
draw_clusters(network.som_layer.weights, clusters);
# Show network stuff.
if (show_layer1 == True):
network.show_som_layer();
if (show_layer2 == True):
network.show_sync_layer();
if (show_clusters == True):
clusters = network.get_clusters();
draw_clusters(sample, clusters);
def templateLengthProcessData(self, file, som_map_size, avg_num_conn, eps, expected_cluster_length):
result_testing = False;
# If phases crosses each other because of random part of the network then we should try again.
for attempt in range(0, 3, 1):
sample = read_sample(file);
network = syncsom(sample, som_map_size[0], som_map_size[1]);
network.process(avg_num_conn, collect_dynamic = False, order = eps);
clusters = network.get_clusters();
obtained_cluster_sizes = [len(cluster) for cluster in clusters];
if (len(sample) != sum(obtained_cluster_sizes)):
continue;
obtained_cluster_sizes.sort();
expected_cluster_length.sort();
#print(obtained_cluster_sizes, expected_cluster_length);
if (obtained_cluster_sizes != expected_cluster_length):
continue;
# Unit-test is passed
result_testing = True;
break;
assert result_testing;
def experiment_execution_one_cluster_dependence(layer_first_size, radius,
order):
print("Experiment: map size =", layer_first_size[0] * layer_first_size[1],
"radius =", radius, "order =", order)
cluster_sizes = [
10, 20, 30, 40, 50, 60, 70, 80, 90, 100, 110, 120, 130, 140, 150
]
for cluster_size in cluster_sizes:
# generate data sets
dataset = []
dataset += [[random(), random()] for _ in range(cluster_size)]
general_value = 0.0
amount_attempt = 5
for _ in range(amount_attempt):
network = syncsom(dataset, layer_first_size[0],
layer_first_size[1], radius)
(ticks, (dyn_time, dyn_phase)) = timedcall(network.process, False,
order)
general_value += ticks
print("Sample: ", cluster_size, "\t\tExecution time: ",
general_value / float(amount_attempt))
print("\n")
def testShowLayersProcessing(self):
sample = read_sample(SIMPLE_SAMPLES.SAMPLE_SIMPLE1)
network = syncsom(sample, 4, 4, 1.0)
network.process(collect_dynamic=False, order=0.99)
network.show_som_layer()
network.show_sync_layer()
def testShowLayersProcessing(self):
sample = read_sample(SIMPLE_SAMPLES.SAMPLE_SIMPLE1);
network = syncsom(sample, 4, 4, 1.0);
network.process(collect_dynamic = False, order = 0.99);
network.show_som_layer();
network.show_sync_layer();
def template_segmentation_image(source, map_som_size = [5, 5], radius = 128.0, sync_order = 0.998, show_dyn = False, show_som_map = False):
data = read_image(source);
network = syncsom(data, map_som_size[0], map_som_size[1], 1.0);
(ticks, (dyn_time, dyn_phase)) = timedcall(network.process, show_dyn, sync_order);
print("Sample: ", source, "\t\tExecution time: ", ticks, "\t\tWinners: ", network.som_layer.get_winner_number(), "\n");
if (show_dyn is True):
draw_dynamics(dyn_time, dyn_phase);
clusters = network.get_clusters();
draw_image_mask_segments(source, clusters);
def template_segmentation_image(source, map_som_size = [5, 5], average_neighbors = 5, sync_order = 0.998, show_dyn = False, show_som_map = False):
data = read_image(source);
network = syncsom(data, map_som_size[0], map_som_size[1]);
(ticks, (dyn_time, dyn_phase)) = timedcall(network.process, average_neighbors, show_dyn, sync_order);
print("Sample: ", source, "\t\tExecution time: ", ticks, "\t\tWinners: ", network.som_layer.get_winner_number(), "\n");
if (show_dyn is True):
draw_dynamics(dyn_time, dyn_phase);
clusters = network.get_clusters();
draw_image_mask_segments(source, clusters);
def templateLengthSomCluster(self, file, som_map_size, avg_num_conn, eps):
sample = read_sample(file);
network = syncsom(sample, som_map_size[0], som_map_size[1]);
network.process(avg_num_conn, collect_dynamic = False, order = eps);
# Check unique
som_clusters = network.get_som_clusters();
indexes = set();
for som_cluster in som_clusters:
for index in som_cluster:
assert (index in indexes) is False;
indexes.add(index);
def templateLengthSomCluster(self, file, som_map_size, avg_num_conn, eps):
sample = read_sample(file)
network = syncsom(sample, som_map_size[0], som_map_size[1])
network.process(avg_num_conn, collect_dynamic=False, order=eps)
# Check unique
som_clusters = network.get_som_clusters()
indexes = set()
for som_cluster in som_clusters:
for index in som_cluster:
assert (index in indexes) is False
indexes.add(index)
def templateLengthSomCluster(self, file, som_map_size, radius, eps):
sample = read_sample(file);
network = syncsom(sample, som_map_size[0], som_map_size[1], radius);
network.process(collect_dynamic = False, order = eps);
# Check unique
som_clusters = network.get_som_clusters();
indexes = set();
for som_cluster in som_clusters:
for index in som_cluster:
assert (index in indexes) is False;
indexes.add(index);
def experiment_execution_one_cluster_dependence(layer_first_size, radius, order):
print("Experiment: map size =", layer_first_size[0] * layer_first_size[1], "radius =", radius, "order =", order);
cluster_sizes = [10, 20, 30, 40, 50, 60, 70, 80, 90, 100, 110, 120, 130, 140, 150];
for cluster_size in cluster_sizes:
# generate data sets
dataset = [];
dataset += [ [random(), random()] for _ in range(cluster_size) ];
general_value = 0.0;
amount_attempt = 5;
for _ in range(amount_attempt):
network = syncsom(dataset, layer_first_size[0], layer_first_size[1], radius);
(ticks, (dyn_time, dyn_phase)) = timedcall(network.process, False, order);
general_value += ticks;
print("Sample: ", cluster_size, "\t\tExecution time: ", general_value / float(amount_attempt));
print("\n");
def template_clustering(file,
map_size,
trust_order,
sync_order=0.999,
show_dyn=False,
show_layer1=False,
show_layer2=False,
show_clusters=True):
# Read sample
sample = read_sample(file)
# Create network
network = syncsom(sample, map_size[0], map_size[1])
# Run processing
(ticks, (dyn_time, dyn_phase)) = timedcall(network.process, trust_order,
show_dyn, sync_order)
print("Sample: ", file, "\t\tExecution time: ", ticks, "\n")
# Show dynamic of the last layer.
if (show_dyn == True):
draw_dynamics(dyn_time,
dyn_phase,
x_title="Time",
y_title="Phase",
y_lim=[0, 2 * 3.14])
if (show_clusters == True):
clusters = network.get_som_clusters()
draw_clusters(network.som_layer.weights, clusters)
# Show network stuff.
if (show_layer1 == True):
network.show_som_layer()
if (show_layer2 == True):
network.show_sync_layer()
if (show_clusters == True):
clusters = network.get_clusters()
draw_clusters(sample, clusters)
def process_syncsom(sample):
instance = syncsom(sample, 1, NUMBER_CLUSTERS)
(ticks, _) = timedcall(instance.process, 0, False, 0.998)
return ticks
from pyclustering.nnet.cnn import cnn_network, cnn_visualizer
dataset = get_train(read_articles('dr_medi'))
# labels = dataset.target
# true_k = np.unique(dataset).shape[0]
true_k = np.unique(dataset[:8]).shape[0]
vectorizer = TfidfVectorizer()
# X = vectorizer.fit_transform(dataset)
X = vectorizer.fit_transform(get_train(read_articles('dr_medi')))
print(X.shape)
svd = TruncatedSVD(true_k)
lsa = make_pipeline(svd, Normalizer(copy=False))
X = lsa.fit_transform(X)
network = syncsom(X, 2, 2, 0.9)
(dyn_time, dyn_phase) = network.process(True, 0.999)
# network.show_sync_layer();
clusters = network.get_clusters()
visualizer = cluster_visualizer()
visualizer.append_clusters(clusters, X)
visualizer.show()
# kmeans_instance = kmeans(X, [[ 0.93036182, -0.20536988, -0.17834053, 0.03102577, 0.01767965, -0.06377806, 0.22167392, -0.077218]]);
#
# kmeans_instance.process();
# clusters = kmeans_instance.get_centers();
# print(clusters)
#
# network = som(5, 5, X, 100);
#
def process_syncsom(sample):
instance = syncsom(sample, 1, NUMBER_CLUSTERS)
(ticks, _) = timedcall(instance.process, 0, False, 0.998)
return ticks
