def plot(self, xy = (0,1)):
"""
Outputs 2d embeded plot based on `pos`
:param xy: specifies the dimsntions of pos to be plotted.
:type xy: tuple, optional
"""
return plot_clusters(self.pos[:,[xy[0],xy[1]]], self.labels, clusters=self._cls)
break
else:
print "Replacing cluster %i with trajectory %i" % (clusterIndex, tjcIndex)
means[clusterIndex] = tjcs[tjcIndex]
meansIndex.append(tjcIndex)
ll_old = 0
i = 0
continue
else:
ll_old = ll_new
i += 1
print "Iteration Number: ", i
# This is to plot the results of the E-M algorithm.
fig2 = plt.figure()
myplt.plot_clusters(clusters, tjcs, "Results after Clustering", "600x500+850+0")
# Save means structre, which represents the clusters typical trayectories into
# a npy file.
timestr = time.strftime("%Y%m%d-%H%M%S")
filename=path.join("data/"+timestr+"-clusters")
np.save(filename, means)
# Let's plot the found clusters.
fig3 = plt.figure()
myplt.plot_time_model(means, "Clusters", "600x500+200+600")
# Plot cluster contribs
cluster_contribs = np.sum(clusters, 0)
fig4 = plt.figure()
wm = plt.get_current_fig_manager()
def main(_):
data_folder = FLAGS.data_folder
filename = FLAGS.filename
if FLAGS.task == 'metrics':
output = FLAGS.output
k = FLAGS.k
# find top-k, mid-k and bottom-k
scores, phrases = read_phrase_list(data_folder=data_folder,
filename=filename)
bottom = bottom_k(scores, phrases, k=k)
mid = mid_k(scores, phrases, k=k)
top = top_k(scores, phrases, k=k)
with open(data_folder + '/' + output, 'w') as out:
out.write('Bottom-k (' + str(k) + ')' + '\n')
for entry in bottom:
out.write(entry)
out.write('\n')
out.write('Mid-k (' + str(k) + ')' + '\n')
for entry in mid:
out.write(entry)
out.write('\n')
out.write('Top-k (' + str(k) + ')' + '\n')
for entry in top:
out.write(entry)
elif FLAGS.task == 'clustering':
clusters_folder = data_folder + '/' + FLAGS.cluster_folder
print('Current step: Reading data set', end='\r')
y_train, x_train = read_dataset(filename=filename,
data_folder=data_folder)
print('Current step: Performing clustering', end='\r')
task = PhraseClustering()
y_pred = task.run(x_train,
n_clusters=FLAGS.n_clusters,
distance=FLAGS.distance)
# create result files
print('Current step: Saving clustered phrases', end='\r')
if not os.path.exists(clusters_folder):
os.makedirs(clusters_folder)
cluster_ids = np.unique(y_pred)
files = {}
for cluster in cluster_ids:
out = open(clusters_folder + '/cluster' + str(cluster) + '.txt',
'w')
files[cluster] = out
for i, cluster in enumerate(y_pred.tolist()):
files[cluster].write(y_train[i] + '\n')
for _, file in files.items():
file.close()
print('Current step: Saving numpy files', end='\r')
save_cluster_numpy(x_train, y_pred, data_folder)
elif FLAGS.task == 'plot':
print('Current step: Reading data set', end='\r')
y_train, x_train = read_dataset(filename=filename,
data_folder=data_folder)
y_pred = load_cluster_numpy(data_folder)
print('Current step: Generating plots', end='\r')
plot_clusters(x_train, y_pred, FLAGS.n_clusters)
elif FLAGS.task == 'compare':
data = read_segmentation_metrics(filename=filename,
data_folder=data_folder)
plot_avg_phrases_curve(data)
plot_total_phrases_curve(data)
if i < iterations - 1:
c_index, c_score = em.worst_cluster(clusters)
t_index, t_score = em.worst_trajectory(clusters, c_index, c_score,
tjcIndex, tjcs, covariance)
# If a worst trajectory is not found.
if t_index == -1:
#break
means[c_index] = -1
else:
#print "Replacing cluster %i with trajectory %i" % ( c_index, t_index )
means[c_index] = tjcs[t_index]
tjcIndex.append( t_index )
# This is for plotting the results.
plt.title("Clustering, Iteration Number: %s" %(i+1))
myplt.plot_clusters(clusters, tjcs)
plt.pause(0.01)
# Save means structre, which represents the clusters typical trayectories into
# a npy file.
np.save("data/clusters.npy", means)
# Let's plot the found clusters.
fig3 = plt.figure()
wm = plt.get_current_fig_manager()
wm.window.wm_geometry("600x500+350+200")
fi3 = plt.title("Clusters")
plt.grid()
plt.xticks(np.arange(0, 16, 1.0))
plt.yticks(np.arange(0, 12, 1.0))
plt.xlabel('X (mts)')
#####################################################################
#####################################################################
best_clustering = get_best_clustering(working_directory)
best_clustering = Clustering.from_dic(best_clustering["clustering"])
scores = []
for a_cluster in best_clustering.clusters:
scores.append((score_cluster(a_cluster, m_handler.distance_matrix, all_metrics), a_cluster))
# Remember: first metric is spawning point, second metric must be energy
most_negative = find_most_negative_cluster(scores)
most_negative_cluster = most_negative[1]
#######################################################################################################################
# Plot clusters
#######################################################################################################################
plot_clusters(os.path.join(base_dir, "clusters.svg"), all_metrics, scores, scores.index(most_negative))
#######################################################################################################################
# Store the elements of the most negative cluster
#######################################################################################################################
# This works because we have only one traj. Merging before should be mandatory.
trajectory_path = os.path.join(os.getcwd(), traj_pdb)
in_handler = open(trajectory_path,"r")
most_negative_path = os.path.join(base_dir,"most_negative_cluster.pdb")
out_handler = open(most_negative_path,"w")
extract_frames_from_trajectory_sequentially(file_handler_in = in_handler,
number_of_frames = get_number_of_frames(trajectory_path),
file_handler_out = out_handler,
frames_to_save = most_negative_cluster.all_elements,
keep_header = True,
write_frame_number_instead_of_correlative_model_number = True)
