def diffusion_payoff(s1, s2):
iterations = 5000
r = np.zeros(iterations)
for i in range(iterations):
g = tools.load_graph(data_set)
tools.diffuse(g, s1, s2)
r[i] = len(g.graph["1"]) - len(g.graph["2"])
return r
def competition(data_set, nodes):
iterations = 1000
r = np.zeros(iterations)
for i in range(iterations):
a1 = ca.action_v_each(k, 2, nodes)
a2 = ca.action_v_each(k, 1, nodes)
s1, s2 = get_seeds(a1, a2)
g = tools.load_graph(data_set)
tools.diffuse(g, s1, s2)
r[i] = len(g.graph["1"]) - len(g.graph["2"])
return r
def competition(a1, a2):
s1, s2 = get_seeds(a1, a2)
g = tools.load_graph(data_set)
tools.diffuse(g, s1, s2)
return len(g.graph["1"]) - len(g.graph["2"])
import numpy as np
import networkx as nx
import tools
import TIM
import time
start_time = time.time()
k = 50
data_set = "wiki_vote"
result_folder = "../results/graph/" + data_set + "/"
g = tools.load_graph(data_set)
theta = TIM.calculate_theta(g, k)
g = tools.load_graph(data_set)
cover_graph = TIM.create_rr_sets_graph(g, theta)
end_time = time.time()
nx.write_gpickle(cover_graph, result_folder + "rr_sets_graph")
file = open(result_folder + "result.txt", "w")
file.write("data set: " + data_set + "\n")
file.write("k: " + str(k) + "\n")
file.write("theta: " + str(theta) + "\n")
file.write("run time: " + str(end_time - start_time) + "\n")
file.close()
def label_images(snapcat_json):
model_file = settings.graph['graph']
label_file = settings.graph['labels']
input_height = settings.graph['input_height']
input_width = settings.graph['input_width']
input_mean = settings.graph['input_mean']
input_std = settings.graph['input_std']
input_layer = settings.graph['input_layer']
output_layer = settings.graph['output_layer']
graph = tools.load_graph(model_file)
labels = tools.load_labels(label_file)
# Create Full path
# todo, should this be outside the loop, one TF session and then parse all the images?
with tf.Session(graph=graph) as sess:
for image in snapcat_json.json_data:
file = snapcat_json.json_data[image]["path"]
if not os.path.isfile(file):
print("***************ERROR - File doesn't exist:", file)
continue
img = cv2.imread(file)
# resize area of interest for classification
#TODO - modify for multiple areas of interest
try:
area_of_interest = snapcat_json.json_data[image][
"area_of_interest"]
x1 = area_of_interest[0]
x2 = area_of_interest[1]
y1 = area_of_interest[2]
y2 = area_of_interest[3]
img = img[y1:y2, x1:x2, :]
except:
pass
resized_image = cv2.resize(img, (input_width, input_height))
t = (np.float32(resized_image) - input_mean) / input_std
input_name = "import/" + input_layer
output_name = "import/" + output_layer
input_operation = graph.get_operation_by_name(input_name)
output_operation = graph.get_operation_by_name(output_name)
# todo, suspect this is what's printing a lot of messages. Attempt to consolidate calls to this function
results = sess.run(output_operation.outputs[0],
{input_operation.outputs[0]: [t]})
results = np.squeeze(results)
# get classification
top_k = results.argsort()[-5:][::-1]
for i in top_k:
# if confidence level is below certain value, put in "unsure" folder
print("%s: %f" % (labels[i], results[i]))
if results[i] >= settings.sort_image[
'cat_confidence_threshold'] and labels[i] == 'cats':
print("cat")
snapcat_json.update(
image, "classifier_label", "cat"
) #TODO - classifier_label will be associated with an area of interest
elif results[i] >= settings.sort_image[
'not_cat_confidence_threshold'] and labels[
i] == 'not cats':
print("not cat")
snapcat_json.update(
image, "classifier_label", "not_cat"
) #TODO - classifier_label will be associated with an area of interest
else:
print("unsure")
snapcat_json.update(
image, "classifier_label", "unsure"
) #TODO - classifier_label will be associated with an area of interest
break
snapcat_json.save()
def label_images(snapcat_json):
model_file = settings.graph['graph']
label_file = settings.graph['labels']
input_height = settings.graph['input_height']
input_width = settings.graph['input_width']
input_mean = settings.graph['input_mean']
input_std = settings.graph['input_std']
input_layer = settings.graph['input_layer']
output_layer = settings.graph['output_layer']
graph = tools.load_graph(model_file)
labels = tools.load_labels(label_file)
# Create Full path
# todo, should this be outside the loop, one TF session and then parse all the images?
with tf.Session(graph=graph) as sess:
for image in snapcat_json.json_data:
file = snapcat_json.json_data[image]["path"]
if not os.path.isfile(file):
print("***************ERROR - File doesn't exist:", file)
continue
# resize area of interest for classification
area_of_interest = snapcat_json.json_data[image][
"area_of_interest"]
x1 = area_of_interest[0]
x2 = area_of_interest[1]
y1 = area_of_interest[2]
y2 = area_of_interest[3]
img = cv2.imread(file)
img = img[y1:y2, x1:x2, :]
resized_image = cv2.resize(img, (224, 224))
resized_image = img_as_ubyte(resized_image)
# TODO - we need to load the area of interest and resize the image here rather than read the file:
tempfile = "tempfile.jpg"
cv2.imwrite(tempfile, resized_image)
cv2.destroyAllWindows()
# TODO - this loads a tensorflow session resulting in a bunch of debug output. Probably will be more efficient to leave the session open if possible
t = tools.read_tensor_from_image_file(tempfile,
input_height=input_height,
input_width=input_width,
input_mean=input_mean,
input_std=input_std)
input_name = "import/" + input_layer
output_name = "import/" + output_layer
input_operation = graph.get_operation_by_name(input_name)
output_operation = graph.get_operation_by_name(output_name)
# todo, suspect this is what's printing a lot of messages. Attempt to consolidate calls to this function
results = sess.run(output_operation.outputs[0],
{input_operation.outputs[0]: t})
results = np.squeeze(results)
# get classification
top_k = results.argsort()[-5:][::-1]
for i in top_k:
# if confidence level is below certain value, put in "unsure" folder
print("%s: %f" % (labels[i], results[i]))
if results[i] >= settings.sort_image[
'cat_confidence_threshold'] and labels[i] == 'cats':
print("cat")
snapcat_json.update(image, "classifier_label", "cat")
elif results[i] >= settings.sort_image[
'not_cat_confidence_threshold'] and labels[
i] == 'not cats':
print("not cat")
snapcat_json.update(image, "classifier_label", "not_cat")
else:
print("unsure")
snapcat_json.update(image, "classifier_label", "unsure")
break
snapcat_json.save()
