def initializeMainGui():
g.init('gui/MotilityTracking.ui')
g.m.trackView = Window(np.zeros((3, 3, 3)))
g.m.histogram = Histogram(title='Mean Single Lag Distance Histogram', labels={'left': 'Count', 'bottom': 'Mean SLD Per Track (Pixels)'})
g.m.trackPlot = TrackPlot()
g.m.trackPlot.tracksChanged.connect(update_plots)
g.m.trackView.imageview.addItem(g.m.trackPlot)
g.m.trackView.imported = pg.ScatterPlotItem()
g.m.trackView.imageview.addItem(g.m.trackView.imported)
g.m.actionImportBin.triggered.connect(lambda : open_file_gui(lambda f: import_mat(open_bin(f)), prompt='Import .bin file of tracks', filetypes='*.bin'))
g.m.actionImportBackground.triggered.connect(lambda : open_file_gui(set_background_image, prompt='Select tif file as background image', filetypes='*.tif *.tiff *.stk'))
g.m.actionImportCoordinates.triggered.connect(lambda : open_file_gui(import_coords, prompt='Import coordinates from txt file', filetypes='*.txt'))
g.m.actionSimulateDistances.triggered.connect(lambda : save_file_gui(simulate_distances, prompt = 'Save simulated distances', filetypes='*.txt'))
g.m.actionExportMSD.triggered.connect(lambda : save_file_gui(exportMSD, prompt='Export Mean Squared Displacement Values', filetypes='*.txt'))
g.m.actionExportHistogram.triggered.connect(lambda : save_file_gui(g.m.histogram.export, prompt='Export Histogram Values', filetypes='*.txt'))
g.m.actionExportTrackLengths.triggered.connect(lambda : save_file_gui(export_track_lengths, prompt='Export Track Lengths', filetypes='*.txt'))
g.m.actionExportOutlined.triggered.connect(g.m.trackPlot.export_gui)
g.m.actionExportDistances.triggered.connect(lambda : save_file_gui(export_real_distances, prompt='Export Distances', filetypes='*.txt'))
g.m.MSLDMinSpin.setOpts(value=0, decimals=2, maximum=1000)
g.m.MSLDMaxSpin.setOpts(value=100, decimals=2, maximum=1000)
g.m.neighborsSpin.setOpts(value=0, maximum=100, int=True, step=1)
g.m.neighborDistanceSpin.setOpts(value=1, decimals=2, maximum=100)
g.m.minLengthSpin.setOpts(value=4, maximum=1000, int=True, step=1)
g.m.maxLengthSpin.setOpts(value=20, maximum=1000, int=True, step=1)
g.m.MSLDGroupBox.toggled.connect(g.m.trackPlot.filter)
g.m.additionalGroupBox.toggled.connect(g.m.trackPlot.filter)
g.m.MSLDMaxSpin.sigValueChanged.connect(g.m.trackPlot.filter)
g.m.MSLDMinSpin.sigValueChanged.connect(g.m.trackPlot.filter)
g.m.neighborDistanceSpin.sigValueChanged.connect(g.m.trackPlot.filter)
g.m.neighborsSpin.sigValueChanged.connect(g.m.trackPlot.filter)
g.m.minLengthSpin.sigValueChanged.connect(g.m.trackPlot.filter)
g.m.maxLengthSpin.sigValueChanged.connect(g.m.trackPlot.filter)
g.m.hideBackgroundCheck.toggled.connect(lambda v: g.m.trackView.imageview.getImageItem().setVisible(not v))
g.m.ignoreOutsideCheck.toggled.connect(g.m.trackPlot.filter)
g.m.plotMeansCheck.toggled.connect(g.m.trackPlot.means.setVisible)
g.m.plotTracksCheck.toggled.connect(g.m.trackPlot.tracks.setVisible)
g.m.viewTab.layout().insertWidget(0, g.m.trackView)
g.m.MSDWidget = pg.PlotWidget(title='Mean Squared Displacement Per Lag', labels={'left': 'Mean Squared Disance (p^2)', 'bottom': 'Lag Count'})
g.m.analysisTab.layout().addWidget(g.m.MSDWidget)
g.m.analysisTab.layout().addWidget(g.m.histogram)
g.m.CDFWidget = CDFWidget()#pg.PlotWidget(title = 'Cumulative Distribution Function', labels={'left': 'Cumulative Probability', 'bottom': 'Single Lag Displacement Squared'})
#g.m.CDFPlot = pg.PlotCurveItem()
g.m.cdfTab.layout().addWidget(g.m.CDFWidget)
g.m.installEventFilter(mainWindowEventEater)
g.m.setWindowTitle('Motility Tracking')
g.m.show()
# Main class to start the game import pygame import random from dungeonGen import dMap from items import Block from player_monster import PlayerMonster import global_vars global_vars.init() PLAYER_SPEED = 3 # Pygame start pygame.init() screen_width = 800 screen_height = 600 screen = pygame.display.set_mode([screen_width, screen_height]) # This is a list of every sprite in the game all_sprites_list = pygame.sprite.Group() # items we can't pass through all_items_list = pygame.sprite.Group() # list of all the blocks in the game (for the dungeon) block_list = pygame.sprite.Group() actor_list = pygame.sprite.Group()
def main():
"""Main function"""
# construct the argument parse and parse the arguments
# init global variables
g.init()
from optparse import OptionParser
# Configure command line options
parser = OptionParser()
parser.add_option("-i", "--input_file", dest="input_file",
help="Input video/image file")
parser.add_option("-o", "--output_file", dest="output_file",
help="Output (destination) video/image file")
parser.add_option("-m", "--method", dest="method", default='m1',
help="m1 -> for method 1 - m2 -> for method 2 ...")
parser.add_option("-v", "--video", dest="video_mode", default='cv2',
help="video_mode choice cv2 or movepi")
parser.add_option("-P", "--preview_only",
action="store_true", default=False,
help="preview annoted video/image")
parser.add_option("-I", "--image_only",
action="store_true", dest="image_only", default=False,
help="Annotate image (defaults to annotating video)")
parser.add_option("-T", "--trackbars",
dest="TRACKBARS", action="store_true", default=False,
help="Enable debug trackbars.")
# Get and parse command line options
options, args = parser.parse_args()
print ('options:',options)
print ('args:',args)
input_file = options.input_file
output_file = options.output_file
image_only = options.image_only
preview_only = options.preview_only
video_mode = options.video_mode
method = options.method
g.trackbar_enabled = options.TRACKBARS
if preview_only:
if video_mode == 'movepi':
if method == 'm1':
annotate_video_preview_1(input_file)
if method == 'm2':
annotate_video_preview_2(input_file)
if method == 'm3':
annotate_video_preview_3(input_file)
if method == 'm4':
annotate_video_preview_4(input_file)
if method == 'm5':
remove_shadows_video_preview(input_file)
if video_mode == 'cv2':
if method == 'm1':
lane_detector_method_1(input_file)
if method == 'm2':
lane_detector_method_2(input_file)
if method == 'm3':
lane_detector_method_3(input_file)
if method == 'm4':
lane_detector_method_4(input_file)
if method == 'm5':
remove_shadows(input_file)
sys.exit()
if image_only:
if method == 'm1':
annotate_image_1(input_file, output_file)
if method == 'm2':
annotate_image_2(input_file, output_file)
if method == 'm3':
annotate_image_3(input_file, output_file)
if method == 'm4':
annotate_image_4(input_file, output_file)
if method == 'm5':
remove_shadows_image(input_file, output_file)
else:
if method == 'm1':
annotate_video_1(input_file, output_file)
if method == 'm2':
annotate_video_2(input_file, output_file)
if method == 'm3':
annotate_video_3(input_file, output_file)
if method == 'm4':
annotate_video_4(input_file, output_file)
if method == 'm5':
remove_shadows_video(input_file, output_file)
if args.mal:
return_dict['mal_suc_count'] = 0
for t in range(args.T):
if not os.path.exists(gv.dir_name + 'global_weights_t%s.npy' % t):
print('No directory found for iteration %s' % t)
break
if args.mal:
# p_eval = Process(target=eval_func, args=(
# X_test, Y_test_uncat, t, return_dict, mal_data_X, mal_data_Y))
eval_func(X_test, Y_test_uncat, t, return_dict, mal_data_X,
mal_data_Y)
else:
# p_eval = Process(target=eval_func, args=(
# X_test, Y_test_uncat, t, return_dict))
eval_func(X_test, Y_test_uncat, t, return_dict)
# p_eval.start()
# p_eval.join()
if args.mal:
print('Malicious agent succeeded in %s of %s iterations' %
(return_dict['mal_suc_count'], (t - 1) * args.mal_num))
if __name__ == "__main__":
gv.init()
tf.set_random_seed(777)
np.random.seed(777)
args = gv.args
main()
def agent(i, X_shard, Y_shard, t, gpu_id, return_dict, X_test, Y_test, lr=None):
tf.keras.backend.set_learning_phase(1)
args = gv.init()
if lr is None:
lr = args.eta
print('Agent %s on GPU %s' % (i,gpu_id))
# set environment
os.environ["CUDA_DEVICE_ORDER"] = "PCI_BUS_ID"
os.environ["CUDA_VISIBLE_DEVICES"] = str(gpu_id)
shared_weights = np.load(gv.dir_name + 'global_weights_t%s.npy' % t, allow_pickle=True)
shard_size = len(X_shard)
if 'theta{}'.format(gv.mal_agent_index) in return_dict.keys():
pre_theta = return_dict['theta{}'.format(gv.mal_agent_index)]
else:
pre_theta = None
# if i == 0:
# # eval_success, eval_loss = eval_minimal(X_test,Y_test,x, y, sess, prediction, loss)
# eval_success, eval_loss = eval_minimal(X_test,Y_test,shared_weights)
# print('Global success at time {}: {}, loss {}'.format(t,eval_success,eval_loss))
if args.steps is not None:
num_steps = args.steps
else:
num_steps = int(args.E * shard_size / args.B)
# with tf.device('/gpu:'+str(gpu_id)):
if args.dataset == 'census':
x = tf.placeholder(shape=(None,
gv.DATA_DIM), dtype=tf.float32)
# y = tf.placeholder(dtype=tf.float32)
y = tf.placeholder(dtype=tf.int64)
else:
x = tf.placeholder(shape=(None,
gv.IMAGE_ROWS,
gv.IMAGE_COLS,
gv.NUM_CHANNELS), dtype=tf.float32)
y = tf.placeholder(dtype=tf.int64)
if 'MNIST' in args.dataset:
agent_model = model_mnist(type=args.model_num)
elif args.dataset == 'census':
agent_model = census_model_1()
elif args.dataset == 'CIFAR-10':
agent_model = cifar10_model()
else:
return
logits = agent_model(x)
if args.dataset == 'census':
# loss = tf.nn.sigmoid_cross_entropy_with_logits(
# labels=y, logits=logits)
loss = tf.reduce_mean(tf.nn.sparse_softmax_cross_entropy_with_logits(
labels=y, logits=logits))
else:
loss = tf.reduce_mean(tf.nn.sparse_softmax_cross_entropy_with_logits(
labels=y, logits=logits))
prediction = tf.nn.softmax(logits)
if args.optimizer == 'adam':
optimizer = tf.train.AdamOptimizer(
learning_rate=lr).minimize(loss)
elif args.optimizer == 'sgd':
optimizer = tf.train.GradientDescentOptimizer(
learning_rate=lr).minimize(loss)
if args.k > 1:
config = tf.ConfigProto(gpu_options=gv.gpu_options)
# config.gpu_options.allow_growth = True
sess = tf.Session(config=config)
elif args.k == 1:
sess = tf.Session()
else:
return
tf.compat.v1.keras.backend.set_session(sess)
sess.run(tf.global_variables_initializer())
if pre_theta is not None:
theta = pre_theta - gv.moving_rate * (pre_theta - shared_weights)
else:
theta = shared_weights
agent_model.set_weights(theta)
# print('loaded shared weights')
start_offset = 0
if args.steps is not None:
start_offset = (t * args.B * args.steps) % (shard_size - args.B)
for step in range(num_steps):
offset = (start_offset + step * args.B) % (shard_size - args.B)
X_batch = X_shard[offset: (offset + args.B)]
Y_batch = Y_shard[offset: (offset + args.B)]
Y_batch_uncat = np.argmax(Y_batch, axis=1)
_, loss_val = sess.run([optimizer, loss], feed_dict={x: X_batch, y: Y_batch_uncat})
if step % 1000 == 0:
print('Agent %s, Step %s, Loss %s, offset %s' % (i, step, loss_val, offset))
# local_weights = agent_model.get_weights()
# eval_success, eval_loss = eval_minimal(X_test,Y_test,x, y, sess, prediction, loss)
# print('Agent {}, Step {}: success {}, loss {}'.format(i,step,eval_success,eval_loss))
local_weights = agent_model.get_weights()
local_delta = local_weights - shared_weights
# eval_success, eval_loss = eval_minimal(X_test,Y_test,x, y, sess, prediction, loss)
eval_success, eval_loss = eval_minimal(X_test, Y_test, local_weights)
print('Agent {}: success {}, loss {}'.format(i, eval_success, eval_loss))
return_dict[str(i)] = np.array(local_delta)
return_dict["theta{}".format(i)] = np.array(local_weights)
np.save(gv.dir_name + 'ben_delta_%s_t%s.npy' % (i, t), local_delta)
return
import global_vars
global_vars.init()
b = global_vars.cvar.b
if b != "string b":
raise RuntimeError("Unexpected string: " + b)
global_vars.cvar.b = "a string value"
b = global_vars.cvar.b
if b != "a string value":
raise RuntimeError("Unexpected string: " + b)
x = global_vars.cvar.x
if x != 1234:
raise RuntimeError("Unexpected x: " + str(x))
global_vars.cvar.x = 9876
x = global_vars.cvar.x
if x != 9876:
raise RuntimeError("Unexpected string: " + str(x))
fail = True
try:
global_vars.cvar.notexist = "something"
except AttributeError, e:
fail = False
if fail:
raise RuntimeError("AttributeError should have been thrown")
fail = True
try:
g = global_vars.cvar.notexist
except AttributeError, e:
elif selection == 1:
p_name = input("Select a Party name: ")
try:
assert p_name + "_party.json" not in listFiles('.')
obj_party = party.Party(p_name)
while True:
print("Let's create our Party Members now.")
try:
m_cnt = input(
"How many Heroes would you like to add? ")
m_cnt = int(m_cnt)
for cnt in range(1, m_cnt + 1):
print(cnt)
except Exception as err:
print("[Exception] :: %s" % err)
pass
break
printJson(obj_party)
except AssertionError:
print("That Party Name already exists!!! Start again...")
pass
else:
print('Invalid selection')
gv.init() # call only once
createParty()
return_dict = manager.dict()
return_dict['eval_success'] = 0.0
return_dict['eval_loss'] = 0.0
if args.mal:
return_dict['mal_suc_count'] = 0
for t in range(args.T):
if not os.path.exists(gv.dir_name + 'global_weights_t%s.npy' % t):
print('No directory found for iteration %s' % t)
break
if args.mal:
p_eval = Process(target=eval_func,
args=(X_test, Y_test_uncat, t, return_dict,
mal_data_X, mal_data_Y))
else:
p_eval = Process(target=eval_func,
args=(X_test, Y_test_uncat, t, return_dict))
p_eval.start()
p_eval.join()
if args.mal:
print('Malicious agent succeeded in %s of %s iterations' %
(return_dict['mal_suc_count'], (t - 1) * args.mal_num))
if __name__ == "__main__":
args = gv.init()
tf.set_random_seed(777)
np.random.seed(777)
main(args)
passwd=passwd)
crawler.crawl(spider)
reactor_control.add_crawler()
crawler.start()
log.msg('Crawler %d started...' % index, log.INFO)
if __name__ == '__main__':
import sys
if len(sys.argv) < 2:
print 'python %s <uid-list-path>' % sys.argv[1]
sys.exit(0)
uid_list_path = sys.argv[1]
# 初始化全局变量
global_vars.init(uid_list_path)
spider_name = 'userinfo'
reactor_control = ReactorControl()
log.start(loglevel=log.DEBUG, logstdout=False)
settings = get_project_settings()
#crawler = Crawler(settings)
# 从配置文件中获取微博账户信息
weibo_accounts = settings.getlist('WEIBO_USER_ACCOUNTS')
for i in range(len(weibo_accounts)):
setup_crawler(spider_name, i, weibo_accounts[i]['username'],
weibo_accounts[i]['passwd'])
reactor.run()
if g.combined_filter_type == 'MAG_DIR_HLS_HSV':
combined[(((mag_bin == 1) & (dir_bin == 1))) | ((hls_bin == 1) |
(hsv_bin == 1))] = 1
if g.combined_filter_type == 'DEFAULT' or g.combined_filter_type == 'SOBEL_MAG_DIR_HLS_HSV':
combined[(abs_bin == 1 | ((mag_bin == 1) & (dir_bin == 1))) |
((hls_bin == 1) | (hsv_bin == 1))] = 1
return combined, abs_bin, mag_bin, dir_bin, hls_bin, hsv_bin # DEBUG
if __name__ == '__main__':
img_file = 'test_images/straight_lines1.jpg'
img_file = 'test_images/test5.jpg'
g.init()
img = mpimg.imread(img_file)
img = cv2.undistort(img, g.mtx, g.dist, None, g.mtx)
combined, abs_bin, mag_bin, dir_bin, hls_bin, hsv_bin = combined_thresh(
img)
plt.subplot(2, 3, 1)
plt.imshow(abs_bin, cmap='gray', vmin=0, vmax=1)
plt.subplot(2, 3, 2)
plt.imshow(mag_bin, cmap='gray', vmin=0, vmax=1)
plt.subplot(2, 3, 3)
plt.imshow(dir_bin, cmap='gray', vmin=0, vmax=1)
plt.subplot(2, 3, 4)
plt.imshow(hls_bin, cmap='gray', vmin=0, vmax=1)
def mal_agent(X_shard, Y_shard, mal_data_X, mal_data_Y, t, gpu_id, return_dict,
mal_visible, X_test, Y_test):
args = gv.init()
shared_weights = np.load(gv.dir_name + 'global_weights_t%s.npy' % t,
allow_pickle=True)
if 'theta{}'.format(gv.mal_agent_index) in return_dict.keys():
pre_theta = return_dict['theta{}'.format(gv.mal_agent_index)]
else:
pre_theta = None
holdoff_flag = 0
if 'holdoff' in args.mal_strat:
print('Checking holdoff')
if 'single' in args.mal_obj:
target, target_conf, actual, actual_conf = mal_eval_single(
mal_data_X, mal_data_Y, shared_weights)
if target_conf == 1:
print('Holding off')
holdoff_flag = 1
# tf.reset_default_graph()
tf.keras.backend.set_learning_phase(1)
print('Malicious Agent on GPU %s' % gpu_id)
# set environment
os.environ["CUDA_DEVICE_ORDER"] = "PCI_BUS_ID"
os.environ["CUDA_VISIBLE_DEVICES"] = str(gpu_id)
if args.dataset == 'census':
x = tf.placeholder(shape=(None, gv.DATA_DIM), dtype=tf.float32)
y = tf.placeholder(dtype=tf.int64)
else:
x = tf.placeholder(shape=(None, gv.IMAGE_ROWS, gv.IMAGE_COLS,
gv.NUM_CHANNELS),
dtype=tf.float32)
y = tf.placeholder(dtype=tf.int64)
if 'MNIST' in args.dataset:
agent_model = model_mnist(type=args.model_num)
elif args.dataset == 'CIFAR-10':
agent_model = cifar10_model()
elif args.dataset == 'census':
agent_model = census_model_1()
else:
return
logits = agent_model(x)
prediction = tf.nn.softmax(logits)
eval_loss = tf.reduce_mean(
tf.nn.sparse_softmax_cross_entropy_with_logits(labels=y,
logits=logits))
config = tf.ConfigProto(gpu_options=gv.gpu_options)
# config.gpu_options.allow_growth = True
sess = tf.Session(config=config)
tf.keras.backend.set_session(sess)
if t >= args.mal_delay and holdoff_flag == 0:
if args.mal_obj == 'all':
final_delta = mal_all_algs(x, y, logits, agent_model,
shared_weights, sess, mal_data_X,
mal_data_Y, t)
elif args.mal_obj == 'single' or 'multiple' in args.mal_obj:
final_delta, penul_delta = mal_single_algs(
x, y, logits, agent_model, shared_weights, sess, mal_data_X,
mal_data_Y, t, mal_visible, X_shard, Y_shard, pre_theta)
else:
return
elif t < args.mal_delay or holdoff_flag == 1:
print('Delay/Hold-off')
final_delta, _ = benign_train(x, y, agent_model, logits, X_shard,
Y_shard, sess, shared_weights)
else:
return
final_weights = shared_weights + final_delta
agent_model.set_weights(final_weights)
print('---Eval at mal agent---')
if 'single' in args.mal_obj:
target, target_conf, actual, actual_conf = mal_eval_single(
mal_data_X, mal_data_Y, final_weights)
print(
'Target:%s with conf. %s, Curr_pred on malicious model for iter %s:%s with conf. %s'
% (target, target_conf, t, actual, actual_conf))
elif 'multiple' in args.mal_obj:
suc_count_local = mal_eval_multiple(mal_data_X, mal_data_Y,
final_weights)
print('%s of %s targets achieved' % (suc_count_local, args.mal_num))
eval_success, eval_loss = eval_minimal(X_test, Y_test, final_weights)
return_dict['mal_success'] = eval_success
print('Malicious Agent: success {}, loss {}'.format(
eval_success, eval_loss))
write_dict = dict()
# just to maintain ordering
write_dict['t'] = t + 1
write_dict['eval_success'] = eval_success
write_dict['eval_loss'] = eval_loss
file_write(write_dict, purpose='mal_eval_loss')
return_dict[str(gv.mal_agent_index)] = np.array(final_delta)
return_dict["theta{}".format(gv.mal_agent_index)] = np.array(final_weights)
np.save(gv.dir_name + 'mal_delta_t%s.npy' % t, final_delta)
if 'auto' in args.mal_strat or 'multiple' in args.mal_obj:
penul_weights = shared_weights + penul_delta
if 'single' in args.mal_obj:
target, target_conf, actual, actual_conf = mal_eval_single(
mal_data_X, mal_data_Y, penul_weights)
print(
'Penul weights ---- Target:%s with conf. %s, Curr_pred on malicious model for iter %s:%s with conf. %s'
% (target, target_conf, t, actual, actual_conf))
elif 'multiple' in args.mal_obj:
suc_count_local = mal_eval_multiple(mal_data_X, mal_data_Y,
penul_weights)
print('%s of %s targets achieved' %
(suc_count_local, args.mal_num))
eval_success, eval_loss = eval_minimal(X_test, Y_test, penul_weights)
print('Penul weights ---- Malicious Agent: success {}, loss {}'.format(
eval_success, eval_loss))
return
def mal_single_algs(x, y, logits, agent_model, shared_weights, sess,
mal_data_X, mal_data_Y, t, mal_visible, X_shard, Y_shard,
pre_theta):
# alg_num = 2
args = gv.init()
alpha_m = 1.0 / args.k
print(mal_visible)
if args.gar == 'avg':
delta_other_prev = est_accuracy(mal_visible, t)
if pre_theta is None:
start_weights = shared_weights
constrain_weights = shared_weights
else:
start_weights = pre_theta - gv.moving_rate * (pre_theta -
shared_weights)
constrain_weights = pre_theta - gv.moving_rate * (pre_theta -
shared_weights)
if len(mal_visible) >= 1 and 'prev_1' in args.mal_strat:
# Starting with weights that account for other agents
start_weights = shared_weights + delta_other_prev
print('Alg 1: Adding benign estimate')
if 'dist' in args.mal_strat:
if 'dist_oth' in args.mal_strat and t >= 1:
constrain_weights = start_weights + delta_other_prev
else:
final_delta, _ = benign_train(x, y, agent_model, logits, X_shard,
Y_shard, sess, shared_weights)
constrain_weights = start_weights + final_delta
tf.keras.backend.set_session(sess)
elif 'add_ben' in args.mal_strat:
ben_delta, loss_val_shard = benign_train(x, y, agent_model, logits,
X_shard, Y_shard, sess,
shared_weights)
elif 'unlimited' in args.mal_strat:
ben_delta, loss_val_shard = benign_train(x, y, agent_model, logits,
X_shard, Y_shard, sess,
shared_weights)
loss1 = tf.reduce_mean(
tf.nn.sparse_softmax_cross_entropy_with_logits(labels=y,
logits=logits))
mal_loss1 = tf.reduce_mean(
tf.nn.sparse_softmax_cross_entropy_with_logits(labels=y,
logits=logits))
prediction = tf.nn.softmax(logits)
if 'dist' in args.mal_strat:
# Adding weight based regularization
loss, loss2, mal_loss = weight_constrain(loss1, mal_loss1, agent_model,
constrain_weights, t)
else:
loss = loss1
mal_loss = mal_loss1
loss2 = None
weights_pl = None
if 'adam' in args.optimizer:
optimizer = tf.train.AdamOptimizer(
learning_rate=args.eta).minimize(loss)
mal_optimizer = tf.train.AdamOptimizer(
learning_rate=args.eta).minimize(mal_loss)
elif 'sgd' in args.optimizer:
mal_optimizer = tf.train.GradientDescentOptimizer(
learning_rate=args.eta).minimize(mal_loss)
optimizer = tf.train.GradientDescentOptimizer(
learning_rate=args.eta).minimize(loss)
sess.run(tf.global_variables_initializer())
if pre_theta is None:
agent_model.set_weights(shared_weights)
else:
theta = pre_theta - gv.moving_rate * (pre_theta - shared_weights)
agent_model.set_weights(theta)
print('loaded shared weights for malicious agent')
mal_data_Y = mal_data_Y.reshape((len(mal_data_Y), ))
shard_size = len(X_shard)
delta_mal = []
for l in range(len(start_weights)):
layer_shape = start_weights[l].shape
delta_mal.append(np.zeros(shape=layer_shape))
# Not including training loss
if 'train' not in args.mal_strat:
num_mal_epochs = args.mal_E
step = 0
mal_loss_val = 100
while mal_loss_val > 1e-6 or step < num_mal_epochs:
step_weight_start = np.array(agent_model.get_weights())
sess.run(mal_optimizer, feed_dict={x: mal_data_X, y: mal_data_Y})
if 'auto' in args.mal_strat:
step_weight_end = agent_model.get_weights()
delta_mal += (1 / args.mal_boost) * (step_weight_end -
step_weight_start)
agent_model.set_weights(start_weights +
(1 / args.mal_boost) * delta_mal)
if step % 100 == 0:
mal_obj_pred, mal_loss_val = sess.run([prediction, mal_loss],
feed_dict={
x: mal_data_X,
y: mal_data_Y
})
if 'single' in args.mal_obj:
print(
'Target:%s w conf.: %s, Curr_pred at step %s:%s, Loss: %s'
% (mal_data_Y, mal_obj_pred[:, mal_data_Y], step,
np.argmax(mal_obj_pred, axis=1), mal_loss_val))
elif 'multiple' in args.mal_obj:
suc_count_local = np.sum(
mal_data_Y == np.argmax(mal_obj_pred, axis=1))
print('%s of %s targets achieved at step %s, Loss: %s' %
(suc_count_local, args.mal_num, step, mal_loss_val))
step += 1
# Including training loss
elif 'train' in args.mal_strat:
# mal epochs different from benign epochs
if args.mal_E > args.E:
num_mal_epochs = args.mal_E
else:
num_mal_epochs = args.E
# fixed number of steps
if args.steps is not None:
num_steps = args.steps
start_offset = (t * args.B * args.steps) % (shard_size - args.B)
else:
num_steps = num_mal_epochs * shard_size / args.B
start_offset = 0
if 'alternate' in args.mal_strat:
if 'unlimited' not in args.mal_strat:
delta_mal_ret = alternate_train(sess, t, optimizer, loss,
mal_optimizer, mal_loss, x, y,
logits, X_shard, Y_shard,
mal_data_X, mal_data_Y,
agent_model, num_steps,
start_offset, loss1, loss2)
elif 'unlimited' in args.mal_strat:
# train until loss matches that of benign trained
alternate_train_unlimited(sess, t, optimizer, loss,
mal_optimizer, mal_loss, x, y,
logits, X_shard, Y_shard, mal_data_X,
mal_data_Y, agent_model, num_steps,
start_offset, loss_val_shard, loss1,
loss2)
elif 'concat' in args.mal_strat:
# training with concatenation
concat_train(sess, optimizer, loss, mal_optimizer, mal_loss, x, y,
logits, X_shard, Y_shard, mal_data_X, mal_data_Y,
agent_model, num_steps, start_offset)
elif 'data_poison' in args.mal_strat:
num_steps += (num_mal_epochs * args.data_rep) / args.B
data_poison_train(sess, optimizer, loss, mal_optimizer, mal_loss,
x, y, logits, X_shard, Y_shard, mal_data_X,
mal_data_Y, agent_model, num_steps, start_offset)
if 'auto' not in args.mal_strat:
# Explicit boosting
delta_naive_mal = agent_model.get_weights() - start_weights
if len(mal_visible) >= 1 and 'prev_2' in args.mal_strat:
print('Alg 2: Deleting benign estimate')
# Algorithm 2: Adjusting weights after optimzation
delta_mal = delta_naive_mal - delta_other_prev
elif len(mal_visible) < 1 or 'prev_2' not in args.mal_strat:
delta_mal = delta_naive_mal
# Boosting weights
if 'no_boost' in args.mal_strat or 'alternate' in args.mal_strat or 'concat' in args.mal_strat or 'data_poison' in args.mal_strat:
print('No boosting')
delta_mal = delta_mal
else:
print('Boosting by %s' % args.mal_boost)
delta_mal = args.mal_boost * delta_mal
if 'add_ben' in args.mal_strat:
print('Direct addition of benign update')
delta_mal += ben_delta
else:
# Implicit boosting
print('In auto mode')
delta_naive_mal = alpha_m * delta_mal_ret
delta_mal = delta_mal_ret
return delta_mal, delta_naive_mal
def benign_train(x, y, agent_model, logits, X_shard, Y_shard, sess,
shared_weights):
args = gv.init()
print('Training benign model at malicious agent')
loss = tf.reduce_mean(
tf.nn.sparse_softmax_cross_entropy_with_logits(labels=y,
logits=logits))
prediction = tf.nn.softmax(logits)
if args.optimizer == 'adam':
optimizer = tf.train.AdamOptimizer(
learning_rate=args.eta).minimize(loss)
elif args.optimizer == 'sgd':
optimizer = tf.train.GradientDescentOptimizer(
learning_rate=args.eta).minimize(loss)
if args.k > 1:
config = tf.ConfigProto(gpu_options=gv.gpu_options)
# config.gpu_options.allow_growth = True
temp_sess = tf.Session(config=config)
elif args.k == 1:
temp_sess = tf.Session()
tf.keras.backend.set_session(temp_sess)
temp_sess.run(tf.global_variables_initializer())
agent_model.set_weights(shared_weights)
shard_size = len(X_shard)
if args.mal_E > args.E:
num_mal_epochs = args.mal_E
else:
num_mal_epochs = args.E
for step in range(int(num_mal_epochs * shard_size / args.B)):
offset = (step * args.B) % (shard_size - args.B)
X_batch = X_shard[offset:(offset + args.B)]
Y_batch = Y_shard[offset:(offset + args.B)]
Y_batch_uncat = np.argmax(Y_batch, axis=1)
_, loss_val = temp_sess.run([optimizer, loss],
feed_dict={
x: X_batch,
y: Y_batch_uncat
})
# if step % 100 == 0:
# print loss_val
final_weights = agent_model.get_weights()
final_delta = final_weights - shared_weights
agent_model.set_weights(final_weights)
num_steps_temp = int(shard_size / args.B)
offset_temp = 0
loss_val_shard = 0.0
for step_temp in range(num_steps_temp):
offset_temp = (offset + step_temp * args.B) % (shard_size - args.B)
X_batch = X_shard[offset:(offset + args.B)]
Y_batch = Y_shard[offset:(offset + args.B)]
Y_batch_uncat = np.argmax(Y_batch, axis=1)
loss_val_shard += temp_sess.run(loss,
feed_dict={
x: X_batch,
y: Y_batch_uncat
})
loss_val_shard = loss_val_shard / num_steps_temp
print('Average loss on the data shard %s' % loss_val_shard)
temp_sess.close()
return final_delta, loss_val_shard
def alternate_train(sess,
t,
optimizer,
loss,
mal_optimizer,
mal_loss,
x,
y,
logits,
X_shard,
Y_shard,
mal_data_X,
mal_data_Y,
agent_model,
num_steps,
start_offset,
loss1=None,
loss2=None):
args = gv.init()
step = 0
num_local_steps = args.ls
shard_size = len(X_shard)
curr_weights = agent_model.get_weights()
delta_mal_local = []
for l in range(len(curr_weights)):
layer_shape = curr_weights[l].shape
delta_mal_local.append(np.zeros(shape=layer_shape))
while step < num_steps:
offset = (start_offset + step * args.B) % (shard_size - args.B)
# Benign
if step < num_steps:
for l_step in range(num_local_steps):
# training
# print offset
offset = (offset + l_step * args.B) % (shard_size - args.B)
X_batch = X_shard[offset:(offset + args.B)]
Y_batch = Y_shard[offset:(offset + args.B)]
Y_batch_uncat = np.argmax(Y_batch, axis=1)
if 'dist' in args.mal_strat:
loss1_val, loss2_val, loss_val = sess.run(
[loss1, loss2, loss],
feed_dict={
x: X_batch,
y: Y_batch_uncat
})
sess.run([optimizer],
feed_dict={
x: X_batch,
y: Y_batch_uncat
})
else:
loss_val = sess.run([loss],
feed_dict={
x: X_batch,
y: Y_batch_uncat
})
sess.run([optimizer],
feed_dict={
x: X_batch,
y: Y_batch_uncat
})
mal_loss_val_bef = sess.run([mal_loss],
feed_dict={
x: mal_data_X,
y: mal_data_Y
})
# Malicious, only if mal loss is non-zero
print(mal_loss_val_bef)
if step >= 0 and mal_loss_val_bef[0] > 0.0:
# print('Boosting mal at step %s' % step)
weights_ben_local = np.array(agent_model.get_weights())
if 'dist' in args.mal_strat:
sess.run([mal_optimizer],
feed_dict={
x: mal_data_X,
y: mal_data_Y
})
else:
sess.run([mal_optimizer],
feed_dict={
x: mal_data_X,
y: mal_data_Y
})
if 'auto' in args.mal_strat:
step_weight_end = agent_model.get_weights()
if 'wt_o' in args.mal_strat:
for l in range(len(delta_mal_local)):
if l % 2 == 0:
delta_mal_local[l] += (1 / args.mal_boost) * (
step_weight_end[l] - weights_ben_local[l])
else:
delta_mal_local += (1 / args.mal_boost) * (
step_weight_end - weights_ben_local)
agent_model.set_weights(curr_weights +
(1 / args.mal_boost) * delta_mal_local)
else:
delta_mal_local = agent_model.get_weights() - weights_ben_local
if 'wt_o' in args.mal_strat:
# Boosting only weights
boosted_delta = delta_mal_local.copy()
for l in range(len(delta_mal_local)):
if l % 2 == 0:
boosted_delta[
l] = args.mal_boost * delta_mal_local[l]
boosted_weights = weights_ben_local + boosted_delta
else:
boosted_weights = weights_ben_local + args.mal_boost * delta_mal_local
agent_model.set_weights(boosted_weights)
mal_loss_val_aft = sess.run([mal_loss],
feed_dict={
x: mal_data_X,
y: mal_data_Y
})
if step % 10 == 0 and 'dist' in args.mal_strat:
print(
'Benign: Loss1 - %s, Loss2 - %s, Loss - %s; Mal: Loss_bef - %s Loss_aft - %s'
% (loss1_val, loss2_val, loss_val, mal_loss_val_bef,
mal_loss_val_aft))
elif step % 10 == 0 and 'dist' not in args.mal_strat:
print('Benign: Loss - %s; Mal: Loss_bef - %s, Loss_aft - %s' %
(loss_val, mal_loss_val_bef, mal_loss_val_aft))
if step % 100 == 0 and t < 5:
np.save(gv.dir_name + 'mal_delta_t%s_step%s.npy' % (t, step),
delta_mal_local)
step += num_local_steps
return delta_mal_local
def initializeMainGui():
g.init('gui/MotilityTracking.ui')
g.m.trackView = Window(np.zeros((3, 3, 3)))
g.m.histogram = Histogram(title='Mean Single Lag Distance Histogram',
labels={
'left': 'Count',
'bottom': 'Mean SLD Per Track (Pixels)'
})
g.m.trackPlot = TrackPlot()
g.m.trackPlot.tracksChanged.connect(update_plots)
g.m.trackView.imageview.addItem(g.m.trackPlot)
g.m.trackView.imported = pg.ScatterPlotItem()
g.m.trackView.imageview.addItem(g.m.trackView.imported)
g.m.actionImportBin.triggered.connect(
lambda: open_file_gui(lambda f: import_mat(open_bin(f)),
prompt='Import .bin file of tracks',
filetypes='*.bin'))
g.m.actionImportBackground.triggered.connect(
lambda: open_file_gui(set_background_image,
prompt='Select tif file as background image',
filetypes='*.tif *.tiff *.stk'))
g.m.actionImportCoordinates.triggered.connect(
lambda: open_file_gui(import_coords,
prompt='Import coordinates from txt file',
filetypes='*.txt'))
g.m.actionSimulateDistances.triggered.connect(
lambda: save_file_gui(simulate_distances,
prompt='Save simulated distances',
filetypes='*.txt'))
g.m.actionExportMSD.triggered.connect(
lambda: save_file_gui(exportMSD,
prompt='Export Mean Squared Displacement Values',
filetypes='*.txt'))
g.m.actionExportHistogram.triggered.connect(
lambda: save_file_gui(g.m.histogram.export,
prompt='Export Histogram Values',
filetypes='*.txt'))
g.m.actionExportTrackLengths.triggered.connect(lambda: save_file_gui(
export_track_lengths, prompt='Export Track Lengths', filetypes='*.txt')
)
g.m.actionExportOutlined.triggered.connect(g.m.trackPlot.export_gui)
g.m.actionExportDistances.triggered.connect(lambda: save_file_gui(
export_real_distances, prompt='Export Distances', filetypes='*.txt'))
g.m.MSLDMinSpin.setOpts(value=0, decimals=2, bounds=(0, 100))
g.m.MSLDMaxSpin.setOpts(value=100, decimals=2, bounds=(0, 100))
g.m.neighborsSpin.setOpts(value=0, bounds=(0, 100), int=True, step=1)
g.m.neighborDistanceSpin.setOpts(value=1, decimals=2, bounds=(0, 100))
g.m.minLengthSpin.setOpts(value=4, bounds=(0, 1000), int=True, step=1)
g.m.maxLengthSpin.setOpts(value=20, bounds=(0, 1000), int=True, step=1)
g.m.MSLDGroupBox.toggled.connect(g.m.trackPlot.filter)
g.m.additionalGroupBox.toggled.connect(g.m.trackPlot.filter)
g.m.MSLDMaxSpin.sigValueChanged.connect(g.m.trackPlot.filter)
g.m.MSLDMinSpin.sigValueChanged.connect(g.m.trackPlot.filter)
g.m.neighborDistanceSpin.sigValueChanged.connect(g.m.trackPlot.filter)
g.m.neighborsSpin.sigValueChanged.connect(g.m.trackPlot.filter)
g.m.minLengthSpin.sigValueChanged.connect(g.m.trackPlot.filter)
g.m.maxLengthSpin.sigValueChanged.connect(g.m.trackPlot.filter)
g.m.hideBackgroundCheck.toggled.connect(
lambda v: g.m.trackView.imageview.getImageItem().setVisible(not v))
g.m.ignoreOutsideCheck.toggled.connect(g.m.trackPlot.filter)
g.m.plotMeansCheck.toggled.connect(g.m.trackPlot.means.setVisible)
g.m.plotTracksCheck.toggled.connect(g.m.trackPlot.tracks.setVisible)
g.m.viewTab.layout().insertWidget(0, g.m.trackView)
g.m.MSDWidget = pg.PlotWidget(title='Mean Squared Displacement Per Lag',
labels={
'left': 'Mean Squared Disance (p^2)',
'bottom': 'Lag Count'
})
g.m.analysisTab.layout().addWidget(g.m.MSDWidget)
g.m.analysisTab.layout().addWidget(g.m.histogram)
g.m.CDFWidget = CDFWidget(
) #pg.PlotWidget(title = 'Cumulative Distribution Function', labels={'left': 'Cumulative Probability', 'bottom': 'Single Lag Displacement Squared'})
#g.m.CDFPlot = pg.PlotCurveItem()
g.m.cdfTab.layout().addWidget(g.m.CDFWidget)
g.m.installEventFilter(mainWindowEventEater)
g.m.setWindowTitle('Motility Tracking')
g.m.show()
crawler.signals.connect(reactor_control.remove_crawler, signal=signals.spider_closed)
spider = crawler.spiders.create(spider_name, index = index, username=username, passwd = passwd)
crawler.crawl(spider)
reactor_control.add_crawler()
crawler.start()
log.msg('Crawler %d started...' % index, log.INFO)
if __name__ == '__main__':
import sys
if len(sys.argv) < 2:
print 'python %s <uid-list-path>' % sys.argv[1]
sys.exit(0)
uid_list_path = sys.argv[1]
# 初始化全局变量
global_vars.init(uid_list_path)
spider_name = 'userinfo'
reactor_control = ReactorControl()
log.start(loglevel = log.DEBUG, logstdout = False)
settings = get_project_settings()
#crawler = Crawler(settings)
# 从配置文件中获取微博账户信息
weibo_accounts = settings.getlist('WEIBO_USER_ACCOUNTS')
for i in range(len(weibo_accounts)):
setup_crawler(spider_name, i, weibo_accounts[i]['username'], weibo_accounts[i]['passwd'])
reactor.run()
