def generate_header(module):
if not Args().skip_header_generation:
text = json.dumps(HeaderGenerator().visit(module), separators=(',', ':'))
with open(Args().get_header_path(), 'w') as file:
file.write(text)
return handle_next_stage(module, generate_code)
def test_header_inclusion(self):
self.reset_module()
a_code = '''module a
type a<`a, `b> = {
A,
B,
C = `a * `b,
D = a<`a, a<bool, string>> * `b
}
let foo = fun(a, b) -> { a + b }
let bar = fun(a, b) -> { foo(a, b) != 0 }
let baz = fun(f, x, y) -> { f(x) = y } '''
b_code = '''module b
type b<`a> = {
A,
B = `a * int
}
let foo = fun(a, b) -> { a = b }
let bar = fun(a) -> { foo(a, 0) }
let baz = fun(f, g, x) -> { f(g(x)) }'''
c_code = '''module c
import "a"
open "b"
let test1 = fun(a, b, c) -> { foo(a.baz(bar, a, b), c) }
let test2 = baz
let test3 = a.bar'''
Args().source = 'a.tml'
parse_source_code(a_code)
self.reset_module()
Args().source = 'b.tml'
parse_source_code(b_code)
self.reset_module()
Args().source = 'c.tml'
Args().skip_header_generation = True
parse_source_code(c_code)
assert_let_types(
self, {
'test1':
fun_type([t_int, t_bool, t_bool], t_bool),
'test2':
fun_type([fun_type([t_a], t_b),
fun_type([t_c], t_a), t_c], t_b),
'test3':
fun_type([t_int, t_int], t_bool)
})
def test_header_to_json_and_back(self):
Args().skip_header_generation = True
code = ''' module test
type list<`a> = {
Empty,
Cons = `a * list<`a>
}
type a<`x, `y> = {
B = ref<ref<ref<`x>>> * `y * `x * ref<ref<ref<`y>>>,
A = a<bool, a<int, float>>
}
let a: `a -> `a -> bool = fun(a, b) -> {True}
let b = fun(f, x, y) -> {f(x) = y}
let c: string -> a<int, string> -> unit = fun(p, q) -> {()}
let d = fun() -> { let a = 4; a + 4 }
let foo = fun(a, b) -> { a + b }
let bar = fun(a, b) -> { foo(a, b) != 0 }
let baz = fun(f, x, y) -> { f(x) = y }'''
parse_source_code(code)
dic = HeaderGenerator().visit(GlobalModule())
module2 = HeaderReader().read_module(dic)
dic2 = HeaderGenerator().visit(module2)
self.assertEqual(json.dumps(dic), json.dumps(dic2))
def infer_types(module):
if Args().stop_before_type_inferring:
print(GlobalTypeInferer().dump())
exit(0)
GlobalTypeInferer().infer()
return handle_next_stage(module, generate_header)
def main():
args = Args().get_args()
kwargs = vars(args)
checkpoint = torch.load(args.checkpoint)
base_classifier = get_architecture(checkpoint["arch"], args.dataset)
base_classifier.load_state_dict(checkpoint['state_dict'])
attacker = SmoothAttack(base_classifier)
smoothed_classifier = Smooth(base_classifier,
get_num_classes(args.dataset), args.sigma)
dataset = get_dataset(args.dataset, 'test')
average_nat = []
average_adv = []
j_header('index', 'nat_y', 'adv_y', 'nat_rad', 'adv_rad', 'success')
figure = FigureSaver()
for i in range(0, len(dataset), args.skip):
(x, label) = dataset[i]
x = x.cuda()
first_x = x.data
nat_pred, nat_rad = smoothed_classifier.certify(
x, args.N0, args.N, args.alpha, args.batch)
if nat_pred is -1:
continue
if args.dataset == DATASETS[0]: # ImageNet
targets = [j for j in range(0, 1000, 100) if j is not label]
else:
targets = [j for j in range(10) if j is not label]
best_rad = -10.0
best_image = None
best_target = -1
for target in targets:
adv_x = attacker.perturb(x=first_x, y=target, **kwargs)
# If you want to do wasserstein attack, uncomment the following and change the attacker to wasserstein
# adv_x = attacker.perturb(x=first_x, y=target, eps=args.sigma, steps=args.steps, batch=args.batch)
adv_pred, adv_rad = smoothed_classifier.certify(
adv_x, args.N0, 2 * args.N0, args.alpha, args.batch)
adv_suc = (adv_pred != label) and (adv_pred != -1) and (nat_pred !=
-1)
adv_rad = adv_rad if adv_suc else -adv_rad
if adv_rad > best_rad:
best_rad = adv_rad
best_image = adv_x.data
best_target = target
figure.save(best_image, i, 'best={}'.format(best_target))
figure.save(first_x, i, 'natural')
best_pred, best_rad = smoothed_classifier.certify(
best_image, args.N0, args.N, args.alpha, args.batch)
j_print(i, label, best_target, nat_rad, best_rad)
average_adv.append(best_rad)
average_nat.append(nat_rad)
average_nat = np.array(average_nat)
average_adv = np.array(average_adv)
print('Average nat radii {}, Average adv radii {}'.format(
average_nat.mean(), average_adv.mean()))
def run_test_code(self, source_path: str, expect_fail: bool):
args = Args()
self.init_args(args)
args.source_path = source_path
args.expect_fail = expect_fail
io = TestIo()
run(args, io)
def parse_source_code(text: str):
ast = parser.parse(text, tracking=True)
if Args().stop_after_parsing and Errors().is_ok():
print(json.dumps(AstToDictVisitor().visit(ast)))
exit(0)
return handle_next_stage(ast, visit_ast)
def get_exp_name() -> str:
args = Args()
name_list = [str(get_exp_no()), _time()]
exp_name = args.get_args().experiment
if str(exp_name) is not 'None':
name_list.append(exp_name)
name_list.append(args.get_name())
return '_'.join(name_list)
def main():
time1 = time.time()
print('Start:',
time.strftime("%a, %d %b %Y %H:%M:%S +0000", time.gmtime(time1)))
args = Args()
print(args)
'''
Ler arquivo args.reference_output
searchLevel (achar abaixo e acima de Fermi)
diferenca entre abaixo e acima (gap ref)
'''
below, above = utils.search_level(args.reference_output)
ref_gap = abs(above - below)
print(ref_gap)
particle = utils.read_particle(args.particle_path)
print('>' + str(particle).strip() + '<')
print(args.specie_type_1)
print(args.specie_nat_1)
print(args.specie_type_2)
print(args.specie_nat_2)
#montar liga na fracao informada pelo usuario
#Sorteio de posicoes
from random import sample
at_species = [args.specie_type_1] * args.n_atoms
random_indices = sample(list(range(args.n_atoms)), args.specie_nat_2)
for idx in random_indices:
at_species[idx] = args.specie_type_2
print(at_species)
particle.at_species = at_species
print('>' + str(particle).strip() + '<')
template_file = open('input_template.in')
content = template_file.read()
template_file.close()
content = content.replace('<#NAT#>', str(args.n_atoms))
content = content.replace('<#NTYP#>', str(args.n_species))
content = content.replace('<#PARTICLE#>', str(particle).strip())
espresso_input_file = open('input.in', 'w')
print(content)
'''
Colocar no template
content = ""
content = content.replace('<#NAT#>', args.n_atoms)
content = content.replace('<#NTYP#>', args.n_species)
content = content.replace('<#PARTICLE#>', conteudo_particula)
#salvar arquivo -> new_content
'''
'''
def main():
args = Args()
lfm = lastfm.init()
user = lfm.get_user(args.user)
artists = get_artists(user, args)
genres = get_genres(artists, args)
chart.save(args, genres)
def main():
"""
Main function
"""
args = Args(sys.argv)
if args.parse_args() == macro.ERROR:
return macro.ERROR
zappy = Zappy(args)
return zappy.launch()
def get_args():
# create args
# You just need to change the parameters here
at = False # Todo: change here
if not at: # for normal training
return Args(
dataset="imdb",
model_short_name="cnn",
batch_size=32,
epochs=75,
adversarial_training=False,
orig_model_prefix="cnn-imdb",
max_length=2500,
# for evaluate
attack_class_for_testing=attack_classes[
1], # test robustness against which model
num_attack_samples=50, # how many samples to test robustness with
# for pre-generate
attack_class_for_training=attack_classes[
1], # specify which attack to generate adv samples on the trained model
adv_sample_file=
"cnn-imdb-bae.csv", # file name of where to save adv. samples
adversarial_samples_to_train=
2000, # how many samples in adv_sample_file
)
else: # for adversarial training
return Args(
dataset="kaggle-toxic-comment",
model_short_name="lstm",
batch_size=32,
epochs=75,
adversarial_training=True,
at_model_prefix="lstm-at-tb-imdb",
adv_sample_file=
"lstm-kaggle-tb.csv", # from which file program will read adv. samples
# for evaluate
attack_class_for_testing=attack_classes[0],
num_attack_samples=250,
)
def CAN(DX, c, k=15, r=-1, islocal=True):
"""
:param DX: a temporal sequence of n x n distance matrix for n data points
:param c: number of clusters
:param k: number of neighbors to determine the initial graph, and the parameter r if r<=0
:param r: paremeter, which could be set to a large enough value. If r<0, then it is determined by algorithm with k
:param islocal:
1: only update the similarities of the k neighbor pairs, faster
0: update all the similarities
:return:
A: num*num learned symmetric similarity matrix
evs: eigenvalues of learned graph Laplacian in the iterations
"""
arg = Args()
N_ITER = arg.n_iter
SDX = []
if arg.debug:
print('\nInitial Parameters:', '\nc = ', c, '\nk = ', k, '\nr = ', r,
'\nislocal = ', islocal, '\nNITER = ', N_ITER)
# Initialization
for distX in DX:
num = distX.shape[0]
distX1 = np.sort(distX, axis=1)
idx = np.argsort(distX, axis=1)
A = np.zeros((num, num))
rr = get_gamma(distX, k)
eps = 10e-10
for i in range(0, num):
A[i, idx[i, 1:k + 2]] = 2 * (distX1[i, k + 1] -
distX1[i, 1:k + 2]) / (rr[i] + eps)
SDX.append(A)
if r < 0:
r = np.mean(rr)
lmd = np.mean(rr)
wt_local = [
np.ones(args.c_dim) / len(connectome_list)
for i in range(0, len(connectome_list))
]
for itr in range(0, N_ITER):
for t in range(0, len(DX)):
distX = DX[t]
A = SDX[t]
return A
def fetch_resource(resource_type):
args = Args(resource_type)
db_manager = DbManager(args)
scraper = get_scraper(args)
if args.refetch:
db_manager.delete_resource(args.db_key)
if not db_manager.resource_exists(args.db_key):
resource_data = scraper.get_resource(args.query_params)
if scraper.driver:
scraper.driver.quit()
db_manager.save_resource(args.db_key, resource_data)
return db_manager.fetch_resource(args.db_key)
def create_brain_net_node_files(sub, tentative_label=[]):
"""
:param sub: subject name
:param tentative_label: optional label list which can influence the label numbers
:return:
"""
args = Args()
data_dir = os.path.join(os.path.join(args.root_directory, os.pardir),
'AD-Data_Organized')
file_parcellation_table = os.path.join(
os.path.join(os.path.join(data_dir, sub),
'helper_files/parcellationTable.json'))
# Reading parcellation table to get the coordinates of each region
with open(file_parcellation_table) as f:
table = json.load(f)
connectomes = readMatricesFromDirectory(os.path.join(
data_dir, sub)) # Reading connectomes
n_components, connectome_labels = get_number_of_components(connectomes)
if len(tentative_label) > 0:
print(len(tentative_label[0]))
connectome_labels = [
replace_labels(connectome_labels[t], tentative_label[0])
for t in range(0, len(connectome_labels))
]
T = len(connectomes) # Number of time points
N = len(table) # Number of regions
node_size = 4
output_dir = os.path.join(data_dir,
sub + '/helper_files/brain_net/node_file')
if not os.path.exists(output_dir):
os.makedirs(output_dir)
for t in range(1, T + 1):
print("\n----", t, "----\n")
print(len(connectome_labels[t - 1]), N)
assert (len(connectome_labels[t -
1]) == N), "Incorrect number of regions"
with open(os.path.join(output_dir, sub + "_t" + str(t) + ".node"),
'w') as node_file:
print("Writing: ", node_file.name)
for i in range(0, N):
node_file.write(
str(table[i]["coord"][0]) + " " +
str(table[i]["coord"][1]) + " " +
str(table[i]["coord"][2]) + " " +
str(connectome_labels[t - 1][i] * 10) + " " +
str(node_size) + " " + table[i]["name"] + "\n")
def main():
args = Args()
app = web.Application()
app.add_routes([
web.post("/connection", routes.connection.connection),
web.get("/printer", routes.get_status.get_status),
web.get("/job", routes.get_job.get_job),
web.post("/printer/command", routes.post_command.post_command),
web.post("/files/local/{file}", routes.print_file.print_file),
web.post("/job", routes.cancel_print.cancel_print)
])
get_printer(args.path)
web.run_app(app, host='0.0.0.0', port=args.port)
def main():
args = Args()
with Display(device=args.device, driver=args.driver) as display:
monitor = display.connect_monitor(args.monitor)
if isinstance(monitor, Virtual):
monitor.orientation = args.orientation
display.scale = 8
now = time.monotonic()
rotori = 0
try:
while True:
start = now
while now - start < 1.0:
now = time.monotonic()
if display.channels == 3:
display.buffer[:] = intro_effect(
display.width, display.height)(now) * 0.75
display.pixels[0:2, 0:2] = np.array((0xff, 0, 0))
display.pixels[2:6, 0] = np.array((0, 0xff, 0))
display.pixels[2:6, 1] = np.array((0, 0, 0))
display.pixels[0, 2:8] = np.array((0, 0, 0xff))
display.pixels[1, 2:8] = np.array((0, 0, 0))
elif display.channels == 1:
display.buffer[:, :, 0] = (
intro_effect(display.width, display.height)(now) *
0.75)[:, :, 1]
display.show()
time.sleep(0.01)
rotori += 1
display.rotation = rotori & 3
if isinstance(display, Virtual):
display.orientation = (rotori >> 2) & 3
display.flip_horizontal = (rotori >> 4) & 1
display.flip_vertical = (rotori >> 5) & 1
except KeyboardInterrupt:
pass
def parse_args(_=None):
arg_parser = argparse.ArgumentParser()
arg_parser.add_argument('source', help='Путь к файлу с исходным кодом.')
arg_parser.add_argument('-p', '--stop-after-parsing',
help='Остановиться после синтаксического анализа и вывести АСД в виде JSON.',
action='store_true')
arg_parser.add_argument('-i', '--stop-before-type-inferring',
help='Остановиться перед выводом типов и вывести систему уравнений.',
action='store_true')
arg_parser.add_argument('-g', '--skip-header-saving', help='Пропустить сохранение заголовочного файла модуля.',
action='store_true')
args = arg_parser.parse_args()
Args(args)
return handle_next_stage(None, read_source_code)
def get_graph_data(dataset, isModelDataset=False):
"""
Given a dataset name, loads in the graphs of that dataset.
Creates a specific argument object for that dataset to allow
for different datasets to be created.
return: If isModelDataset: returns args, train, val, test split
else: returgn args, data
"""
args_data = Args()
args_data.change_dataset(dataset)
# Load the graph data - Consider using presaved datasets! with graph load list
graphs = create(args_data)
graphs_len = len(graphs)
random.seed(123)
shuffle(graphs)
# Display some graph stats
graph_node_avg = 0
for graph in graphs:
graph_node_avg += graph.number_of_nodes()
graph_node_avg /= graphs_len
print('Average num nodes', graph_node_avg)
args_data.max_num_node = max(
[graphs[i].number_of_nodes() for i in range(graphs_len)])
max_num_edge = max(
[graphs[i].number_of_edges() for i in range(graphs_len)])
min_num_edge = min(
[graphs[i].number_of_edges() for i in range(graphs_len)])
# show graphs statistics
print('total graph num: {}'.format(graphs_len))
print('max number node: {}'.format(args_data.max_num_node))
print('max/min number edge: {}; {}'.format(max_num_edge, min_num_edge))
print('max previous node: {}'.format(args_data.max_prev_node))
if isModelDataset:
# split datasets
graphs_len = len(graphs)
graphs_test = graphs[int(0.8 * graphs_len):]
graphs_train = graphs[0:int(0.8 * graphs_len)]
graphs_validate = graphs[0:int(0.2 * graphs_len)]
return args_data, graphs_train, graphs_validate, graphs_test
return args_data, graphs
def main():
logger = set_log()
logger.info('程序运行开始')
args = Args().get_all_args()
set_seed(args)
# 进行模型的训练
# 根据参数选择 trainer
Trainer = getattr(trainers, args.trainer_name)
trainer = Trainer(args)
args.do_train = True
if args.do_train:
trainer.train()
# 在做 test 之前，应该需要 load 以前保存的最佳模型
if args.do_test:
trainer.test()
def main() -> None:
args = Args()
args.parse(sys.argv[1:]) # strip off the first argument (program name)
if args.show_info:
logging.basicConfig(level=logging.INFO)
success = False
try:
io = UserIo()
run(args, io)
success = True
except FileNotFoundError as e:
logger.error(e)
except ParseError as e:
logger.error('Syntax error: ' + str(e))
except ProgramError as e:
logger.error('Program failed: ' + str(e))
if not success:
exit(1)
def main(args):
a = Args()
if args.experiment == "nearestneighbor":
nearestneighbor.run(a)
elif args.experiment == "mcts":
mcts.run(a)
elif args.experiment == "exact":
exact.run(a)
elif args.experiment == "gurobi":
gurobi.run(a)
elif args.experiment == "insertion":
insertion.run(a)
elif args.experiment == "policy":
policy.run(a)
elif args.experiment == "parallel":
parallel.run(a)
elif args.experiment == "selfplay":
selfplay.run(a)
else:
raise ValueError("Invalid experiment selection.")
class Logger():
args = Args().args
cwd = os.getcwd()
with open(cwd + "/logging.json", 'r') as logging_configuration_file:
config_dict = json.load(logging_configuration_file)
if 'console' in config_dict['handlers']:
config_dict['handlers']['console']['level'] = args.verbosity
config_dict['handlers']['console']['formatter'] = args.verbosity
# if 'level' in config_dict['root']:
# config_dict['root']['level'] = args.verbosity
for handler in config_dict['handlers']:
# config_dict['handlers'][handler]['formatter'] = args.verbosity
if 'filename' in config_dict['handlers'][handler]:
if path.isfile(config_dict['handlers'][handler]['filename']):
print "Deleting logfile %s" % (
config_dict['handlers'][handler]['filename'])
remove(config_dict['handlers'][handler]['filename'])
logging.config.dictConfig(config_dict)
logger = logging.getLogger(__name__)
logger.info("Logger Initialized")
logger.debug("DEBUG Logger Initialized")
from debug import Debug # noqa
from recovery import Recovery # noqa
from util import Util # noqa
# Sikuli settings
sikuli.Settings.MinSimilarity = Globals.DEFAULT_SIMILARITY
sikuli.Settings.WaitScanRate = Globals.SIKULI_SCANRATE
sikuli.Settings.ObserveScanRate = Globals.SIKULI_SCANRATE
sikuli.Settings.OcrTextRead = True
sikuli.Settings.AutoWaitTimeout = 1
sikuli.Settings.RepeatWaitTime = 0
# check run-time args
args = None
if len(sys.argv) > 1:
args = Args(sys.argv)
# check args, and if none provided, load default config
if args and args.mode == 'cfg':
config = Config(args.cfg)
elif args and args.mode == 'debug':
Debug.find(args.window, args.target, args.similarity)
sys.exit(0)
elif args and args.mode == 'debugc':
Debug.continuously_find(args.window, args.target, args.similarity)
sys.exit(0) # never actually reached
else:
config = Config('config.ini')
kcauto = KCAuto(config)
def parse_args(self):
"""Parse CLI args."""
Args(self.tcex.parser)
self.args = self.tcex.args
def correction(self, inconnue):
# dans notre cas il n'y a qu'un seul jeu d'entrées
self.datasets = [ Args(inconnue) ]
def check(inconnue):
return self.connue + inconnue + self.connue
return ExerciceFunction.correction(self, check)
def parse_args(self):
"""Parse CLI args."""
self.tcex.log.info('Parsing Args.')
Args(self.tcex)
self.args = self.tcex.args
def CAN(distX, c, k=15, r=-1, islocal=True):
"""
:param distX: n x n distance matrix for n data points
:param c: number of clusters
:param k: number of neighbors to determine the initial graph, and the parameter r if r<=0
:param r: paremeter, which could be set to a large enough value. If r<0, then it is determined by algorithm with k
:param islocal:
1: only update the similarities of the k neighbor pairs, faster
0: update all the similarities
:return:
A: num*num learned symmetric similarity matrix
evs: eigenvalues of learned graph Laplacian in the iterations
"""
num = distX.shape[0]
arg = Args()
NITER = arg.n_iter
if arg.debug:
print('\nInitial Parameters:',
'\nc = ', c,
'\nk = ', k,
'\nr = ', r,
'\nislocal = ', islocal,
'\ndimension = ', num,
'\nNITER = ', NITER)
distX1 = np.sort(distX, axis=1)
idx = np.argsort(distX, axis=1)
A = np.zeros((num, num))
rr = get_gamma(distX, k)
if r < 0:
r = np.mean(rr)
lmd = np.mean(rr)
eps = 10e-10
for i in range(0, num):
A[i, idx[i, 1: k + 2]] = 2 * (distX1[i, k + 1] - distX1[i, 1: k + 2]) / (r + eps)
print("\nAverage number of non zero elements per row before optimizing: ", (A > 0).sum() / 148,
"\nNumber of negative elements: ", (A < 0).sum())
A0 = (A + A.T) / 2
A0 = row_normalize(A0)
D0 = np.diag(A0.sum(axis=1))
L0 = D0 - A0
evs, F = get_eigen(L0, c + 1) # Taking c + 1 eig values
F = F[:, 0:c] # removing last one
ev = []
if sum(evs) < 10e-10:
raise Exception('The number of connected component in the graph is greater than {}', c)
for iter in range(0, NITER):
distF = L2_distance(F.T, F.T)
distF = np.sqrt((distF >= 0) * distF)
A = np.zeros((num, num))
for i in range(0, num):
if islocal:
idxa0 = idx[i, 1:k+1]
else:
idxa0 = np.arange(num)
dfi = distF[i, idxa0]
dxi = distX[i, idxa0]
d = dxi + lmd * dfi
ad = -d / (2 * r)
res, _ = EProjSimplex(ad)
A[i, idxa0] = res
np.fill_diagonal(A, 0)
A = (A + A.T) / 2
A = row_normalize(A)
D = np.diag(A.sum(axis=1))
L = D - A
F_old = F
evs, F = get_eigen(L, c + 1)
ev.append(evs)
F = F[:, 0:c] # removing last one
if sum(evs[0:c]) > 10e-10:
lmd = 2 * lmd
elif sum(evs) < 10e-10:
lmd = lmd / 2
F = F_old
else:
break
print("Iter:", iter,
"lambda = ", lmd,
"Regularization Parameter r = ", r)
return A, ev
F = F_old
else:
break
print("Iter:", iter,
"lambda = ", lmd,
"Regularization Parameter r = ", r)
return A, ev
if __name__ == '__main__':
# Read data
data_dir = os.path.join(os.path.dirname(os.getcwd()), 'AD-Data_Organized')
sub = '027_S_2336'
connectome_list = readMatricesFromDirectory(os.path.join(data_dir, sub))
args = Args()
output_dir = os.path.join(data_dir, sub + '_smoothed')
if not os.path.exists(output_dir):
os.makedirs(output_dir)
smoothed_connectomes = []
for t in range(0, len(connectome_list)):
A = connectome_list[t]
dX = np.sqrt(1 - A)
np.fill_diagonal(dX, 2)
S, _ = CAN(dX, args.n_module, k=args.k, islocal=True)
#S = (S.T + S)/2
np.fill_diagonal(S, 0)
S = row_normalize(S)
smoothed_connectomes.append(S)
def main():
args = Args().get_args()
Core(args).recommend()