def printBlock(self, height):
latest = self.db.get('latest')
current_block = utils.deserialize(self.db.get(latest))
while current_block.height != height:
current_block = utils.deserialize(
self.db.get(current_block.prev_hash))
current_block.printBlock()
def printChain(self):
latest = self.db.get('latest')
current_block = utils.deserialize(self.db.get(latest))
while True:
current_block.printBlock()
if current_block.prev_hash == None:
break
current_block = utils.deserialize(
self.db.get(current_block.prev_hash))
def calculate_req_water(type, age, temperature, shower, moisture):
min_l_reg = deserialize('min_l_reg')
rec_l_reg = deserialize('rec_l_reg')
X = pd.DataFrame({'TYPE': type, 'AGE': age, 'TEMPERATURE': temperature, 'SHOWER': shower, 'MOISTURE': moisture},
index=[0])
X = X[['TYPE', 'AGE', 'TEMPERATURE', 'SHOWER', 'MOISTURE']]
min_prediction = min_l_reg.predict(X)[0]
rec_prediction = rec_l_reg.predict(X)[0]
# return [min_prediction, rec_prediction]
return [int(min_prediction), int(rec_prediction)]
def remote_gateway_info(self, gateway):
'''
Return remote gateway information for the specified gateway string id.
@param gateways : gateway id string to search for
@type string
@return remote gateway information
@rtype gateway_msgs.RemotGateway or None
'''
firewall = self._redis_server.get(
hub_api.create_rocon_gateway_key(gateway, 'firewall'))
ip = self._redis_server.get(
hub_api.create_rocon_gateway_key(gateway, 'ip'))
if firewall is None:
return None # equivalent to saying no gateway of this id found
else:
remote_gateway = gateway_msgs.RemoteGateway()
remote_gateway.name = gateway
remote_gateway.ip = ip
remote_gateway.firewall = True if int(firewall) else False
remote_gateway.public_interface = []
encoded_advertisements = self._redis_server.smembers(
hub_api.create_rocon_gateway_key(gateway, 'advertisements'))
for encoded_advertisement in encoded_advertisements:
advertisement = utils.deserialize_connection(
encoded_advertisement)
remote_gateway.public_interface.append(advertisement.rule)
remote_gateway.flipped_interface = []
encoded_flips = self._redis_server.smembers(
hub_api.create_rocon_gateway_key(gateway, 'flips'))
for encoded_flip in encoded_flips:
[target_gateway, name, connection_type,
node] = utils.deserialize(encoded_flip)
remote_rule = gateway_msgs.RemoteRule(
target_gateway,
gateway_msgs.Rule(connection_type, name, node))
remote_gateway.flipped_interface.append(remote_rule)
remote_gateway.pulled_interface = []
encoded_pulls = self._redis_server.smembers(
hub_api.create_rocon_gateway_key(gateway, 'pulls'))
for encoded_pull in encoded_pulls:
[target_gateway, name, connection_type,
node] = utils.deserialize(encoded_pull)
remote_rule = gateway_msgs.RemoteRule(
target_gateway,
gateway_msgs.Rule(connection_type, name, node))
remote_gateway.pulled_interface.append(remote_rule)
return remote_gateway
def get_classes_dict(serialized_file: str,
sparql_file: str,
repository: str,
endpoint: str,
endpoint_type: str,
limit: int = 1000) -> ResourceDictionary:
"""
Return a ResourceDictionary with the list of classes in the ontology
:param serialized_file: The file where the properties ResourceDictionary is serialized
:param sparql_file: The file containing the SPARQL query
:param repository: The repository containing the ontology
:param endpoint: The SPARQL endpoint
:param endpoint_type: GRAPHDB or VIRTUOSO (to change the way the endpoint is called)
:param limit: The sparql query limit
:return: A ResourceDictionary with the list of classes in the ontology
"""
classes_dictionary = deserialize(serialized_file)
if classes_dictionary:
return classes_dictionary
classes_dictionary = ResourceDictionary()
classes_sparql_query = open(sparql_file).read()
classes_sparql_query_template = Template(classes_sparql_query +
" limit $limit offset $offset ")
for class_uri in get_sparql_results(classes_sparql_query_template,
["class"], endpoint, repository,
endpoint_type, limit):
classes_dictionary.add(class_uri[0])
serialize(classes_dictionary, serialized_file)
return classes_dictionary
def get_properties_dict(serialized_file: str,
sparql_file: str,
repository: str,
endpoint: str,
endpoint_type: str,
limit: int = 1000) -> ResourceDictionary:
"""
Return a ResourceDictionary with the list of properties in the ontology
:param serialized_file: The file where the properties ResourceDictionary is serialized
:param sparql_file: The file containing the SPARQL query
:param repository: The repository containing the ontology
:param endpoint: The SPARQL endpoint
:param endpoint_type: GRAPHDB or VIRTUOSO (to change the way the endpoint is called)
:param limit: The sparql query limit
:return: A ResourceDictionary with the list of properties in the ontology
"""
global_properties_dict = deserialize(serialized_file)
if global_properties_dict:
return global_properties_dict
global_properties_dict = ResourceDictionary()
global_properties_dict.add(RDF.type)
properties_sparql_query = open(sparql_file).read()
properties_sparql_query_template = Template(
properties_sparql_query + " limit $limit offset $offset ")
for rdf_property in get_sparql_results(properties_sparql_query_template,
["property"], endpoint, repository,
endpoint_type, limit):
global_properties_dict.add(rdf_property[0])
serialize(global_properties_dict, serialized_file)
return global_properties_dict
def dataset(Dir, fontList = None):
imgs, other_info = utils.deserialize(Dir)#, fontList)
labels = np.concatenate(
[np.ones(i.shape[0], np.int32) * k for k, i in enumerate(imgs)],
0
)
imgs = np.concatenate(imgs, 0).astype(np.float32) / 255
imgs = np.expand_dims(imgs, -1)
random.seed(23333)
indices = [i for i in range(imgs.shape[0])]
random.shuffle(indices)
imgs = imgs[indices]
labels = labels[indices]
num_train = imgs.shape[0] * 3 // 5
num_valid = imgs.shape[0] // 5
train = tf.data.Dataset.from_tensor_slices(
(imgs[:num_train], labels[:num_train])
).map(
lambda img, label: (tf.image.resize_image_with_pad(
tf.random_crop(img, [48, 48, 1]),
64, 64
), label)
).shuffle(256).repeat().batch(batch_size)
valid = tf.data.Dataset.from_tensor_slices((imgs[num_train:num_train+num_valid], labels[num_train:num_train+num_valid])).repeat().batch(batch_size)
test = tf.data.Dataset.from_tensor_slices((imgs[num_train+num_valid:], labels[num_train+num_valid:])).batch(batch_size)
return train, valid, test, imgs.shape[0]*3//5
def main():
random.seed(23333)
imgs, other_infos = utils.deserialize("../data", fontList)
imgs_shuffled = []
for i in imgs:
indices = [j for j in range(i.shape[0])]
random.shuffle(indices)
imgs_shuffled.append(i[indices])
train = np.array([
utils.daisy(img) for i in imgs_shuffled for img in i[:len(i) * 4 // 5]
])
train = np.reshape(train, [train.shape[0], -1])
train_label = np.array(
[i for i, _ in enumerate(imgs) for j in _[:len(_) * 4 // 5]])
test = np.array([
utils.daisy(img) for i in imgs_shuffled for img in i[len(i) * 4 // 5:]
])
test = np.reshape(test, [test.shape[0], -1])
test_label = np.array(
[i for i, _ in enumerate(imgs) for j in _[len(_) * 4 // 5:]])
print(train.shape)
label, center, cnt = meanshift(train, .7, 20, 4e-2, uniform_weight)
center_label = np.zeros([label.ntree, len(fontList)], int)
for i in range(train.shape[0]):
center_label[label.classidx[label.find(i)], train_label[i]] += 1
tp = 0
for idx, t in enumerate(test):
lst, dist = utils.nearest_neighbour(t, center)
if np.argmax(center_label[lst[0]]) == test_label[idx]:
tp += 1
print('acc =', tp / test.shape[0])
def test_on_Jaconbsen_with_50_stages():
# # Just to document how the test DAG was created
# from equations import read_bipartite_graph
# g, eqs, forbidden = read_bipartite_graph('JacobsenILOSimpBounds')
# for eq, var in g.edges_iter(eqs):
# g[eq][var]['weight'] = 1 if (eq, var) in forbidden else 10
# mate = nx.max_weight_matching(g, maxcardinality=True)
# # Orient according to the matching, and also label
# dig = nx.DiGraph()
# for eq, var in g.edges_iter(eqs):
# if mate[eq]==var:
# dig.add_edge(eq, var, weight=1, orig_edges=[(eq,var)])
# else:
# dig.add_edge(var, eq, weight=1, orig_edges=[(var,eq)])
# assert not nx.is_directed_acyclic_graph(dig)
# from utils import serialize
# serialize(dig, 'data/JacobsenILOSimpBounds_as_DAG.pkl.gz')
dig = deserialize(DATADIR + 'JacobsenILOSimpBounds_as_DAG.pkl.gz')
# # Uncomment to prove that this graph has more than 10M simple cycles:
# cutoff = 10000000
# from itertools import islice
# n_cycles = sum(1 for _ in islice(nx.simple_cycles(dig), cutoff+1))
# if n_cycles == cutoff+1:
# print('More than', cutoff, 'simple cycles, giving up...')
# else:
# print('There are', n_cycles, 'simple cycles in total')
_, cost = solve_problem(dig)
print('Cost with ILP:', cost) # 107 with this matching; optimal tearing 53
cost, _ = run_mfes_heuristic(dig, try_one_cut=True, is_labeled=True)
print('Cost with heuristic:', cost) # 160
def upload_model(self,model_class,result):
from google.appengine.ext import db
from utils import deserialize
save = []
to_delete = []
for entity in result:
if "id" in entity:
existing_entry = model_class.get(db.Key.from_path(model_class.kind(),entity ["id"]))
if existing_entry:
to_delete.append(existing_entry) # Remove the existing entry with numeric ID
object = deserialize(model_class , entity)
save.append(object)
if len(to_delete) > 100:
db.delete(to_delete)
to_delete = []
if len(save) > 100:
db.put(save)
save = []
db.delete(to_delete)
db.put(save)
def send_message(address, command, data, response=False):
message = prepare_message(command, data)
with socket.socket(socket.AF_INET, socket.SOCK_STREAM) as s:
s.connect(address)
s.sendall(serialize(message))
if response:
return deserialize(s.recv(5000))
def load_nar_module_resources(nar_encoders_dict_path):
nar_encoders_dict = \
deserialize(nar_encoders_dict_path)
print("Read NAR label encoders dict for: {}".format(nar_encoders_dict.keys()))
return nar_encoders_dict
def handle(self):
message_bytes = self.request.recv(5000).strip()
message = deserialize(message_bytes)
command = message["command"]
data = message["data"]
logger.info(f"Received {message}")
if command == "ping":
self.respond(command="pong", data="")
if command == "tx":
try:
bank.handle_tx(data)
self.respond(command="tx-response", data="accepted")
except:
self.respond(command="tx-response", data="rejected")
if command == "block":
bank.handle_block(data)
if command == "utxos":
balance = bank.fetch_utxos(data)
self.respond(command="utxos-response", data=balance)
if command == "balance":
balance = bank.fetch_balance(data)
self.respond(command="balance-response", data=balance)
def handle(self):
message_bytes = self.request.recv(1024 * 4).strip()
message = deserialize(message_bytes)
command = message["command"]
data = message["data"]
if command == "ping":
self.respond(command="pong", data="")
if command == "block":
# If the block isn't new, ignore it
chain_index, height = node.find_block(data)
if chain_index == height == None:
logging.info(f"Received block from peer")
node.handle_block(data)
# Tell the mining thread mine the new tip
logger.info(f"in tcphandler. mempool has {len(node.mempool)}")
mining_interrupt.set()
# logging.info(f"Ignoring block: {data}")
if command == "tx":
node.handle_tx(data)
if command == "balance":
balance = node.fetch_balance(data)
self.respond(command="balance-response", data=balance)
if command == "utxos":
utxos = node.fetch_utxos(data)
self.respond(command="utxos-response", data=utxos)
def get_numeric_jacobian(problem_name):
# Always return the Jacobian as list of (i, j, J_ij)
jacfile = 'data/' + problem_name + '.jac.pkl.gz'
if isfile(jacfile):
print('Using cached Jacobian for', problem_name)
return deserialize(jacfile)
#
print('Computing then saving the Jacobian for', problem_name)
g, problem = create_dag(problem_name)
code = numeric_diff(g, problem)
globals_ = {}
try:
exec_(code, globals_)
except:
print(
'===============================================================')
print(code)
print(
'===============================================================')
raise
J = globals_['J']
#
Jrows = get_J_rowwise(problem)
jac = [(i, j, J[i, j]) for i, cols in enumerate(Jrows) for j in cols]
serialize(jac, 'data/' + problem.name + '.jac.pkl.gz')
return jac
def print_utxo(self):
utxos = []
for _, outs in self._bucket.kv.items():
outs = utils.deserialize(outs)
for out in outs:
print(out.value)
def handle(self):
message_data = self.request.recv(5000).strip()
message = deserialize(message_data)
print(f"got a message: {message}")
command = message["command"]
if command == "ping":
self.respond("pong", "")
if command == "balance":
public_key = message["data"]
balance = bank.fetch_balance(public_key)
self.respond("balance-response", balance)
if command == "utxo":
public_key = message["data"]
utxo = bank.fetch_utxo(public_key)
self.respond("utxo-response", utxo)
if command == "tx":
tx = message["data"]
try:
bank.handle_tx(tx)
self.respond("tx-response", data="accepted")
except:
self.respond("tx-response", data="rejected")
def test_packing(self):
ik, tk = KG(length=8)
u = ik.encode(b'\x0f')
serialize(ik, PACK_TEST_IK_FILE)
serialize(tk, PACK_TEST_TK_FILE)
serialize(u, PACK_TEST_CT_FILE)
ik1 = deserialize(PACK_TEST_IK_FILE)
tk1 = deserialize(PACK_TEST_TK_FILE)
u1 = deserialize(PACK_TEST_CT_FILE)
#call to_affine on all GE objects in codec
self.assertEqual(u.all_to_affine(), u1)
self.assertEqual(ik.all_to_affine(), ik1)
self.assertEqual(tk.all_to_affine(), tk1)
def run_training(n_hidden=256, ds_type='2d_star_11', n_parallel=1000, n_epochs=1000000,
random_bias=False, act='relu', n_layers=1, device_number=0, version=0):
print('Start time:', datetime.datetime.now())
if n_layers > 1:
name = f'{n_hidden}x{n_layers}-{n_parallel}-{random_bias}-{act}-v{version}'
else:
name = f'{n_hidden}-{n_parallel}-{random_bias}-{act}-v{version}'
x_train, y_train = get_2d_star_dataset(k=11, dist=0.1)
print(f'Running model for {n_epochs} epochs on dataset {ds_type}: {name}')
base_dir = Path(get_results_path())
file_dir = base_dir/ds_type/name
file_path = file_dir/'model_trainer.p'
if utils.existsFile(file_path):
print('Loading existing model')
mt = utils.deserialize(file_path)
mt.to(get_device(device_number))
else:
print('Creating new model')
mt = ModelTrainer(x_train, y_train, n_parallel=n_parallel,
hidden_sizes=[n_hidden] * n_layers, n_virtual_samples=n_hidden**2,
random_bias=random_bias, act=act, device_number=device_number, version=version)
mt.train(n_epochs)
mt.to('cpu')
utils.serialize(file_path, mt)
utils.serialize(file_dir/'config.p', dict(ds_type=ds_type, n_parallel=n_parallel, n_layers=n_layers,
random_bias=random_bias, act=act, n_epochs=n_epochs, version=version))
print('Saved trained model')
print('End time:', datetime.datetime.now())
def handle(self):
message_bytes = self.request.recv(1024 * 8).strip()
message = deserialize(message_bytes)
command = message["command"]
data = message["data"]
# logger.info(f"Received {message}")
if command == "ping":
self.respond(command="pong", data="")
if command == "tx":
try:
node.handle_tx(data)
self.respond(command="tx-response", data="accepted")
except:
self.respond(command="tx-response", data="rejected")
if command == "block":
# handle only if it builds on the tip
if data.prev_id == node.blocks[-1].id:
node.handle_block(data)
# interrupt mining thread
mining_interrupt.set()
if command == "utxos":
balance = node.fetch_utxos(data, True)
self.respond(command="utxos-response", data=balance)
if command == "balance":
balance = node.fetch_balance(data, True)
self.respond(command="balance-response", data=balance)
def handle(self):
while True:
message_data = self.request.recv(5000).strip()
# Need to break early if we got empty bytes,
# otherwise we get a deserialization error
# trying to deserialize empty bytes
if message_data == b"":
print("Closing connection, bye!")
break
message = deserialize(message_data)
print(f'Received {message}')
if message["command"] == "ping":
self.respond("pong", "")
if message["command"] == "balance":
public_key = message["data"]
balance = bank.fetch_balance(public_key)
self.respond("balance-response", balance)
if message["command"] == "utxo":
public_key = message["data"]
utxo = bank.fetch_utxo(public_key)
self.respond("utxo-response", utxo)
if message["command"] == "tx":
tx = message["data"]
try:
bank.handle_tx(tx)
self.respond("tx-response", data="accepted")
except:
self.respond("tx-response", data="rejected")
def process_result(result):
value_format = result.headers['Content-Type']
device_body = utils.deserialize(result, value_format)
utils.save_data(device_body, value_format)
print("\nHTTP Response successful!") #debug
def handle(self):
message_bytes = self.request.recv(1024 * 4).strip()
message = deserialize(message_bytes)
command = message["command"]
data = message["data"]
logger.info(f"received {command}")
if command == "ping":
self.respond(command="pong", data="")
if command == "block":
if data.prev_id == node.blocks[-1].id:
node.handle_block(data)
# Interrupt mining thread
mining_interrupt.set()
if command == "tx":
node.handle_tx(data)
if command == "balance":
balance = node.fetch_balance(data)
self.respond(command="balance-response", data=balance)
if command == "utxos":
utxos = node.fetch_utxos(data)
self.respond(command="utxos-response", data=utxos)
def __deserialise_from_split(self, name, split):
data_dir = utils.DATA_DIR
split_dir = utils.DATA_DIR + '/split-' + str(split)
utils.DATA_DIR = split_dir
data = utils.deserialize(name)
utils.DATA_DIR = data_dir
return data, split_dir
def solve_difficult_for_ilp(solve_function, log):
for filename, opt in DIFFICULT_FOR_ILP:
log(filename)
g, eqs = deserialize(DATADIR + filename)
_, _, _, tear_set, _ = solve_function(g, eqs)
cost = len(tear_set)
assert opt == cost, (opt, cost)
log()
def handle(self):
message_bytes = self.request.recv(1024 * 4).strip()
message = deserialize(message_bytes)
command = message["command"]
data = message["data"]
if command == "ping":
self.respond(command="pong", data="")
if command == "block":
if data == None:
logger.info(f"Initial block download complete")
node.syncing = False
return
logging.info(f"Received block from peer")
try:
node.handle_block(data)
mining_interrupt.set()
except:
pass
# If syncing, request next block
if node.syncing:
node.initial_block_download()
if command == "tx":
node.handle_tx(data)
if command == "balance":
balance = node.fetch_balance(data)
self.respond(command="balance-response", data=balance)
if command == "utxos":
utxos = node.fetch_utxos(data)
self.respond(command="utxos-response", data=utxos)
if command == "join":
node.peers.add(data)
self.respond(command="peers", data=node.peers)
logger.info("received join msg")
if command == "peers":
logger.info("received peer list")
if command == "get_blocks":
next_block = None
# locate the block in the main chain
# FIXME: this should call a general-purpose function
for block in node.active_chain:
if block.prev_id == data:
next_block = block
break
# Says the IBD is done
send_message(self.peer(), command="block", data=next_block)
logger.info(f"sent 'block' message: {next_block}")
def run_finer_lrs(init_param='kaiming', device='cpu'):
dist_grid = [ExampleDistribution()
] + [RadialDataDistribution(d=2**k) for k in range(7)]
std_grid = [0.1, 0.5, 1.0, 2.0]
# bi_grid = [('zero', 0.0), ('he+5', 0.0), ('he+1', 0.0), ('kink_uniform', 0.0)] \
# + [(bim, big) for big in std_grid for bim in ['normal', 'uniform']] \
# + [('pos-unif', 1.0), ('neg-unif', 1.0), ('kink-unif', 1.0), ('kink-neg-unif', 1.0),
# ('kink-neg-point', 0.0)]
bi_grid = [('zero', 0.0), ('unif', 1.0), ('unif-pos', 1.0),
('unif-neg', 1.0), ('kink-neg-unif', 1.0), ('pytorch', 1.0),
('kink-neg-point', 0.0)]
for opt in ['gd', 'gd-mom', 'adam']:
for dist in dist_grid:
d = dist.get_x_dim()
for bim, big in bi_grid:
folder_name = f'{init_param}_{opt}_{dist.get_name()}_{bim}-{big:g}'
path = Path(custom_paths.get_results_path()
) / 'nn_comparison' / folder_name
best_lr_file = Path(custom_paths.get_results_path(
)) / 'nn_comparison' / f'{folder_name}_bestlr.pkl'
if not utils.existsFile(best_lr_file):
sys.stderr.write(
'best lr file {best_lr_file} does not exist!\n')
continue
best_lr = utils.deserialize(best_lr_file)
lr_grid = [best_lr * (2**(k / 8)) for k in range(-3, 4)]
for lr in lr_grid:
print(f'Running combination {folder_name} with lr {lr:g}')
file = path / f'{lr:g}.pkl'
utils.ensureDir(file)
if utils.existsFile(file):
continue
n_rep = 2 if d == 64 else 1
trainer = SimpleParallelTrainer(n_parallel=100 // n_rep,
n_train=256 * d,
n_valid=1024,
n_test=1024,
data_distribution=dist,
lr=lr,
bias_init_gain=big,
batch_size=256,
bias_init_mode=bim,
init_param=init_param,
n_epochs=8192 // d,
seed=0,
device=device,
n_hidden=512,
opt=opt,
valid_epoch_interval=64 //
d,
n_rep=n_rep)
results = trainer.fit(do_plot=False, verbose=False)
if results is None:
print('Got NaN values')
utils.serialize(file, {
'trainer': trainer,
'results': results
})
def test_serialization(self):
serialize(MATRIX, MATRIX_TEST_SERIALIZE_FILE)
n_matrix = deserialize(MATRIX_TEST_SERIALIZE_FILE)
MATRIX.all_to_affine()
self.assertEqual(n_matrix, MATRIX)
self.assertTrue(MATRIX.all_validate())
self.assertTrue(n_matrix.all_validate())
def test_serialization(self):
serialize(VECTOR, VECTOR_TEST_SERIALIZE_FILE)
n_vector = deserialize(VECTOR_TEST_SERIALIZE_FILE)
VECTOR.all_to_affine()
self.assertEqual(n_vector, VECTOR)
self.assertTrue(VECTOR.all_validate())
self.assertTrue(n_vector.all_validate())
def send_message(command, data):
with socket.socket(socket.AF_INET, socket.SOCK_STREAM) as s:
s.connect(client_address)
message = prepare_message(command, data)
serialized_message = serialize(message)
s.sendall(serialized_message)
message_data = s.recv(5000)
message = deserialize(message_data)
print(f'Received {message}')
def send_message(command, data):
sock = socket.socket(socket.AF_INET, socket.SOCK_STREAM)
sock.connect(address)
message = serialize(prepare_message(command, data))
sock.sendall(message)
message_data = sock.recv(5000)
message = deserialize(message_data)
print(f"Received data: {message}")
return message
def from_raw(raw):
result = Message(deserialize(raw))
result._serialized_size = len(raw)
return result
def load_workflow(path):
'''Load workflow from directory path.'''
tokenizer = cPickle.load(gzip.open(path + "/tok.pkl.gz"))
scaler = cPickle.load(gzip.open(path + "/scaler.pkl.gz"))
projector = utils.deserialize(cPickle.load(gzip.open(path + "/model.univ.pkl.gz")))
return Workflow(tokenizer, projector, scaler)
