def testFindMatchingProductions(self):
# Providing no productions should result in no matches.
gen = Generator()
g = Graph()
self.assertEquals( len(gen._findMatchingProductions(g, [])), 0)
# We have a production, but the LHS can't be found in the graph.
# No solutions.
g = Graph()
g.addEdge(Vertex('g0', 'A'), Vertex('g1', 'B'))
lhs = Graph()
lhs.addEdge(Vertex('g0', 'C'), Vertex('g1', 'D'))
rhs = Graph()
p1 = Production(lhs, rhs)
gen = Generator()
self.assertEquals( len(gen._findMatchingProductions(g, [p1])), 0)
# One matching production, a simple vertex "A".
g = Graph()
g.addEdge(Vertex('g0', 'A'), Vertex('g1', 'B'))
lhs = Graph()
lhs.addVertex(Vertex('g0', 'A', '1'))
rhs = Graph()
p1 = Production(lhs, rhs)
self.assertEquals( len(gen._findMatchingProductions(g, [p1])), 1)
# Two matching productions.
g = Graph()
g.addEdge(Vertex('g0', 'A'), Vertex('g1', 'B'))
lhs = Graph()
lhs.addVertex(Vertex('g0', 'A', '2'))
rhs = Graph()
p1 = Production(lhs, rhs)
p2 = Production(lhs, rhs)
self.assertEquals( len(gen._findMatchingProductions(g, [p1, p2])), 2)
def __init__(self, template_dir, basepackage):
Generator.__init__(self,None,None,None,None)
self.basepackage = basepackage
if not template_dir:
self.template_path='./templates/pom/template.txt'
else:
self.template_path=join(template_dir,'template.txt')
def logistic_regression(n,
mx1,
vx1,
my1,
vy1,
mx2,
vx2,
my2,
vy2,
optimizer='SGD'):
"""
weights : shape=(k=3,1)
Input
-----
optimizer : 'SGD' or 'NTM'
'SGD' == 'Steepest Gradient Descent'
'NTM' == 'Newton's Method'
"""
inputs = []
labels = []
D1_label = 0.0
D2_label = 1.0
bias_term = 1.0
for _ in range(n):
# Data 1
D1x = Generator.univariate_gaussian(mx1, vx1)
D1y = Generator.univariate_gaussian(my1, vy1)
inputs.append([bias_term, D1x, D1y])
labels.append([D1_label])
# Data 2
D2x = Generator.univariate_gaussian(mx2, vx2)
D2y = Generator.univariate_gaussian(my2, vy2)
inputs.append([bias_term, D2x, D2y])
labels.append([D2_label])
inputs = Mat(inputs)
labels = Mat(labels)
# init weights
weights = Mat([[-6.0], [1.0], [-0.1]])
print('inputs shape:\t', inputs.shape)
print('labels shape:\t', labels.shape)
print('weights shape:\t', weights.shape)
# optimization
if optimizer == 'SGD':
weights = steepest_gradient_descent(weights, inputs, labels)
elif optimizer == 'NTM':
weights = newton_method(weights, inputs, labels)
else:
raise AttributeError('{} is not a valid optimizor'.format(optimizer))
# inference
logits = inference(weights, inputs)
# evaluate model
CM = ConfusionMatrix(logits, labels)
CM.show_matrix()
CM.show_accuracy()
CM.show_sensitivity()
CM.show_specificity()
def on_create_train(self):
msg = ""
if self.selectedPathes:
for p in self.selectedPathes:
if os.path.isdir(p):
self.generator = Generator(p, self.logger)
self.generator.createCSVLabelMap()
else:
self.generator = Generator(self.directory, self.logger)
self.generator.createCSVLabelMap()
def build_generators(self):
self.generator_A = Generator(
n_filters = self.generator_filters,
image_size = 128,
image_channels = 3
)
self.generator_B = Generator(
n_filters = self.generator_filters,
image_size = 128,
image_channels = 3
)
def test_get_values_for_recursive_simple_functions(self):
right_result = [
'114BT00F', '114BT0OF', '114BTO0F', '114BTOOF', '11ABT00F',
'11ABT0OF', '11ABTO0F', '11ABTOOF', '1L4BT00F', '1L4BT0OF',
'1L4BTO0F', '1L4BTOOF', '1LABT00F', '1LABT0OF', '1LABTO0F',
'1LABTOOF', 'L14BT00F', 'L14BT0OF', 'L14BTO0F', 'L14BTOOF',
'L1ABT00F', 'L1ABT0OF', 'L1ABTO0F', 'L1ABTOOF', 'LL4BT00F',
'LL4BT0OF', 'LL4BTO0F', 'LL4BTOOF', 'LLABT00F', 'LLABT0OF',
'LLABTO0F', 'LLABTOOF', '114BT3KS4B', '114BT3KSAB', '114BTEKS4B',
'114BTEKSAB', '11ABT3KS4B', '11ABT3KSAB', '11ABTEKS4B',
'11ABTEKSAB', '1L4BT3KS4B', '1L4BT3KSAB', '1L4BTEKS4B',
'1L4BTEKSAB', '1LABT3KS4B', '1LABT3KSAB', '1LABTEKS4B',
'1LABTEKSAB', 'L14BT3KS4B', 'L14BT3KSAB', 'L14BTEKS4B',
'L14BTEKSAB', 'L1ABT3KS4B', 'L1ABT3KSAB', 'L1ABTEKS4B',
'L1ABTEKSAB', 'LL4BT3KS4B', 'LL4BT3KSAB', 'LL4BTEKS4B',
'LL4BTEKSAB', 'LLABT3KS4B', 'LLABT3KSAB', 'LLABTEKS4B',
'LLABTEKSAB'
]
patterns = Pattern('|upper(leet(reverse(hobbies)))|')
tokens = patterns.get_tokens()
manipulated_result = []
for token in tokens:
# function = token.get_function()
function_and_args = token.get_function_and_args()
category = token.get_category()
generator = Generator(patterns, self.profile)
result = generator.get_values_for_simple_token(category)
while function_and_args:
cur_function, args = function_and_args.pop()
result = Utils.manipulate_list(cur_function, result)
manipulated_result += result
self.assertEqual(right_result, manipulated_result)
def run(fread, fwrite, models, factory):
gold, lexicalized, realized = [], {}, {}
for amr in utils.parse_corpus(fread, True):
print amr['sentence']
try:
gen = Generator(amr=amr['amr'],
erg_factory=factory,
models=models,
beam_n=10)
candidates = gen.run()
for i in range(10):
if i not in lexicalized:
lexicalized[i] = []
if i < len(candidates):
tree = candidates[i].tree
lexicalized[i].append(tree.realize(root=tree.root,
text=''))
# print tree.realize(root=tree.root, text=''), ' \t', tree.prettify(root=tree.root, isRule=False)
else:
lexicalized[i].append('-')
except:
print 'Error'
for i in range(10):
if i not in lexicalized:
lexicalized[i] = []
lexicalized[i].append('-')
gold.append(amr['sentence'])
write(gold, os.path.join(fwrite, 'gold'))
write(lexicalized, os.path.join(fwrite, 'lexicalized'))
def __init__(self, monet_path, photo_path, image_shape, batch_size):
# Load Datasets
self.monet_path = monet_path
self.photo_path = photo_path
self.monets = glob.glob(self.monet_path)
self.photos = glob.glob(self.photo_path)
self.num_monets = len(self.monets)
self.num_photos = len(self.photos)
self.batch_size = batch_size
# Data shape
self.image_shape = image_shape
# Instantiate models
self.generator = Generator(self.monets, self.photos, self.image_shape,
self.batch_size)
self.generator.build()
self.generator.model.compile(loss='binary_crossentropy')
print('-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=' * 3)
self.discriminator = Discriminator(self.monets, self.photos,
self.image_shape, self.batch_size)
self.discriminator.build()
self.discriminator.model.compile(loss='binary_crossentropy')
def main():
clientGenerator = Generator(UniformDistribution(8, 12))
firstQueue = []
secondQueue = []
operators = [
Operator(firstQueue, UniformDistribution(15, 25)),
Operator(firstQueue, UniformDistribution(30, 50)),
Operator(secondQueue, UniformDistribution(20, 60))
]
processors = [
Processor(firstQueue, UniformDistribution(15, 15)),
Processor(secondQueue, UniformDistribution(30, 30))
]
totalRequests = 3000
tStart = time()
res = modeling(clientGenerator, operators, processors, totalRequests)
print('time (secs)', time() - tStart)
for key in res.keys():
print(key, res[key])
print('lost', res['lost'] / totalRequests)
def testApplyProduction_Blackbox4(self):
# More complex black-box test. This time we have several productions
# in various configurations, and we build a big graph (50 vertices).
input = """
configuration {
min_vertices = 50;
}
productions {
# Start graph
A->B, A->C;
# Productions
A->C, A->B ==> A->D->C, A->B;
A->D ==> A->D->E;
D->E ==> D->F->E, D->G;
G ==> G->A->D;
}
"""
gen = Generator()
f = gen._parseGrammarFile(input)
logging.debug('start graph is...')
logging.debug(f.startGraph)
gen.applyProductions(f.startGraph, f.productions, f.config)
logging.info(f.startGraph)
self.assertEqual(f.startGraph.numVertices, 50)
def addPass():
# ask the user what catagory the password is in, and where the password is used
if PasswordManager.mstrPWStor.keys().__str__() == "dict_keys([])":
catagory = input("What do you want to call your first catagory? ")
passFor = input("Where is this password going to be used? ")
else:
while True:
catagory = input(f"What category is this password; {useful.Strings.lstToStr(PasswordManager.mstrPWStor, ', ', False)}, or enter a new name to create a new catagory? ")
passFor = input("Where is this password going to be used? ")
break
# ask the user what the password is, or generate one for them
ui = getpass("Type a password, or put a number 1 to 3 for a password with that strength to be generated, 1 being super weak, 3 being super strong: ")
while True:
useful.Terminal.clear()
if ui.isdigit():
if 3 <= int(ui) + 2 <= 5:
if not catagory in PasswordManager.mstrPWStor:
PasswordManager.mstrPWStor[catagory] = {}
randPass = Password( Generator.genPass(int(ui) + 2) )
PasswordManager.mstrPWStor[catagory][passFor] = randPass.getPass()
break
else:
passTemp = Password(ui)
if passTemp.check():
if not catagory in PasswordManager.mstrPWStor:
PasswordManager.mstrPWStor[catagory] = {}
PasswordManager.mstrPWStor[catagory][passFor] = passTemp.getPass()
break
ui = getpass("That password was insecure, type a password, or put a number 1 to 3 for a password with that strength to be generated, 1 being super weak, 3 being super strong: ")
def test(FLAGS):
sample_size = FLAGS.eval_size
z_size = FLAGS.zsize
cuda = FLAGS.cuda
g_path = FLAGS.gpath
d_path = FLAGS.dpath
map_location = 'cuda' if cuda else 'cpu'
# Load the models
dckpt = torch.load(d_path, map_location=map_location)
gckpt = torch.load(g_path, map_location=map_location)
D = Discriminator(784, 128, 1)
G = Generator(100, 32, 784)
D.load_state_dict(dckpt['state_dict'])
G.load_state_dict(gckpt['state_dict'])
# Define some latent vectors
z = np.random.uniform(-1, 1, size=(sample_size, z_size))
z = torch.from_numpy(z).float()
if cuda:
z = z.cuda()
# Eval mode
G.eval()
rand_images = G(z)
view_samples(0, [rand_images])
def __init__(self, experiment_name, vizualize, num_epochs, n_observations):
# Create Experiment name dir for records
self.experiment_name = experiment_name
self.n_observations = n_observations
self.viz = create_viz('{}_{}'.format(
name_env, self.experiment_name)) if vizualize else None
self.dataset, self.dataloader, self.device = load_dataset(
self.viz, folder_name=self.experiment_name)
self.netG = Generator(ngpu).to(self.device)
self.netD = Discriminator(ngpu).to(self.device)
self.start_epoch = self.filehandling_experiment()
self.num_epochs = num_epochs
# We create a fixed subset of random for the latent variable, this way we can evauate our progress.
self.fixed_noise = torch.randn(64, nz, 1, 1, device=self.device)
# Setup Adam optimizers for both G and D
self.optimizerD = optim.RMSprop(self.netD.parameters(), lr=lr)
self.optimizerG = optim.RMSprop(self.netG.parameters(), lr=lr)
# Fixed noise for visualisation
self.fixed_noise = torch.randn(64, nz, 1, 1, device=self.device)
def main():
client_generator = Generator(EvenDistribution(8, 12))
first_queue = []
second_queue = []
operators = [
Operator(first_queue, EvenDistribution(15,
25)), # самый производительный
Operator(first_queue, EvenDistribution(30, 50)),
Operator(second_queue,
EvenDistribution(20, 60)) # наименее производительный
]
processors = [
Processor(first_queue, EvenDistribution(15, 15)), # ровно 15 минут
Processor(second_queue, EvenDistribution(30, 30)) # ровно 30 минут
]
total_requests = 300
t_start = time()
res = modeling(client_generator, operators, processors, total_requests)
print('time seconds', time() - t_start)
for key in res.keys():
print(key, res[key])
print('lost', res['lost'] / total_requests)
def get_row_key(datetime_str, use_old_generator=True):
rk_generator = DatetimeKeyGenerator()
if use_old_generator:
rk_generator = Generator()
row_key = rk_generator.generateTimeRowKeyStart(datetime_str)
return row_key
def testApplyProduction_Blackbox3(self):
# Another black-box test. This time with a split LHS: A->B,A->C
input = """
# Grammar file for testing.
configuration {
min_vertices = 4;
}
productions {
# Start graph
A->B, A->C;
# Productions
A->C, A->B ==> A->D->C, A->B;
}
"""
gen = Generator()
f = gen._parseGrammarFile(input)
logging.debug('start graph is...')
logging.debug(f.startGraph)
gen.applyProductions(f.startGraph, f.productions, f.config)
self.assertEqual(f.startGraph.numVertices, 4)
def setUp(self):
self.data = open('namesBoys.txt', 'r')
self.min_name_length = 2
self.max_name_length = 5
self.number_of_names = 5
self.model_order = 2
self.model = Generator(self.data, self.min_name_length, self.max_name_length, self.model_order, self.number_of_names)
def testApplyProductions(self):
# Start graph already has the minimum number of vertices. Nothing done.
g = Graph()
c = {'min_vertices':0}
gen = Generator()
gen.applyProductions(g, None, c)
self.assertEqual(len(g._vertices), 0)
# No matching productions raises an error.
c = {'min_vertices':1}
self.assertRaises(RuntimeError, gen.applyProductions, g, [], c)
# When we're done, g has more at least min_vertices.
g.addEdge(Vertex('g0', 'A'), Vertex('g1', 'A'))
c = {'min_vertices':10}
# Production is A1->A2 ==> A1->A->A2
lhs = Graph()
lhs.addEdge(Vertex('l0', 'A', 1), Vertex('l1', 'A', 2))
rhs = Graph()
rhs.addEdge(Vertex('r0', 'A', 1), Vertex('r1', 'A'))
rhs.addEdge('r1', Vertex('r2', 'A', 2))
p = Production(lhs, rhs)
gen.applyProductions(g, [p], c)
logging.debug(g)
self.assertEqual(len(g._vertices), 10)
def process_torrent(self, info):
if self.log.isEnabledFor(logging.DEBUG):
self.log.debug('metainfo: %s', pprint.pformat(info))
self.generator = Generator(info, self.fileFinder, self.media_dirs,
self.dest_dir)
pieces = StringIO.StringIO(info['pieces'])
# Iterate through pieces
last_file_pos = 0
for piece in self.generator.pieces_generator():
if self.generator.torrent_corrupted:
self.log.warning('torrent corrupted: %s', info['name'])
break
# Compare piece hash with expected hash
piece_hash = hashlib.sha1(piece).digest()
# seek the offset (skip unwanted files)
if self.generator.new_candidate:
#save the actual position of pieces corresponding to the
#0th byte of any relevant file
last_file_pos = pieces.tell()
pieces.seek(
self.generator.get_last_number_of_skipped_pieces() * 20,
os.SEEK_CUR)
if piece_hash != pieces.read(20):
self.generator.corruption()
pieces.seek(last_file_pos)
# ensure we've read all pieces
if pieces.read():
self.generator.corruption()
def __init__(self):
self.dataset, self.dataloader, self.device = load_dataset()
self.netG = Generator(ngpu).to(self.device)
self.netD = Discriminator(ngpu).to(self.device)
# Initialise Weights
self.netG.apply(weights_init)
self.netD.apply(weights_init)
# define loss function
self.criterion = nn.BCELoss()
# We create a fixed subset of random for the latent variable, this way we can evauate our progress.
self.fixed_noise = torch.randn(64, nz, 1, 1, device=self.device)
# Establish convention for real and fake labels during training
self.real_label = 1
self.fake_label = 0
# Setup Adam optimizers for both G and D
self.optimizerD = optim.Adam(self.netD.parameters(),
lr=lr,
betas=(beta1, 0.999))
self.optimizerG = optim.Adam(self.netG.parameters(),
lr=lr,
betas=(beta1, 0.999))
# Fixed noise for visualisation
self.fixed_noise = torch.randn(64, nz, 1, 1, device=self.device)
def sequential_estimate(mean, variance):
"""
Sample Mean:
Xbar(n) = Xbar(n-1) + (X(n)-Xbar(n-1))/n
Sample Variance:
S(n) = (n-2)/(n-1)*S(n-1) + (X(n)-Xbar(n-1))**2/n
Avoid numerical instability
M(2,n) = sum(X-Xbar)**2
M(2,n) = M(2,n-1) + (X(n)-Xbar(n-1))*(X(n)-Xbar(n))
S(n) = M(2,n)/n-1
var(n) - M(2,n)/n
"""
estimated_mean = math.inf
estimated_variance = math.inf
last_estimated_mean = -math.inf
M = math.inf
iteration_idx = 0
while abs(estimated_mean - last_estimated_mean)>1e-4:
iteration_idx += 1
data_point = Generator.univariate_gaussian(mean, variance)
if iteration_idx==1:
estimated_mean = data_point
estimated_variance = 0.0
M = 0.0
else:
last_estimated_mean = estimated_mean
estimated_mean = estimated_mean + (data_point-estimated_mean)/iteration_idx
M = M + (data_point-last_estimated_mean)*(data_point-estimated_mean)
estimated_variance = M/(iteration_idx-1)
print('Iteration: {}'.format(iteration_idx))
print('New Data Point: {}'.format(data_point))
print('Estimated Mean: {}'.format(estimated_mean))
print('Estimated Variance: {}'.format(estimated_variance))
def main():
firstQueueGroup = [[], [], []]
secondQueueGroup = [[], []]
clientGenerator = Generator(UniformDistribution(1, 5), firstQueueGroup)
operators = [
Operator(firstQueueGroup[0], secondQueueGroup, UniformDistribution(2, 5)),
Operator(firstQueueGroup[1], secondQueueGroup, UniformDistribution(4, 8)),
Operator(firstQueueGroup[2], secondQueueGroup, UniformDistribution(10, 15))
]
Attractions = [
Attraction(secondQueueGroup[0], UniformDistribution(5, 9)),
Attraction(secondQueueGroup[1], UniformDistribution(10, 25))
]
totalVisitors = 300
tStart = time()
res = modeling(clientGenerator, operators, Attractions, totalVisitors)
print('time (secs)', time() - tStart)
for key in res.keys():
print(key, res[key])
print('lost', res['lost'] / totalVisitors)
def testDeleteMissingVertices(self):
# lhs has no vertices(!). Nothing done.
g = Graph()
lhs = Graph()
rhs = Graph()
p = Production(lhs, rhs)
gen = Generator()
gen._deleteMissingVertices(g, p, {})
self.assertEqual(len(g._vertices), 0)
# lhs has vertices, but they all appear in the rhs. Nothing done.
g.addVertex(Vertex('g0', 'A', 1))
lhs.addVertex(Vertex('l0', 'A', 1))
rhs.addVertex(Vertex('r0', 'A', 1))
p = Production(lhs, rhs)
gen._deleteMissingVertices(g, p, {'l0':'g0'})
self.assertEqual(len(g._vertices), 1)
# lhs has a vertex (A2) that don't appear in the rhs. It should be
# deleted from g.
g.addVertex(Vertex('g1', 'A', 2))
lhs.addVertex(Vertex('l1', 'A', 2))
p = Production(lhs, rhs)
self.assertEqual(len(g._vertices), 2)
gen._deleteMissingVertices(g, p, {'l0':'g0', 'l1':'g1'})
self.assertEqual(len(g._vertices), 1)
def generate(ir_path, vis_path, model_path, index, output_path = None):
ir_img = imread(ir_path) / 255.0
vis_img = imread(vis_path) / 255.0
ir_dimension = list(ir_img.shape)
vis_dimension = list(vis_img.shape)
ir_dimension.insert(0, 1)
ir_dimension.append(1)
vis_dimension.insert(0, 1)
vis_dimension.append(1)
ir_img = ir_img.reshape(ir_dimension)
vis_img = vis_img.reshape(vis_dimension)
with tf.Graph().as_default(), tf.Session() as sess:
SOURCE_VIS = tf.placeholder(tf.float32, shape = vis_dimension, name = 'SOURCE_VIS')
SOURCE_ir = tf.placeholder(tf.float32, shape = ir_dimension, name = 'SOURCE_ir')
# source_field = tf.placeholder(tf.float32, shape = source_shape, name = 'source_imgs')
G = Generator('Generator')
output_image = G.transform(vis = SOURCE_VIS, ir = SOURCE_ir)
# D1 = Discriminator1('Discriminator1')
# D2 = Discriminator2('Discriminator2')
# restore the trained model and run the style transferring
saver = tf.train.Saver()
saver.restore(sess, model_path)
output = sess.run(output_image, feed_dict = {SOURCE_VIS: vis_img, SOURCE_ir: ir_img})
output = output[0, :, :, 0]
imsave(output_path + str(index) + '.bmp', output)
def testApplyProduction(self):
# A basic test that tests all four cases: add and remove vertex,
# and add and remove edge.
# Graph starts with A->B
g = Graph()
g.addEdge(Vertex('g0', 'A'), Vertex('g1', 'B'))
g1 = g._vertices['g1']
# Production lhs matches A->B
lhs = Graph()
lhs.addEdge(Vertex('l0', 'A', 1), Vertex('l1', 'B', 1))
# Production rhs transforms that to A->C
rhs = Graph()
rhs.addEdge(Vertex('r0', 'A', 1), Vertex('r1', 'C'))
p = Production(lhs,rhs)
gen = Generator()
gen._applyProduction(g, p, {'l0':'g0','l1':'g1'})
# g has a new vertex, <v2,C>.
self.assertEqual(len(g._vertices), 2)
self.assertEqual(g._vertices['v1'].label, 'C')
# <g0,A> points to <v2,C>
self.assertEqual(len(g._edges['g0']), 1)
self.assertEqual(g._edges['g0'][0].id, 'v1')
self.assertEqual(g._vertices['v1'].label, 'C')
# <g0,A> no longer points to <g1,B>
self.assertNotIn(g1, g._edges['g0'])
# Vertex <g1,B> has been deleted.
self.assertNotIn('g1', g._vertices)
def test_evolution(self):
generator = Generator()
card = UnitCard(Unit, 1, 60, 1, 0.01, 0.01)
generator.evolution(card, 100)
self.assertEqual(2, card.max_count)
def make_model(src_vocab,
tgt_vocab,
N=6,
d_model=512,
d_ff=2048,
h=8,
dropout=0.1):
"Helper: Construct a model from hyperparameters."
c = copy.deepcopy
attn = MultiHeadedAttention(h, d_model)
ff = PositionwiseFeedForward(d_model, d_ff, dropout)
position = PositionalEncoding(d_model, dropout)
model = EncoderDecoder(
Encoder(EncoderLayer(d_model, c(attn), c(ff), dropout), N),
Decoder(DecoderLayer(d_model, c(attn), c(attn), c(ff), dropout), N),
nn.Sequential(Embeddings(d_model, src_vocab), c(position)),
nn.Sequential(Embeddings(d_model, tgt_vocab), c(position)),
Generator(d_model, tgt_vocab))
# This was important from their code.
# Initialize parameters with Glorot / fan_avg.
for p in model.parameters():
if p.dim() > 1:
nn.init.xavier_uniform_(p)
"""https://zhuanlan.zhihu.com/p/74274453
#權值初始化 Xavier均勻分佈"""
return model
def testAddNewVertices(self):
# Production rhs has no vertices, so nothing done.
g = Graph()
lhs = Graph()
rhs = Graph()
p = Production(lhs, rhs)
gen = Generator()
self.assertEqual(len(g._vertices), 0)
gen._addNewVertices(g, p, {})
self.assertEqual(len(g._vertices), 0)
# Production rhs has vertices, but they all appear in the LHS. Hence
# they aren't new and nothing is done.
lhs.addVertex(Vertex('l1', 'A', 1))
rhs.addVertex(Vertex('r1', 'A', 1))
self.assertEqual(len(g._vertices), 0)
gen._addNewVertices(g, p, {})
self.assertEqual(len(g._vertices), 0)
# rhs has one new vertex not in the lhs.
rhsMapping = {}
rhs.addVertex(Vertex('r2', 'B', 2))
self.assertEqual(len(g._vertices), 0)
gen._addNewVertices(g, p, rhsMapping)
self.assertEqual(len(g._vertices), 1)
self.assertIn('v0', g._vertices) # new vertex is v0
self.assertEqual(g._vertices['v0'].label, 'B') # with label B
self.assertEqual(g._vertices['v0'].number, 2) # with number 2
self.assertIn('r2', rhsMapping) # now appears in rhsMapping
self.assertEqual(rhsMapping['r2'], 'v0') # r2 mapped to v0 (the newly added vertex) in graph
def __init__(self):
torch.random.manual_seed(42)
self.dim_latent_g = 128
self.dim_output_g = 28 * 28
self.dim_hidden_g = 128
self.dim_output_d = 1
self.dim_input_d = 28 * 28
self.batch_size = 64
self.learning_rate = 1e-4
self.num_epochs = 50 # default for time saving on colab
self.display_freq = 50
self.d_train_iter = 2 # This is possible to have a relation with TTsUR
self.g_train_iter = 1
self.dist_latent = torch.distributions.normal.Normal(loc=0, scale=1) # Gaussian(0, 1)
self.g = Generator(self.dist_latent, self.dim_latent_g, self.dim_hidden_g, self.dim_output_g).cuda()
self.d = Discriminator(self.dim_input_d, self.dim_output_d).cuda()
self.transforms = transforms.Compose(
[transforms.ToTensor(), transforms.Normalize((0.5,), (0.5,)), ]
)
self.dataset = datasets.MNIST(root="dataset/", transform=self.transforms, download=True)
self.loader = DataLoader(self.dataset, batch_size=self.batch_size, shuffle=True)
# TTsUR => different learning rates
self.optimizer_G = optim.Adam(self.g.parameters(), lr=self.learning_rate / 2)
self.optimizer_D = optim.Adam(self.d.parameters(), lr=self.learning_rate * 2)
self.criterion = nn.BCEWithLogitsLoss() # This changed a little bit some loss values
def test_disc_loss(max_tests=10):
z_dim = 64
gen = Generator(z_dim).to(device)
gen_opt = torch.optim.Adam(gen.parameters(), lr=lr)
disc = Discriminator().to(device)
disc_opt = torch.optim.Adam(disc.parameters(), lr=lr)
num_steps = 0
for real, _ in dataloader:
cur_bath_size = len(real)
real = real.view(cur_bath_size, -1).to(device)
# Zero out the gradient before backpropagation
disc_opt.zero_grad()
disc_loss = get_disc_loss(gen, disc, criterion, real, cur_bath_size,
z_dim, device)
assert (disc_loss - 0.68).abs() < 0.05
# Update the gradients
disc_loss.backward(retain_graph=True)
assert gen.gen[0][0].weight.grad is None
old_weight = disc.disc[0][0].weight.data.clone()
# Updata optimizer
disc_opt.step()
new_weight = disc.disc[0][0].weight.data
assert not torch.all(torch.eq(old_weight, new_weight))
num_steps += 1
if num_steps >= max_tests:
break
def test_menu(self):
generator = Generator()
space = MagicMock()
generator.init_levels()
generator.menu(space)
self.assertTrue(space.clear.called)
def __init__(self, name, head):
Generator.__init__(self, name, head)
self.stimulusTimes = []
self.currentStimulus = 0
def __init__(self, name, head, position=[0, 0, 0], mean=0, stddev=0):
Generator.__init__(self, name, head, position)
self.mean = mean
self.stddev = stddev
def __init__(self):
Generator.__init__(self, NameGenerator.filenames)
output_file = args['--output']
ast = None
try:
with open(input_file, 'r') as f:
Parser.build("Program")
source = f.read()
ast = Parser.parse(source)
except IOError:
print 'Error opening file %s. Please check the file or ' \
'the directory.' % input_file
sys.exit(1)
if ast is None:
error_list = list(Parser.error_list.keys())
error_list.sort()
for key in error_list:
sys.stdout.write(Parser.error_list[key])
sys.stdout.flush()
sys.exit(-1)
if args['-O']:
ast = Optimizer.optimize(ast)
try:
with open(output_file, 'w') as f:
f.write(Generator.generate(ast))
except IOError:
print 'Error writing to file %s. Please check the file or ' \
'the directory.' % output_file
sys.exit(1)
def __init__(self, name, head, position = [0,0,0], frequency = 2, phaseShift = 0):
Generator.__init__(self, name, head, position)
self.frequency = frequency
self.phaseShift = phaseShift
self.currentPhase = self.phaseShift
self.lastTimePoint = 0
