def setup(cls):
def multipleOf4Criterion(number):
return number % 4 == 0
def multipleOf8Criterion(number):
return number % 8 == 0
cls.generatorA = Generator(65, 16807, criterion=multipleOf4Criterion)
cls.generatorB = Generator(8921, 48271, criterion=multipleOf8Criterion)
def generateNewImage(self):
if (self.dataset.get() == 1):
path = "models/faces/celeba" + str(self.num.get() * 10) + ".model"
firstModel = torch.load(path)
netG = Generator(ngpu).to(self.device)
netG.load_state_dict(firstModel["generator_state_dict"])
netG.eval()
fake1 = netG(self.noise).detach().cpu()
array = np.transpose(
vutils.make_grid(fake1, padding=2, normalize=True),
(1, 2, 0)).numpy()
im = np.array(
Image.fromarray((array * 255).astype(np.uint8)).convert('RGB'))
self.img = ImageTk.PhotoImage(image=Image.fromarray(im))
self.imageMatrix.configure(image=self.img)
self.imageMatrix.image = self.img
self.imageMatrix.pack()
elif (self.dataset.get() == 2):
path = "models/landscapes/scenery" + str(self.num.get()) + ".model"
firstModel = torch.load(path)
netG = Generator(ngpu).to(self.device)
netG.load_state_dict(firstModel["generator_state_dict"])
netG.eval()
fake1 = netG(self.noise).detach().cpu()
array = np.transpose(
vutils.make_grid(fake1, padding=2, normalize=True),
(1, 2, 0)).numpy()
im = np.array(
Image.fromarray((array * 255).astype(np.uint8)).convert('RGB'))
self.img = ImageTk.PhotoImage(image=Image.fromarray(im))
self.imageMatrix.configure(image=self.img)
self.imageMatrix.image = self.img
self.imageMatrix.pack()
elif (self.dataset.get() == 3):
path = "models/art/art" + str(self.num.get() * 100) + ".model"
firstModel = torch.load(path)
netG = Generator(ngpu).to(self.device)
netG.load_state_dict(firstModel["generator_state_dict"])
netG.eval()
fake1 = netG(self.noise).detach().cpu()
array = np.transpose(
vutils.make_grid(fake1, padding=2, normalize=True),
(1, 2, 0)).numpy()
im = np.array(
Image.fromarray((array * 255).astype(np.uint8)).convert('RGB'))
self.img = ImageTk.PhotoImage(image=Image.fromarray(im))
self.imageMatrix.configure(image=self.img)
self.imageMatrix.image = self.img
self.imageMatrix.pack()
def lanzar_dados(dado1, jugador, dado2, total, n):
h = 0
n = n
for i in range(n):
dado1.append(
Generator(m=i * 5 + 2, low=1, high=6, rounded=True).throw())
dado2.append(Generator(m=9 + i, low=1, high=6, rounded=True).throw())
if i > 0:
while dado1[i] + dado2[i] in total:
h += 1
dado1[i] = Generator(a=h, low=1, high=6, rounded=True).throw()
dado2[i] = Generator(c=h + 4, low=1, high=6,
rounded=True).throw()
total.append(dado1[i] + dado2[i])
return total, dado1, dado2
def showImages(path):
device = torch.device("cuda:0" if (
torch.cuda.is_available() and ngpu > 0) else "cpu")
fixed_noise = torch.randn(64, nz, 1, 1, device=device)
#total = torch.load(path)
firstModel = torch.load(".art0.model")
print(firstModel["generator_state_dict"])
print("______________________________________________________________")
netG = Generator(ngpu).to(device)
netG.load_state_dict(firstModel["generator_state_dict"])
netG.eval()
fake1 = netG(fixed_noise).detach().cpu()
firstModel = torch.load(".art1.model")
print(firstModel["generator_state_dict"])
netG.load_state_dict(firstModel["generator_state_dict"])
netG.eval()
fake2 = netG(fixed_noise).detach().cpu()
plt.figure(figsize=(15, 15))
plt.subplot(1, 2, 2)
plt.axis("off")
plt.title("Fake 1")
plt.imshow(
np.transpose(vutils.make_grid(fake1, padding=2, normalize=True),
(1, 2, 0)))
plt.subplot(1, 2, 1)
plt.axis("off")
plt.title("Fake2")
plt.imshow(
np.transpose(vutils.make_grid(fake2, padding=2, normalize=True),
(1, 2, 0)))
plt.show()
def __init__(self, input_size, word_vec_dim, hidden_size, output_size,
n_layers=4, dropout_p=.2):
self.input_size = input_size
self.word_vec_dim = word_vec_dim
self.hidden_size = hidden_size
self.output_size = output_size
self.n_layers = n_layers
self.dropout_p = dropout_p
super(Seq2Seq, self).__init__()
self.emb_src = nn.Embedding(input_size, word_vec_dim)
self.emb_dec = nn.Embedding(output_size, word_vec_dim)
self.encoder = Encoder(word_vec_dim, hidden_size,
n_layers=n_layers,
dropout_p=dropout_p)
self.decoder = Decoder(word_vec_dim, hidden_size,
n_lyaers=n_layers,
dropout_p=dropout_p)
self.attn = Attention(hidden_size)
self.concat = nn.Linear(hidden_size*2, hidden_size)
self.tanh = nn.Tanh()
self.generator = Generator(hidden_size, output_size)
def __init__(self, scene):
self.scene = scene
self.scene.onKeyboard = self.onKeyBoard
self.background1 = Background("Images/background/background.png", 0)
self.background1.locate(self.scene, self.background1.x, 0)
self.background1.show()
self.background2 = Background("Images/background/background.png", 1280)
self.background2.locate(self.scene, self.background2.x, 0)
self.background2.show()
self.backgroundTimer = BackgroundTimer(0.01, self.scene,
self.background1,
self.background2)
self.backgroundTimer.start()
self.user = User("Images/character/user/run/user_1.png", self.scene,
0.9, self)
self.user.locate(self.scene, self.user.x, self.user.y)
self.user.show()
def up_background_velocity():
self.backgroundTimer.velocity += 1
def down_background_velocity():
self.backgroundTimer.velocity = 5
self.user.up_background_velocity = up_background_velocity
self.user.down_background_velocity = down_background_velocity
self.generator = Generator(self.scene, self.user)
def generate(outputDir, db, kernels, libxsmmGenerator, architecture, prefix=''):
Expr.analyseKernels(db, kernels)
for matrixInfo in db.itervalues():
matrixInfo.generateMemoryLayout(architecture, alignStartrow=matrixInfo.leftMultiplication)
for prototype in kernels:
prototype.kernel.generateMemoryLayout(architecture, alignStartrow=True)
print('\nKernels')
print('-------')
for prototype in kernels:
print(u'{}: {}'.format(prototype.name, prototype.kernel.symbol))
print('\nMemory layout')
print('-------------')
keys = db.keys()
keys.sort(key=lambda s: s.lower())
for key in keys:
name = db[key].name
for block in db[key].blocks:
print('{:16} {}'.format(name, block))
name = ''
generator = Generator.Generator(db, libxsmmGenerator, architecture, prefix)
generator.generateKernels(outputDir, kernels)
generator.generateInitializer(outputDir)
generator.generateUnitTests(outputDir, kernels)
def init_static_dialog_agent(args) :
print "reading in Ontology"
ont = Ontology.Ontology(sys.argv[1])
print "predicates: " + str(ont.preds)
print "types: " + str(ont.types)
print "entries: " + str(ont.entries)
print "reading in Lexicon"
lex = Lexicon.Lexicon(ont, sys.argv[2])
print "surface forms: " + str(lex.surface_forms)
print "categories: " + str(lex.categories)
print "semantic forms: " + str(lex.semantic_forms)
print "entries: " + str(lex.entries)
print "instantiating Feature Extractor"
f_extractor = FeatureExtractor.FeatureExtractor(ont, lex)
print "instantiating Linear Learner"
learner = LinearLearner.LinearLearner(ont, lex, f_extractor)
print "instantiating KBGrounder"
grounder = KBGrounder.KBGrounder(ont)
load_parser_from_file = False
if len(args) > 4 :
if args[4].lower() == 'true' :
load_parser_from_file = True
if load_parser_from_file :
parser = load_model('static_parser')
grounder.parser = parser
grounder.ontology = parser.ontology
else :
print "instantiating Parser"
parser = Parser.Parser(ont, lex, learner, grounder, beam_width=10, safety=True)
print "instantiating Generator"
generator = Generator.Generator(ont, lex, learner, parser, beam_width=sys.maxint, safety=True)
print "instantiating DialogAgent"
static_policy = StaticDialogPolicy.StaticDialogPolicy()
A = StaticDialogAgent(parser, generator, grounder, static_policy, None, None)
if not load_parser_from_file :
print "reading in training data"
D = A.read_in_utterance_action_pairs(args[3])
if len(args) > 4 and args[4] == "both":
print "training parser and generator jointly from actions"
converged = A.jointly_train_parser_and_generator_from_utterance_action_pairs(
D, epochs=10, parse_beam=30, generator_beam=10)
else:
print "training parser from actions"
converged = A.train_parser_from_utterance_action_pairs(
D, epochs=10, parse_beam=30)
print "theta: "+str(parser.learner.theta)
save_model(parser, 'static_parser')
return A
def __init__(self, dimensions=(100, 100)):
self.dimensions = dimensions
self.layers = {}
self.generator = Generator.Generator()
for layer_name in self.generator.generation_order:
self.add_layer(layer_name)
def gird_model(name,
day,
timeDs,
engine,
priceRatio=0.2,
geneRatio=0.2,
loadRatio=0.1):
#pd.set_option('display.max_columns', None)
#pd.set_option('display.max_rows', None)
res = []
generator = Generator(name, day, engine, frash_ratio=geneRatio)
load = Load(name, day, engine, frash_ratio=loadRatio)
storage = Storage(name, engine)
pricetable = Para_PriceTable(day, engine, frash_ratio=priceRatio)
dayplan = Para_DayAheadPlan(day, engine)
timenow = datetime(int(day[0:4]), int(day[5:7]), int(day[8:10]))
timeD = timedelta(minutes=timeDs)
while timenow < datetime(int(day[0:4]), int(day[5:7]), int(
day[8:10])) + timedelta(days=1):
print('Do:', timenow, '-------------------------------------------')
policy = gird_model_dynamic(timenow, storage, load, generator,
pricetable, dayplan)
log = Do_policy(timenow, timeD, policy,
pricetable.get_PP_bytime_buy(timenow).price,
pricetable.get_PP_bytime_sale(timenow).price, storage,
load, generator, dayplan)
res.append(log.tolist())
timenow += timeD
generator.refrash(timenow)
load.refrash(timenow)
pricetable.refrash(timenow)
return pd.DataFrame(res, columns=Para_DoLog.get_names())
def generate_noisy_dataset():
Gen = Generator()
Gen.load_params(generator_weights)
contexts = pickle.load(open("contexts",'r'))
for c in contexts.iterkeys():
contexts[c] = Gen.res_gen([c])
return contexts
def norepalgo():
contador1 = 0
contador1S = 0
contador2 = 0
contador2S = 0
contador3 = 0
contador3S = 0
contador4 = 0
contador4S = 0
contador5 = 0
contador5S = 0
contador6 = 0
contadorG = 0
rarrayt = []
i = 0
j = 3
for n in range(19):
while contadorG != 19:
i += 1
j += 1
r = Generator(a=i, c=j / 2, low=1, high=6, rounded=True).throw()
if r == 1 and contador1 != 4:
contador1 += 1
rarrayt.append(r + contador1S)
contador1S += 10
elif r == 2 and contador2 != 4:
contador2 += 1
rarrayt.append(r + contador2S)
contador2S += 10
elif r == 3 and contador3 != 4:
contador3 += 1
rarrayt.append(r + contador3S)
contador3S += 10
elif r == 4 and contador4 != 3:
contador4 += 1
rarrayt.append(r + contador4S)
contador4S += 10
elif r == 5 and contador5 != 3:
contador5 += 1
rarrayt.append(r + contador5S)
contador5S += 10
elif r == 6 and contador6 != 1:
contador6 += 1
rarrayt.append(r)
contadorG = contador1 + contador2 + contador3 + contador4 + contador5 + contador6
print("Arreglo global de los terrenos del tablero")
return (rarrayt)
print("")
def __init__(self,
input_shape,
depth_layers_discriminator=[64, 128, 256, 512],
depth_layers_generator=[1024, 512, 256, 128],
dim_noise=100,
model="simple",
data="MNIST",
flip_discri_labels=False,
final_generator_activation="tanh",
test_name='_1'):
"""DCGAN model (for each parameter, the default value is the value used in the DCGAN paper)
:param input_shape: format [height, width, depth]
:param depth_layers_discriminator: the depth of the different layers used by the discriminator, only for the
dcgan models
:param depth_layers_generator: the depth of the different layers used by the generator, only for the
dcgan models
:param dim_noise: size of the input noise used by the generator
:param model: type of model to use (simple, intermediate, dcgan_custom, dcgan_vanilla)
:param data: dataset used
:param flip_discri_labels: flip labels in the computation of the discrimination loss (10% flip)
:param final_generator_activation: activation function for the output layer of the generator
:param test_name: to give a name to the current execution"""
#Saving param
self.test_name = test_name
# Global model
self.model = model
self.data = data
# Dimension of data
self.output_height = input_shape[0]
self.output_width = input_shape[1]
self.output_depth = input_shape[2]
self.dim_noise = dim_noise
# Useful variables
self.real_images_probabilities = 0
self.fake_images_probabilities = 0
# Build input variables
#self.X_batch = tf.placeholder(dtype=tf.float32, shape=[None, self.output_depth*self.output_height*self.output_width], name='X')
self.X_batch = tf.placeholder(dtype=tf.float32,
shape=[
None, self.output_height,
self.output_width, self.output_depth
],
name='real_images')
self.noise_batch = tf.placeholder(dtype=tf.float32,
shape=[None, self.dim_noise],
name='noise')
# Build both components
self.final_generator_activation = final_generator_activation
self.discriminator = Discriminator(input_shape,
depth_layers_discriminator,
model=model)
self.generator = Generator(input_shape,
depth_layers=depth_layers_generator,
model=model,
data=data,
final_activation=final_generator_activation)
# Construct the graph
self.build_graph(flip_discri_labels=flip_discri_labels)
def _generate_cpp(self , header):
filename = self.filename + ".cpp"
cppfile = Generator.Generator(filename)
self._write_author( self.cpp , cppfile )
cppfile.Write('#include "{0}"\n'.format(header))
self._write_using_namespace(cppfile)
self._write_constructor(cppfile)
self._write_deconstructor(cppfile)
cppfile.Destory()
def start():
global dataset
# THIS IS THE DATA THAT IS RETRIEVED IN DATASET
dataset = DataR.DatasetRetrieval()
dataset = dataset.retrieveImages() # first half is orig images, 2nd half is pixelated images
print('Loaded ', dataset[0].shape, dataset[0].shape[1:], " Image sizes")
# Image shape is 96 x 96 x 3 in this dataset
image_shape = dataset[0].shape[1:]
# define descriminator model
descrim_model = Discriminator.Discriminator(image_shape)
descrim_model= descrim_model.define_discriminator()
# Define generator model
gen_model = Generator.Generator((32,32,3))
gen_model= gen_model.define_gen()
# GAN MODEL IMPLEMENTS BOTH GENERATOR AND DESCRIMINATOR INSIDE
gan_model = define_GAN(gen_model, descrim_model,image_shape)
n_patch = descrim_model.get_output_shape_at(1)[1] # size 1
n_batches= dataset[0].shape[0]
# unpack dataset
train_hr, train_lr = dataset
# num of batches per epoch
####################################
#
# Train Discriminator...
#
#####################################
bat_per_epo = int(len(train_hr) / 1)
# Calculates total iterations needed based on epochs (100 epochs)
n_steps = bat_per_epo * 1000 #100,000 iterations...
# iterate through each epoch through steps
for i in range(n_steps):
# retrieve real samples
X_real_hr, X_real_lr,real_y1= generate_real_samples(n_batches, n_patch)
# generate fake images
X_fakeB,fake_y = Generator.generateFakeSamples(gen_model, X_real_lr, n_patch,)
#X_real_hr = (X_real_hr + 1) / 2.0
#X_real_lr = (X_real_lr + 1) / 2.0
#X_fakeB = (X_fakeB + 1) / 2.0
# Loss function of first set of real images
_,d_loss_real = descrim_model.train_on_batch(X_real_hr,real_y1)
# Loss function for fake images
_,d_loss_fake = descrim_model.train_on_batch(X_fakeB,fake_y)
d_loss= 0.5 * np.add(d_loss_real,d_loss_fake)# d_loss[0]--> shape(2,)
_,g_loss,_= gan_model.train_on_batch(X_real_lr, [X_real_hr,real_y1]) #in_src,[gen_out,dis_out] model
# Loss functions printed out
print('>%d, dreal[%.4f], dfake[%.4f], g[%.4f]' % (i + 1, d_loss_real, d_loss_fake,g_loss))
# save data after epoch
if (i + 1) % (200) == 0:
summarize_performance(i, gen_model)
class Tester():
minSup = 2
delta = 5
upperDelta = 20
size = 20
nEvents = 6
debug = False
gen = Generator(42)
eventSeq = gen.generate(size, nEvents)
print('Event sequence:')
print(eventSeq)
bf = BruteForce(eventSeq, minSup, delta, upperDelta, debug)
t1 = time.time()
F1 = bf.apply()
t2 = time.time()
print('F1')
print(F1)
m = Meselo(eventSeq, minSup, delta, upperDelta, debug)
t3 = time.time()
F2 = m.apply()
t4 = time.time()
print('F2')
print(F2)
print("Comparison:" + str(F1 == F2))
btime = (t2 - t1)
mtime = (t4 - t3)
print('btime:' + str(btime) + "VS" + 'mtime' + str(mtime))
print('Other small custom test...')
customSeq = gen.generateCustom()
print('CustomSequence:')
print(customSeq)
bf2 = BruteForce(customSeq, 1, 3, 3, debug)
m2 = Meselo(customSeq, 1, 3, 3, debug)
t5 = time.time()
F3 = bf2.apply()
t6 = time.time()
print('F3')
print(F3)
t7 = time.time()
F4 = m2.apply()
t8 = time.time()
print('F4')
print(F4)
btime2 = (t6 - t5)
mtime2 = (t8 - t7)
print('btime:' + str(btime2) + "VS" + 'mtime' + str(mtime2))
print('Comparison:' + str(F3 == F4))
# print(eventSeq)
print('end........')
def one_shot(name):
global dims,n_layers
r_net = net.FFNN(n_layers)
g_net = net.FFNN(n_layers)
b_net = net.FFNN(n_layers)
r_build = gen.Generator(dims,r_net)
g_build = gen.Generator(dims,g_net)
b_build = gen.Generator(dims,b_net)
colours = [r_build,g_build,b_build]
mapping = []
for x in colours:
x.genEnv(0)
mapping.append(x.getEnv())
main = scene.Draw(dims,mapping)
main.save_image(f"Images//PerlinNoise{name}","PNG",main.res)
def _generate_cpp(self , header):
filename = self.filename + ".cpp"
cppfile = Generator.Generator(filename)
self._write_author( self.cpp , cppfile )
cppfile.Write('#include "gtest/gtest.h"\n'.format(header))
cppfile.Write('#include "{0}.h"\n'.format(self.cpp.name))
cppfile.Write('#include "{0}.h"\n'.format(header))
self._write_using_namespace(cppfile)
self._write_gtest(cppfile)
cppfile.Destory()
def Adveraisl_training(G, g_op, D, d_op, d_loader):
G_beta = Generator(embed_size,
hidden_size,
vocab_size,
num_layers,
use_cuda=cuda)
restore(G_beta, G_path)
log = open('save/log.txt', 'w')
print('start Adveraisl_training!')
for i in range(epochs):
#train G
start = time.time()
buffer = 'start epoch{}'.format(i + 1)
print(buffer)
log.write(buffer)
for _ in range(g_step):
g_op.zero_grad()
#Frist generate a whole seqence
samples, preds = G.generate_sample(batch_size, seq_len)
#Then get reward step by step
rewards = get_rewards(G_beta, D, samples, rollout_num)
loss = G.PG_loss(samples, preds, rewards)
buffer = 'reward = {:.3f} PGloss={:.3f}'.format(
(rewards.mean()).data.item(), (loss).data.item())
print(buffer)
log.write(buffer)
loss.backward()
clip_grad_norm_(G.parameters(), max_norm=max_norm)
g_op.step()
#updata G_bata
updataG_beta(G_beta, G, update_rate)
#teain D
for _ in range(d_step):
G.get_negsample(neg_path, neg_num, seq_len)
d_loader.create_batch(pos_num, batch_size)
for r in range(3):
d_loader.reset_pointer()
losses = 0
for j in range(d_loader.batch_num):
d_op.zero_grad()
sentence, labels = d_loader.next_batch()
loss = D.loss_fn(sentence, labels)
losses += loss
loss.backward()
d_op.step()
loss = test(G, g_loader)
buffer = 'test loss={:.3f}'.format((loss))
log.write(buffer)
print('epoch{}Done! cost{:.2f}s'.format(i + 1, time.time() - start))
save(D, D_path)
save(G, G_path)
def main():
with tf.Graph().as_default():
with tf.device("/gpu:1"):
session_conf = tf.ConfigProto(
allow_soft_placement=FLAGS.allow_soft_placement,
log_device_placement=FLAGS.log_device_placement)
sess = tf.Session(config=session_conf)
with sess.as_default(), open(log_precision, "w") as log, open(
loss_precision, "w") as loss_log:
DIS_MODEL_FILE = "model/irgan_eval_pre-trained.model" # overfitted DNS
param = pickle.load(open(DIS_MODEL_FILE, "rb")) #add
loss_type = "pair"
with tf.name_scope('discriminator_setting') as scope:
discriminator = Discriminator.Discriminator(
sequence_length=FLAGS.max_sequence_length,
batch_size=FLAGS.batch_size,
vocab_size=len(vocab),
embedding_size=FLAGS.embedding_dim,
filter_sizes=list(
map(int, FLAGS.filter_sizes.split(","))),
num_filters=FLAGS.num_filters,
learning_rate=FLAGS.learning_rate,
l2_reg_lambda=FLAGS.l2_reg_lambda,
# embeddings=embeddings,
embeddings=None,
paras=param,
loss=loss_type)
with tf.name_scope('generator_setting') as scope:
generator = Generator.Generator(
sequence_length=FLAGS.max_sequence_length,
batch_size=FLAGS.batch_size,
vocab_size=len(vocab),
embedding_size=FLAGS.embedding_dim,
filter_sizes=list(
map(int, FLAGS.filter_sizes.split(","))),
num_filters=FLAGS.num_filters,
learning_rate=FLAGS.learning_rate * 0.1,
l2_reg_lambda=FLAGS.l2_reg_lambda,
# embeddings=embeddings,
embeddings=None,
paras=param,
loss=loss_type)
sess.run(tf.global_variables_initializer())
tw = tf.summary.FileWriter("log_dir", graph=sess.graph) #add
evaluation(sess, discriminator, log, 0)
'''
def __init__(self, master, controller):
# parameters that you want to send through the Frame class.
Frame.__init__(self, master)
self.controller = controller
#reference to the master widget, which is the tk window
self.master = master
self.device = torch.device("cuda:0" if (
torch.cuda.is_available() and ngpu > 0) else "cpu")
self.noise = torch.randn(64, nz, 1, 1, device=self.device)
self.model = Generator(ngpu).to(self.device)
self.dataset = IntVar()
self.num = IntVar()
self.init_window()
def __init__(self,item,customer):
self.item = item
self.customer = customer
self.seller = Seller(item,customer)
self.customer_sequence = []
self.generator = Generator(customer)
self.revenue = {}
#self.revenue_upper_bound = {}
#只需要获取最后一天的upper_bound
self.revenue_upper_bound=0
self.time_length = 0
self.IB_function_type = ''
self.IB_function_type_sequence = []
self.customer_choose_mode = ''
self.log = {}
def main():
''' Iterate over all .txt files in the directory. This is useful for
making a script that periodically scans all feeds for updates, without
opening a new browser for each feed '''
for feed in os.listdir("."):
if feed.endswith(".txt"):
input_file = open(feed, "r")
information = input_file.readlines()
''' Strip newline from every line of input '''
i = 0
info_len = len(information)
while i < info_len:
information[i] = information[i].rstrip('\n')
i += 1
''' Grab the information to send to the generator '''
title = information[0]
link = information[1]
description = information[2]
browser.get(link)
title_selector = information[3]
link_selector = information[4]
description_selector = information[5]
generator = Generator(browser, title, link, description)
generator.find_new_articles(title_selector, link_selector,
description_selector)
if generator.error_no_new_articles() is True:
continue
if os.path.isfile(generator.get_filename()) is False:
generator.generate_new_feed()
else:
generator.append_feed()
generator.archive_new_feeds()
input_file.close()
browser.quit()
def uppercase():
url = request.GET.get('s', default=None)
if url is not None:
#url = "http://www.apple.com/legal/internet-services/itunes/us/terms.html"
sc = Scrapper.Scrapper()
obj = sc.scrap(url)
#print obj
#obj = json.dumps({'privacy':{'p':[{'name':'first para in privacy'},{'name':'second para in privacy'}] },
# 'taxes':{'p':[{'name':'first para in taxes'},{'name':'second para in taxes'}] }
#})
obj = Analysis.Analysis().analysis(obj)
#obj = json.dumps({'privacy':{'classify':'computer software','p':[{'name':'first para in privacy','summarizer':'example of summarized paragraph','tags':['tag1','tag2','tag3']},{'name':'second para in privacy','summarizer':'example of summarized paragraph','tags':['tag1','tag2','tag3']}] },
# 'taxes':{'classify':'computer software','p':[{'name':'first para in taxes','summarizer':'example of summarized paragraph','tags':['tag1','tag2','tag3']},{'name':'second para in taxes','summarizer':'example of summarized paragraph','tags':['tag1','tag2','tag3']}] }
#})
obj = Generator.Generator().generate(obj)
return json.dumps(obj, indent=4)
def _genator_header(self):
filename = self.filename + ".h"
hfile = Generator.Generator(filename)
self._write_author(self.cpp, hfile)
defineHeader = self.filename.upper()
hfile.Write("#ifndef __{0}_H__\n".format(defineHeader))
hfile.Write("#define __{0}_H__\n".format(defineHeader))
self._write_include(hfile)
self._write_namespace_begin(hfile)
hfile.Write("{0}class {1} {2}\n".format(self.format, self.filename,
self.extend))
hfile.Write("{0}{1}\n".format(self.format, '{'))
hfile.Write("{0}public:\n".format(self.format))
self._write_constuctor(hfile)
self._write_deconstructor(hfile)
hfile.Write("{0}{1};\n".format(self.format, '}'))
self._write_namespace_end(hfile)
hfile.Write("#endif\n".format())
hfile.Destory()
return filename
def open_map(self):
# Clear all map tiles
for i in range(15, 20):
self.image_tiles[i]["tilesImages"] = []
# Decides whether to generate a new map
new = raw_input("Generate New map (Y/N): ")
if new == "Y" or new == "y":
g = Generator.Generator(self.TILE_SIZE, self.image_tiles)
name = g.generate_maps()
self.parse_data_file(name)
else:
while True:
name = raw_input("map Name? ")
if os.path.exists(name):
break
else:
print "map doesn't exist"
self.parse_data_file(name)
self.build_tiles()
def setup(cls):
cls.generatorA = Generator(65, 16807)
cls.generatorB = Generator(8921, 48271)
def PatternTheory(inputFile, semanticBondPath, topK, outFile):
BondID = count(0)
genID = count(0)
localSwapSpace = {}
topKLabel = 0
priorScale = 10 #changed from 100 to 10 on 3/28/2020 7:29 PM
#Load Feature labels and create feature generators
equivalence = {}
feature_generators = []
with open(inputFile) as f:
for featFile in f:
feat = str.split(featFile.replace('\n', ''))
feat = str.split(featFile.replace('/home/saakur/Desktop/', './'))
equivalence = {}
with open(feat[1]) as fl:
for line in fl:
l = str.split(line.replace('\n', ''))
equivalence[l[0]] = float(l[1])
if topKLabel > 0:
sorted_equiv = sorted(equivalence.items(),
key=operator.itemgetter(1),
reverse=True)
equivalence = {}
for k, v in sorted_equiv[:topKLabel]:
equivalence[k] = v * supBondWeight
G = gen.Generator(next(genID), feat[0], feat[0], "Feature",
feat[0])
supBG = sup_b.SupportBond(next(BondID), "SupportBond", "OUT",
G.generatorID)
supBG.compatible = deepcopy(equivalence)
G.addOutBond(supBG)
feature_generators.append(G)
localSwapSpace[G.feature] = []
filelist = [f for f in os.listdir(semanticBondPath) if f.endswith(".txt")]
semBondDict = {}
#Load semantic bonds
for semanticBondFile in filelist:
semBondName = str.split(
str.split(semanticBondFile.replace('\n', ''), '_').pop(), '.')[0]
if semBondName not in semBondDict.keys():
semBondDict[semBondName] = {}
equivalence = {}
with open(semanticBondPath + semanticBondFile) as fl:
for line in fl:
semBondDict_Concept = {}
l = str.split(line.replace('\n', ''), ',')
label = l.pop(0)
label = str.split(label, '-')[0]
for l1 in l:
l2 = str.split(l1.replace('\n', ''), ':')
if (semBondName != "Similarity"):
semBondDict_Concept[l2[0]] = float(
l2[1]) * priorScale * semBondWeight
else:
semBondDict_Concept[l2[0]] = float(
l2[1]) * semBondWeight * priorScale
equivalence[label] = dict(semBondDict_Concept.copy())
semBondDict[semBondName] = dict((equivalence.copy()))
#For each of the feature generators, create generators for each of the label possibilties
# and bonds for each of the generators
for g in feature_generators:
#Determine the label category to create generators for them
genType = g.generatorName
#For each of the outbound bonds of the feature generators, create the generators for each of the label candidates
for bondID in g.outBonds.keys():
b = g.outBonds[bondID]
#For each of the each of the label candidates, create the generators and bonds
for label in b.compatible:
#Create the generator
G = gen.Generator(next(genID), label, label, genType,
g.generatorName)
#Create and add complementary support bond
supBG_IN = sup_b.SupportBond(next(BondID), "SupportBond", "IN",
G.generatorID)
G.addInBond(supBG_IN)
for semBondName in semBondDict.keys():
if label in semBondDict[semBondName].keys():
#Create a semantic bond
semBG_OUT = sem_b.SemanticBond(next(BondID),
semBondName, "OUT",
G.generatorID)
#Add the equivalence table for the semantic bond
semBG_OUT.compatible = deepcopy(
semBondDict[semBondName][label])
#Create complementary semantic bond
semBG_IN = sem_b.SemanticBond(next(BondID),
semBondName, "IN",
G.generatorID)
#Add created bonds to the generator
G.addOutBond(semBG_OUT)
G.addInBond(semBG_IN)
#Add created generator to list with generators of equivalent modality
localSwapSpace[G.feature].append(G)
globalSwapSpace = {}
for fg in feature_generators:
globalSwapSpace[fg.generatorID] = fg
debugFIle = "./debugFile.txt"
# globalProposalChance Helps avoid local minima. Need to be less than 1 when using MCMC candidate proposal
globalProposalChance = 1.0
PS = inf.Inference(localSwapSpace, globalSwapSpace, False, debugFIle, topK,
globalProposalChance)
topKConfig = PS.run_inference()
topVal = 1
ftFile = open(outFile, 'w')
ftFile.write("Top %s results:\n" % topK)
ftFile.close()
for key in sorted(topKConfig):
topKConfig[key].printConfig(outFile, localSwapSpace.keys())
topVal += 1
# Clean up for Anneal
filelist = [f for f in os.listdir(".") if f.endswith(".state")]
for f in filelist:
os.remove(f)
def process(fold_index=0):
with tf.Graph().as_default():
#with tf.device('/gpu:0'):
session_conf = tf.ConfigProto(
allow_soft_placement=FLAGS.allow_soft_placement,
log_device_placement=FLAGS.log_device_placement)
sess = tf.Session(config=session_conf)
with sess.as_default(), open(precision_log + str(fold_index),
'w') as log:
if len(sys.argv) > 1:
paramG = cPickle.load(open(pre_trained_path + sys.argv[1],
'r'))
paramD = cPickle.load(open(pre_trained_path + sys.argv[2],
'r'))
else:
paramG = None
paramD = None
eb_samples, pro_samples = cPickle.load(
open(path + 'train_samples' + str(fold_index), 'r'))
query_prof_dim = len(pro_samples[0][0])
response_prof_dim = len(pro_samples[0][-1])
batch_size = FLAGS.batch_size
num_batches = int(math.ceil(len(eb_samples) / batch_size))
print np.shape(eb_samples), np.shape(
pro_samples), query_prof_dim, response_prof_dim
dis = Discriminator.Discriminator(
FLAGS.max_sequence_len, FLAGS.batch_size, len(vocab),
FLAGS.embedding_dim,
list(map(int, FLAGS.filter_sizes.split(","))),
FLAGS.num_filters, query_prof_dim, response_prof_dim,
FLAGS.dropout, FLAGS.l2_reg, FLAGS.learning_rate, paramD, None,
FLAGS.loss, True, FLAGS.score_type)
gen = Generator.Generator(
FLAGS.max_sequence_len, FLAGS.batch_size, len(vocab),
FLAGS.embedding_dim,
list(map(int, FLAGS.filter_sizes.split(","))),
FLAGS.num_filters, query_prof_dim, response_prof_dim,
FLAGS.dropout, FLAGS.l2_reg, FLAGS.learning_rate / 50, paramG,
None, FLAGS.loss, True, FLAGS.score_type)
sess.run(tf.global_variables_initializer())
for i in range(FLAGS.num_epochs):
#g_batch_size = len(eb_samples)
for g_epoch in range(FLAGS.g_epochs_num):
#if i == 0:
# break
step, current_loss, positive,negative, pos_score, neg_score = 0, 0.0, 0.0, 0.0, [],[]
for _index, row in enumerate(eb_samples):
#candidate_index = range(len(eb_samples))
#sampled_index = list(np.random.choice(candidate_index, size=[batch_size],replace=False))
#if _index not in sampled_index:
# sampled_index[-1] = _index
#sampled_index = list(candidate_index)
for ib in range(1):
ib = _index / batch_size
end_index = min((ib + 1) * batch_size,
len(eb_samples))
eb_sample_pools = eb_samples[end_index -
batch_size:end_index]
pro_sample_pools = pro_samples[
end_index - batch_size:end_index]
feed_dict = {
gen.input_x_1: [row[0]] * batch_size,
gen.input_x_2: [row[1]] * batch_size,
gen.input_x_3: eb_sample_pools[:, 2],
gen.prof_1:
[pro_samples[_index][0]] * batch_size,
gen.prof_2:
[pro_samples[_index][1]] * batch_size,
gen.prof_3: list(pro_sample_pools[:, 2])
}
predicted = sess.run(gen.gan_score, feed_dict)
exp_rating = np.exp(np.array(predicted) * 10)
prob = exp_rating / np.sum(exp_rating)
neg_index = np.random.choice(range(batch_size),
size=[FLAGS.gan_k],
p=prob,
replace=False)
feed_dict = {
dis.input_x_1: [row[0]] * FLAGS.gan_k,
dis.input_x_2: [row[1]] * FLAGS.gan_k,
dis.input_x_3: eb_sample_pools[neg_index][:,
2],
dis.prof_1:
[pro_samples[_index][0]] * FLAGS.gan_k,
dis.prof_2:
[pro_samples[_index][1]] * FLAGS.gan_k,
dis.prof_3:
list(pro_sample_pools[neg_index][:, 2])
}
reward = sess.run(dis.reward, feed_dict)
feed_dict = {
gen.input_x_1: eb_sample_pools[:, 0],
gen.input_x_2: eb_sample_pools[:, 1],
gen.input_x_3: eb_sample_pools[:, 2],
gen.prof_1: list(pro_sample_pools[:, 0]),
gen.prof_2: list(pro_sample_pools[:, 1]),
gen.prof_3: list(pro_sample_pools[:, 2]),
gen.neg_index: neg_index,
gen.reward: reward
}
_, step, current_loss, gan_score, pos, neg, pos_score, neg_score = sess.run(
[
gen.gan_updates, gen.global_step,
gen.gan_loss, gen.gans, gen.positive,
gen.negative, gen.pos_score, gen.neg_score
], feed_dict)
#print pos_score[:1], neg_score[:1],current_loss, step, _index, len(eb_samples)
line = (
"%s: GEN step %d %d, loss %f gan score %f ,pos score %f, neg score %f, total step: %d "
% (timestamp(), step, i, current_loss, gan_score, pos,
neg, FLAGS.num_epochs * FLAGS.g_epochs_num *
len(eb_samples)))
print line
log.write(line + "\n")
d_eb_samples, d_pro_samples = gan_samples(
eb_samples, pro_samples, sess, gen)
for d_epoch in range(FLAGS.d_epochs_num):
step, current_loss, accuracy = 0, 0.0, 0.0
for ib in range(num_batches):
end_index = min((ib + 1) * batch_size,
len(d_eb_samples))
eb_batch = d_eb_samples[end_index -
batch_size:end_index]
pro_batch = d_pro_samples[end_index -
batch_size:end_index]
feed_dict = {
dis.input_x_1: eb_batch[:, 0],
dis.input_x_2: eb_batch[:, 1],
dis.input_x_3: eb_batch[:, 2],
dis.prof_1: list(pro_batch[:, 0]),
dis.prof_2: list(pro_batch[:, 1]),
dis.prof_3: list(pro_batch[:, 2])
}
_, step, current_loss, accuracy, pos_score, neg_score = sess.run(
[
dis.updates, dis.global_step, dis.loss,
dis.accuracy, dis.positive, dis.negative
], feed_dict)
line = (
"%s: Dis step %d %d, loss %f with acc %f, pos score %f, neg score %f, total step: %d "
% (timestamp(), step, i, current_loss, accuracy,
pos_score, neg_score, FLAGS.num_epochs *
FLAGS.d_epochs_num * num_batches))
print line
log.write(line + '\n')
if i != FLAGS.num_epochs - 1:
evaluation(sess, gen, log, batch_size,
path + 'test_samples' + str(fold_index))
evaluation(sess, dis, log, batch_size,
path + 'test_samples' + str(fold_index))
evaluation(sess, gen, log, batch_size,
path + 'test_samples' + str(fold_index), True)
evaluation(sess, dis, log, batch_size,
path + 'test_samples' + str(fold_index), True)
import chainer
from chainer import serializers
import nn.DCGAN import Generator
from utils import config_parse as cp
from computations.trainer import generate_training_video
MODEL_PATH = './models/generator.npz'
N_GENERATIONS = 100
conf_parser = cp.Config('setup.ini')
params = conf_parser.get_params()
generator = Generator(params.ch, params.latent_dim)
serializers.load_npz(MODEL_PATH, generator)
for i in range(N_GENERATIONS):
print(f'Generator video {i} / {N_GENERATIONS}', end='\r')
generate_training_video(generator, filename=f'./gifs/gif_{i}.gif')