def __init__(self,
ntokens,
nsampled,
nhid,
tied_weight,
n_proj,
sparse=True):
super(SampledSoftmax, self).__init__()
# Parameters
self.ntokens = ntokens
self.nsampled = nsampled
self.sampler = LogUniformSampler(self.ntokens)
self.n_proj = n_proj
if n_proj != nhid:
self.projection = nn.Linear(n_proj, nhid)
else:
self.projection = None
if not sparse:
self.params = nn.Linear(nhid, ntokens)
if tied_weight is not None:
self.params.weight = tied_weight
else:
util.initialize(self.params.weight)
else:
self.softmax_b = torch.nn.Embedding(ntokens, 1, sparse=True)
self.softmax_w = torch.nn.Embedding(ntokens, nhid, sparse=True)
self.sparse = sparse
print("-- Used sampled softmax with " + str(self.nsampled) +
" samples.")
def __init__(self, ntokens, nsampled, nhid, tied_weight):
super(SampledSoftmax, self).__init__()
# Parameters
self.ntokens = ntokens
self.nsampled = nsampled
self.sampler = LogUniformSampler(self.ntokens)
self.params = nn.Linear(nhid, ntokens)
if tied_weight is not None:
self.params.weight = tied_weight
else:
util.initialize(self.params.weight)
def __init__(self, ntokens, ninp, nhid, nout, nlayers, proj, dropout):
super(RNNModel, self).__init__()
# Parameters
self.nhid = nhid
self.nlayers = nlayers
# Create Layers
self.drop = nn.Dropout(dropout)
self.rnn = nn.LSTM(ninp, nhid, nlayers, dropout=dropout)
if proj:
self.proj = nn.Linear(nhid, nout)
util.initialize(self.proj.weight)
else:
self.proj = None
def main():
if len(sys.argv) > 1 and \
sys.argv[1] == "--about":
about()
return
pygame.init()
screen = pygame.display.set_mode(LAUNCH_SCREEN_RECT.size, 0)
util.initialize()
pygame.display.set_caption("DVONN for One")
icon = pygame.transform.scale(util.get_image("icon"), (32, 32))
pygame.display.set_icon(icon)
clock = pygame.time.Clock()
l = Launcher(screen, clock)
l.run()
def main():
if len(sys.argv) > 1 and \
sys.argv[1] == "--about":
about()
return
pygame.init()
screen = pygame.display.set_mode(LAUNCH_SCREEN_RECT.size,0)
util.initialize()
pygame.display.set_caption("DVONN for One")
icon = pygame.transform.scale(util.get_image("icon"),(32,32))
pygame.display.set_icon(icon)
clock = pygame.time.Clock()
l = Launcher(screen,clock)
l.run()
def viewBySelected(self):
mlog.debug("> upcoming.Window.viewBySelected()")
# switch to next view by
self.viewBy += 1
if self.viewBy >= self.VIEW_BY_MAX:
self.viewBy = self.VIEW_BY_MIN + 1
# store the setting change
self.settings.put("upcoming_view_by",
"%d" % self.viewBy,
shouldCreate=1)
# refresh the listing
self.loadList()
mlog.debug("< upcoming.Window.viewBySelected()")
# =============================================================================
if __name__ == "__main__":
try:
loadingWin = xbmcgui.DialogProgress()
loadingWin.create("Upcoming Shows", "Loading Upcoming Shows",
"Please Wait...")
util.initialize()
showWindow(loadingWin)
except Exception, ex:
traceback.print_exc()
ui.Dialog().ok('Shit blew up!', str(ex))
def pretrain_manipulators(params):
speaker_categs = torch.load(params.speaker_categs_path)
num_speakers, speaker_feature_dim = speaker_categs.size()
describer_model = util.load_model(params.header + DESCRIBER_FOOTER)
describer = Describer(describer_model, speaker_feature_dim)
describer.load_state_dict(torch.load(
'snapshots/' + params.header + DESCRIBER_FOOTER + '.pth'))
describer.eval()
reconstructor_model = util.load_model(params.header + RECONSTRUCTOR_FOOTER)
reconstructor = Reconstructor(reconstructor_model, params.log_frac)
reconstructor.load_state_dict(torch.load(
'snapshots/' + params.header + RECONSTRUCTOR_FOOTER + '.pth'))
latent_forger_model = util.load_model(params.header + LATENT_FORGER_FOOTER)
latent_forger = LatentForger(latent_forger_model)
util.initialize(latent_forger)
if example_tensor.is_cuda:
speaker_categs = speaker_categs.cuda()
describer = describer.cuda()
reconstructor = reconstructor.cuda()
latent_forger = latent_forger.cuda()
optim = torch.optim.Adam(latent_forger.parameters(), lr=params.lr)
data_loader = VCTKLoader(
params.data_path, example_tensor, features='log')
data_iterator = iter(data_loader)
latent_loss_sum_batch = 0.0
reconst_loss_sum_batch = 0.0
loss_sum_batch = 0.0
num_in_batch = 0
period = 0
while period < params.num_periods:
period += 1
loss_sum = 0.0
loss_count = 0
print(util.COMMENT_HEADER, end='')
for _ in range(params.period_size):
orig, orig_speaker = next(data_iterator)
orig_categ = speaker_categs[orig_speaker].unsqueeze(0)
forgery_categ = speaker_categs[
np.random.randint(num_speakers)].unsqueeze(0)
orig_latent, metadata = describer.latent(orig)
orig_latent = orig_latent.detach()
pretend_latent = latent_forger.modify_latent(
orig_latent, forgery_categ, orig_categ)
pretend_reconst = reconstructor.reconst(pretend_latent, metadata)
latent_loss = latent_forger.pretrain_loss(
pretend_latent, orig_latent)
reconst_loss = reconstructor.reconst_loss(pretend_reconst, orig)
loss = latent_loss + reconst_loss
latent_loss_sum_batch += latent_loss.item()
reconst_loss_sum_batch += reconst_loss.item()
loss_sum_batch = loss_sum_batch + loss
num_in_batch += 1
if num_in_batch >= params.batch_size:
mean_loss = loss_sum_batch / num_in_batch
optim.zero_grad()
mean_loss.backward()
optim.step()
print("(" + "|".join([
"%0.3f" % (latent_loss_sum_batch / num_in_batch),
"%0.3f" % (reconst_loss_sum_batch / num_in_batch)]) + ")",
end=' ', flush=True)
latent_loss_sum_batch = 0.0
reconst_loss_sum_batch = 0.0
loss_sum_batch = 0.0
num_in_batch = 0
loss_sum += loss.item()
loss_count += 1
print('')
loss_mean = loss_sum / loss_count
metrics = [
('period', period),
('loss', round(loss_mean, 3))
]
util.print_metrics(metrics)
torch.save(
latent_forger.state_dict(),
'snapshots/' + params.header + LATENT_FORGER_FOOTER + '.pth')
def initialize(doc=__doc__):
util.initialize(doc)
blockify = Blockify(blocklist.Blocklist())
return blockify
if not queue: # Can't save an empty queue.
error_msg(outbar, 'Queue is empty.')
return
if fn is None: # No argument.
error_msg(outbar, 'Missing argument to write.')
return
path = join(expanduser('~'), '.local', 'share', 'pmcli', 'playlists')
if not exists(path): # No playlists directory.
error_msg(outbar, 'Path to playlists does not exist.')
elif exists(join(path, fn)):
error_msg(outbar, 'Playist %s already exists.' % fn)
else: # Write the playlist.
with open(join(path, fn), 'a') as f:
json.dump(queue, f)
addstr(outbar, 'Wrote queue to %s.' % fn)
if __name__ == '__main__':
colour, main, inbar, infobar, outbar = initialize()
addstr(infobar, 'Enter \'h\' or \'help\' if you need help.')
queue = Queue()
global content
content = None
while True:
transition(get_input())
def train(params):
encoder = util.load_model(params.header + ENCODER_FOOTER)
util.initialize(encoder)
encoder.train()
discriminator = util.load_model(params.header + DISCRIMINATOR_FOOTER)
util.initialize(discriminator)
discriminator.train()
if example_tensor.is_cuda:
encoder = encoder.cuda()
discriminator = discriminator.cuda()
optim = torch.optim.Adam(torch.nn.Sequential(encoder,
discriminator).parameters(),
lr=params.lr)
mse = torch.nn.MSELoss()
bce = torch.nn.BCEWithLogitsLoss()
speaker_averages = {}
average_decay = 1.0 - 1.0 / params.average_count
data_loader = loader.VCTKLoader(
params.data_path,
example_tensor,
features='log',
speaker_take_count=params.speaker_take_count,
utterance_take_count=params.utterance_take_count)
def update_average(speaker, encoded):
if speaker not in speaker_averages:
speaker_averages[speaker] = encoded
else:
speaker_averages[speaker] = (
average_decay * speaker_averages[speaker] +
(1.0 - average_decay) * encoded)
cluster_loss_sum_batch = 0.0
discrim_loss_sum_batch = 0.0
loss_sum_batch = 0.0
num_in_batch = 0
data_iterator = iter(data_loader)
period = 0
while period < params.num_periods:
period += 1
loss_sum_print = 0.0
loss_count_print = 0
print(util.COMMENT_HEADER, end='')
for _ in range(params.period_size):
utterance_1, speaker_1 = next(data_iterator)
utterance_2, speaker_2 = next(data_iterator)
encoded_1 = encoder(utterance_1)
encoded_2 = encoder(utterance_2)
update_average(speaker_1, encoded_1.detach())
update_average(speaker_2, encoded_2.detach())
discrim_1 = discriminator(torch.cat([encoded_1, encoded_1], dim=1))
discrim_2 = discriminator(torch.cat([encoded_2, encoded_2], dim=1))
discrim_3 = discriminator(torch.cat([encoded_1, encoded_2], dim=1))
discrim_4 = discriminator(torch.cat([encoded_2, encoded_1], dim=1))
compare = 1 if speaker_1 == speaker_2 else 0
compare = example_tensor.new_tensor(compare).reshape(1, 1)
same = example_tensor.new_tensor(1).reshape(1, 1)
discrim_loss = (bce(discrim_1, same) + bce(discrim_2, same) + bce(
discrim_3, compare) + bce(discrim_4, compare)) / 4.0
cluster_loss = (mse(encoded_1, speaker_averages[speaker_1]) +
mse(encoded_2, speaker_averages[speaker_2])) / 2.0
loss = params.cluster_term * cluster_loss + discrim_loss
cluster_loss_sum_batch += cluster_loss.item()
discrim_loss_sum_batch += discrim_loss.item()
loss_sum_batch = loss_sum_batch + loss
num_in_batch += 1
if num_in_batch >= params.batch_size:
mean_loss = loss_sum_batch / num_in_batch
optim.zero_grad()
mean_loss.backward()
optim.step()
loss_sum_print += mean_loss.item()
loss_count_print += 1
print("(" + "|".join([
"%0.3f" % (cluster_loss_sum_batch / num_in_batch),
"%0.3f" % (discrim_loss_sum_batch / num_in_batch)
]) + ")",
end=' ',
flush=True)
cluster_loss_sum_batch = 0.0
discrim_loss_sum_batch = 0.0
loss_sum_batch = 0.0
num_in_batch = 0
print('')
loss_mean = loss_sum_print / loss_count_print
metrics = [('period', period), ('loss', round(loss_mean, 3))]
util.print_metrics(metrics)
torch.save(encoder.state_dict(),
'snapshots/' + params.header + ENCODER_FOOTER + '.pth')
torch.save(
discriminator.state_dict(),
'snapshots/' + params.header + DISCRIMINATOR_FOOTER + '.pth')
s.subtitle()))
self.controls['schedule_lbl'].control.setLabel(self.translator.get(84) % totRecs)
self.controls['guidedata'].control.reset()
self.controls['guidedata'].control.addLabel(self.conn.getGuideData())
except:
log.exception("refresh")
#***************************************************************************#
# End Window Class #
#***************************************************************************#
try:
log.info("\n\t\tInitializing...\n\n")
updateProgress("Initializing...")
util.initialize()
log.info(">> MythBox Version: %s <<" % __version__)
updateProgress("Creating Main Window")
import mythtv
import mythdb
import mythtv
import util
updateProgress("Checking Settings")
# TODO: Springify later...
# Injected dependencies that get passed all over the place
# 1
nsampled = 8192
train_corpus = StreamGBWDataset(vocabulary, os.path.join(args.data, "training-monolingual.tokenized.shuffled/*"))
test_corpus = StreamGBWDataset(vocabulary, os.path.join(args.data, "heldout-monolingual.tokenized.shuffled/*"), deterministic=True)
print("load dataset - complete")
'''
###############################################################################
# Build the model
###############################################################################
eval_batch_size = 1
net = model.RNNModel(ntokens, args.emsize, args.nhid, args.nlayers,
args.dropout)
encoder = nn.Embedding(ntokens, args.emsize)
util.initialize(encoder, ntokens)
twht = None
if args.tied:
if args.nhid != args.emsize:
raise ValueError(
'When using the tied flag, hidden must be equal to embedding size')
twht = encoder.weight
ss = model.SampledSoftmax(ntokens, nsampled, args.nhid, tied_weight=twht)
net.add_module("encoder", encoder)
net.add_module("decoder", ss)
net.cuda(0)
encoder.cuda(0)
def train_analysts(params):
speaker_categs = torch.load(params.speaker_categs_path)
speaker_feature_dim = speaker_categs.size()[1]
describer_model = util.load_model(params.header + DESCRIBER_FOOTER)
describer = Describer(describer_model, speaker_feature_dim)
util.initialize(describer)
reconstructor_model = util.load_model(params.header + RECONSTRUCTOR_FOOTER)
reconstructor = Reconstructor(reconstructor_model, params.log_frac)
util.initialize(reconstructor)
examiner_model = util.load_model(params.header + EXAMINER_FOOTER)
distinguisher_model = util.load_model(params.header + DISTINGUISHER_FOOTER)
discriminator = PairDiscriminator(examiner_model, distinguisher_model)
util.initialize(discriminator)
discriminator.train()
if example_tensor.is_cuda:
speaker_categs = speaker_categs.cuda()
describer = describer.cuda()
reconstructor = reconstructor.cuda()
discriminator = discriminator.cuda()
optim = torch.optim.Adam(
torch.nn.Sequential(describer, reconstructor).parameters(),
lr=params.lr)
advers_optim = torch.optim.Adam(
discriminator.parameters(),
lr=params.advers_lr)
data_loader = VCTKLoader(
params.data_path, example_tensor, features='log')
data_iterator = iter(data_loader)
categ_loss_sum_batch = 0.0
robustness_loss_sum_batch = 0.0
reconst_loss_sum_batch = 0.0
gen_loss_batch = 0.0
advers_loss_batch = 0.0
loss_sum_batch = 0.0
discrim_loss_sum_batch = 0.0
num_in_batch = 0
period = 0
while period < params.num_periods:
period += 1
loss_sum = 0.0
loss_count = 0
print(util.COMMENT_HEADER, end='')
for _ in range(params.period_size):
orig, orig_speaker = next(data_iterator)
orig_categ = speaker_categs[orig_speaker].unsqueeze(0)
describer.eval()
center_categ = describer.categ(orig).detach()
describer.train()
(latent, metadata, pred_categ) = describer.describe(orig)
reconst = reconstructor.reconst(latent, metadata)
truth = 0 if (np.random.random() < 0.5) else 1
if truth == 0:
decision = discriminator.discriminate(
orig_categ, orig, reconst)
else:
decision = discriminator.discriminate(
orig_categ, reconst, orig)
categ_loss = describer.categ_loss(pred_categ, orig_categ)
robustness_loss = describer.categ_loss(pred_categ, center_categ)
reconst_loss = reconstructor.reconst_loss(reconst, orig)
gen_loss = discriminator.gen_loss(decision, truth)
advers_loss = discriminator.advers_loss(decision, truth)
loss = (
params.categ_term * categ_loss +
params.robustness_term * robustness_loss +
reconst_loss +
params.advers_term * gen_loss)
discrim_loss = advers_loss
categ_loss_sum_batch += categ_loss.item()
robustness_loss_sum_batch += robustness_loss.item()
reconst_loss_sum_batch += reconst_loss.item()
gen_loss_batch += gen_loss.item()
advers_loss_batch += advers_loss.item()
loss_sum_batch = loss_sum_batch + loss
discrim_loss_sum_batch = discrim_loss_sum_batch + discrim_loss
num_in_batch += 1
if num_in_batch >= params.batch_size:
mean_discrim_loss = discrim_loss_sum_batch / num_in_batch
if gen_loss_batch / num_in_batch <= 10.0:
advers_optim.zero_grad()
mean_discrim_loss.backward(retain_graph=True)
advers_optim.step()
mean_loss = loss_sum_batch / num_in_batch
optim.zero_grad()
mean_loss.backward()
optim.step()
print("(" + "|".join([
"%0.3f" % (categ_loss_sum_batch / num_in_batch),
"%0.3f" % (robustness_loss_sum_batch / num_in_batch),
"%0.3f" % (reconst_loss_sum_batch / num_in_batch),
"%0.3f" % (gen_loss_batch / num_in_batch),
"%0.3f" % (advers_loss_batch / num_in_batch)]) + ")",
end=' ', flush=True)
categ_loss_sum_batch = 0.0
robustness_loss_sum_batch = 0.0
reconst_loss_sum_batch = 0.0
gen_loss_batch = 0.0
advers_loss_batch = 0.0
loss_sum_batch = 0.0
discrim_loss_sum_batch = 0.0
num_in_batch = 0
loss_sum += loss.item()
loss_count += 1
print('')
loss_mean = loss_sum / loss_count
metrics = [
('period', period),
('loss', round(loss_mean, 3))
]
util.print_metrics(metrics)
torch.save(
describer.state_dict(),
'snapshots/' + params.header + DESCRIBER_FOOTER + '.pth')
torch.save(
reconstructor.state_dict(),
'snapshots/' + params.header + RECONSTRUCTOR_FOOTER + '.pth')
def train_manipulators(params):
speaker_categs = torch.load(params.speaker_categs_path)
num_speakers, speaker_feature_dim = speaker_categs.size()
describer_model = util.load_model(params.header + DESCRIBER_FOOTER)
describer = Describer(describer_model, speaker_feature_dim)
describer.load_state_dict(torch.load(
'snapshots/' + params.header + DESCRIBER_FOOTER + '.pth'))
describer.eval()
reconstructor_model = util.load_model(params.header + RECONSTRUCTOR_FOOTER)
reconstructor = Reconstructor(reconstructor_model, params.log_frac)
reconstructor.load_state_dict(torch.load(
'snapshots/' + params.header + RECONSTRUCTOR_FOOTER + '.pth'))
latent_forger_model = util.load_model(params.header + LATENT_FORGER_FOOTER)
latent_forger = LatentForger(latent_forger_model)
latent_forger.load_state_dict(torch.load(
'snapshots/' + params.header + LATENT_FORGER_FOOTER + '.pth'))
latent_forger.train()
examiner_model = util.load_model(params.header + EXAMINER_FOOTER)
distinguisher_model = util.load_model(params.header + DISTINGUISHER_FOOTER)
discriminator = PairDiscriminator(examiner_model, distinguisher_model)
util.initialize(discriminator)
discriminator.train()
if example_tensor.is_cuda:
speaker_categs = speaker_categs.cuda()
describer = describer.cuda()
reconstructor = reconstructor.cuda()
latent_forger = latent_forger.cuda()
discriminator = discriminator.cuda()
optim = torch.optim.Adam(
latent_forger.parameters(),
lr=params.lr)
advers_optim = torch.optim.Adam(
discriminator.parameters(),
lr=params.advers_lr)
data_loader = VCTKLoader(
params.data_path, example_tensor, features='log')
data_iterator = iter(data_loader)
forgery_categ_loss_sum_batch = 0.0
activity_loss_sum_batch = 0.0
pretend_latent_loss_sum_batch = 0.0
pretend_reconst_loss_sum_batch = 0.0
gen_loss_batch = 0.0
advers_loss_batch = 0.0
loss_sum_batch = 0.0
discrim_loss_sum_batch = 0.0
num_in_batch = 0
period = 0
while period < params.num_periods:
period += 1
loss_sum = 0.0
loss_count = 0
print(util.COMMENT_HEADER, end='')
for _ in range(params.period_size):
orig, orig_speaker = next(data_iterator)
target, target_speaker = next(data_iterator)
orig_categ = speaker_categs[orig_speaker].unsqueeze(0)
target_categ = speaker_categs[target_speaker].unsqueeze(0)
target_latent, metadata = describer.latent(target)
target_reconst = reconstructor.reconst(target_latent, metadata)
target_reconst = target_reconst.detach()
orig_latent, metadata = describer.latent(orig)
orig_latent = orig_latent.detach()
forgery_latent_raw = latent_forger.modify_latent(
orig_latent, orig_categ, target_categ)
forgery = reconstructor.reconst(forgery_latent_raw, metadata)
(forgery_latent, metadata, pred_forgery_categ) = \
describer.describe(forgery)
activity_orig = torch.exp(orig).mean(dim=1)
activity_forgery = torch.exp(forgery).mean(dim=1)
pretend_latent = latent_forger.modify_latent(
forgery_latent, target_categ, orig_categ)
pretend_reconst = reconstructor.reconst(pretend_latent, metadata)
truth = 0 if (np.random.random() < 0.5) else 1
if truth == 0:
decision = discriminator.discriminate(
target_categ, target_reconst, forgery)
else:
decision = discriminator.discriminate(
target_categ, forgery, target_reconst)
forgery_categ_loss = describer.categ_loss(
pred_forgery_categ, target_categ)
activity_loss = ((activity_orig - activity_forgery) ** 2).mean(
dim=list(range(activity_orig.dim())))
pretend_latent_loss = describer.latent_loss(
pretend_latent, orig_latent)
pretend_reconst_loss = reconstructor.reconst_loss(
pretend_reconst, orig)
gen_loss = discriminator.gen_loss(decision, truth)
advers_loss = discriminator.advers_loss(decision, truth)
loss = (params.categ_term * forgery_categ_loss +
params.activity_term * activity_loss +
pretend_latent_loss +
pretend_reconst_loss +
params.advers_term * gen_loss)
discrim_loss = advers_loss
forgery_categ_loss_sum_batch += forgery_categ_loss.item()
activity_loss_sum_batch += activity_loss.item()
pretend_latent_loss_sum_batch += pretend_latent_loss.item()
pretend_reconst_loss_sum_batch += pretend_reconst_loss.item()
gen_loss_batch += gen_loss.item()
advers_loss_batch += advers_loss.item()
loss_sum_batch = loss_sum_batch + loss
discrim_loss_sum_batch = discrim_loss_sum_batch + discrim_loss
num_in_batch += 1
if num_in_batch >= params.batch_size:
mean_discrim_loss = discrim_loss_sum_batch / num_in_batch
if gen_loss_batch / num_in_batch <= 10.0:
advers_optim.zero_grad()
mean_discrim_loss.backward(retain_graph=True)
advers_optim.step()
mean_loss = loss_sum_batch / num_in_batch
optim.zero_grad()
mean_loss.backward()
if period >= 1:
optim.step()
print("(" + "|".join([
"%0.3f" % (forgery_categ_loss_sum_batch / num_in_batch),
"%0.3f" % (activity_loss_sum_batch / num_in_batch),
"%0.3f" % (pretend_latent_loss_sum_batch / num_in_batch),
"%0.3f" % (pretend_reconst_loss_sum_batch / num_in_batch),
"%0.3f" % (gen_loss_batch / num_in_batch),
"%0.3f" % (advers_loss_batch / num_in_batch)
]) + ")", end=' ', flush=True)
forgery_categ_loss_sum_batch = 0.0
activity_loss_sum_batch = 0.0
pretend_latent_loss_sum_batch = 0.0
pretend_reconst_loss_sum_batch = 0.0
gen_loss_batch = 0.0
advers_loss_batch = 0.0
loss_sum_batch = 0.0
discrim_loss_sum_batch = 0.0
num_in_batch = 0
loss_sum += loss.item()
loss_count += 1
print('')
loss_mean = loss_sum / loss_count
metrics = [
('period', period),
('loss', round(loss_mean, 3))
]
util.print_metrics(metrics)
torch.save(
latent_forger.state_dict(),
'snapshots/' + params.header + LATENT_FORGER_FOOTER + '.pth')
import kaldi_active_grammar
if __name__ == '__main__':
import util
compiler, decoder = util.initialize()
##### Set up a rule mixing strict commands with free dictation
rule = kaldi_active_grammar.KaldiRule(compiler, 'TestRule')
fst = rule.fst
dictation_nonterm = '#nonterm:dictation'
end_nonterm = '#nonterm:end'
# Optional preface
previous_state = fst.add_state(initial=True)
next_state = fst.add_state()
fst.add_arc(previous_state, next_state, 'cap')
fst.add_arc(previous_state, next_state,
None) # Optionally skip, with an epsilon (silent) arc
# Required free dictation
previous_state = next_state
extra_state = fst.add_state()
next_state = fst.add_state()
# These two arcs together (always use together) will recognize one or more words of free dictation (but not zero):
fst.add_arc(previous_state, extra_state, dictation_nonterm)
fst.add_arc(extra_state, next_state, None, end_nonterm)
# Loop repetition, alternating between a group of alternatives and more free dictation
previous_state = next_state
def __init__(self):
self.client = carla.Client("127.0.0.1", 2000)
self.client.set_timeout(20.0)
change_to_Town06(self.client)
self.world = self.client.get_world()
self.world_snapshot = self.world.get_snapshot()
self.time_step_count = 0
self.time_step = 0.025 # Seconds. Have to fully divide 1
self.curr_time = 0
self.subject_path_time_res = 0.5
self.warmup_time = 3
# 0. Reset Scene
initialize(self.world, self.client, self.time_step)
self.lane_marker_linestring = (
get_lane_marker_linestring_from_right_lane_road_and_lane_id(
self.world, 15, -6))
# 1. Spawn vehicles
self.subject_behavior = "very_aggressive"
(
self.ego_vehicle,
self.subject_vehicle,
self.current_lane_waypoints,
self.subject_agent,
) = setup_scenario(
self.world,
self.client,
synchronous_master=True,
subject_behavior=self.subject_behavior,
)
# 2. Get the path follower object
self.path_follower = PathFollower(self.world, self.ego_vehicle,
self.time_step)
# 3. Coarse Trajectory Genertion
road = Road()
actions = Actions()
constraints = Constraints()
termination_conditions = TerminationConditions(max_time=15,
max_position_x=180)
start_state = State([
self.ego_vehicle.get_location().x,
self.ego_vehicle.get_location().y
], 0, 0)
cost_calculator = CostCalculator(
subject_path_time_res=self.subject_path_time_res)
self.latticeGenerator = LatticeGenerator(
road=road,
actions=actions,
constraints=constraints,
cost_calculator=cost_calculator,
termination_conditions=termination_conditions,
start_state=start_state,
ego=self.ego_vehicle,
subject=self.subject_vehicle,
)
# 3. Get the controller object
args_lateral = {
"K_P": 1.0,
"K_D": 0.0,
"K_I": 0.0,
"dt": self.time_step
}
args_longitudinal = {
"K_P": 1.0,
"K_D": 0.00,
"K_I": 0.00,
"dt": self.time_step,
}
self.controller = VehiclePIDController(self.ego_vehicle, args_lateral,
args_longitudinal)
# 4. Placeholder for the concatenated trajectory
self.traj_to_track = []
self.has_lane_change_happend = 0
self.is_lane_change_happening = 0
self.planned_path = None
self.next_timestep = None
self.latest_tracked_speed_ego = get_speed(self.ego_vehicle)
# self.regenerate_traj_flag = False
if self.subject_behavior == "manual":
self.path_predictor = PathPredictor(self.world,
self.subject_vehicle,
self.subject_path_time_res)
else:
self.path_predictor = PathPredictor(self.world,
self.subject_agent.vehicle,
self.subject_path_time_res)
self.subject_traj = []
self.ego_traj = []
def initialize(doc=__doc__):
util.initialize(doc)
blockify = Blockify(blocklist.Blocklist())
return blockify
ntokens = len(vocabulary)
nsampled = 8192
train_corpus = StreamGBWDataset(vocabulary, os.path.join(args.data, "training-monolingual.tokenized.shuffled/*"))
test_corpus = StreamGBWDataset(vocabulary, os.path.join(args.data, "heldout-monolingual.tokenized.shuffled/*"), deterministic=True)
print("load dataset - complete")
'''
###############################################################################
# Build the model
###############################################################################
eval_batch_size = 1
net = model.RNNModel(ntokens, args.emsize, args.nhid, args.emsize, args.nlayers, args.proj, args.dropout)
encoder = nn.Embedding(ntokens, args.emsize)
util.initialize(encoder.weight)
twht = None
if args.tied:
if args.nhid != args.emsize and not args.proj:
raise ValueError('When using the tied flag, hidden must be equal to embedding size')
twht = encoder.weight
D = args.emsize if args.proj else args.nhid
ss = model.SampledSoftmax(ntokens, nsampled, D, tied_weight=twht)
net.add_module("encoder", encoder)
net.add_module("decoder", ss)
net.cuda()
criterion = nn.CrossEntropyLoss()
def learn(env,
total_timesteps,
seed=None,
nsteps=1024,
ent_coef=0.01,
lr=0.01,
vf_coef=0.5,
p_coef=1.0,
max_grad_norm=None,
gamma=0.99,
lam=0.95,
nminibatches=15,
noptepochs=4,
cliprange=0.2,
save_interval=100,
copeoperation=False,
human_ent_coef=0.01,
human_vf_coef=0.5,
human_p_coef=1.0):
set_global_seeds(seed)
sess = get_session()
global_summary = tf.summary.FileWriter(
'summaries/' + 'feeding' +
datetime.datetime.now().strftime('%d-%m-%y%H%M'), sess.graph)
if isinstance(lr, float): lr = constfn(lr)
else: assert callable(lr)
if isinstance(cliprange, float): cliprange = constfn(cliprange)
else: assert callable(cliprange)
# Get the nb of env
nenvs = env.num_envs
# Calculate the batch_size
nbatch = nenvs * nsteps
nbatch_train = nbatch // nminibatches
if copeoperation == True:
human_model = Model(env=env,
nbatch_act=nenvs,
nbatch_train=nbatch_train,
ent_coef=human_ent_coef,
vf_coef=human_vf_coef,
p_coef=human_p_coef,
max_grad_norm=max_grad_norm,
human=True,
robot=False)
robot_model = Model(env=env,
nbatch_act=nenvs,
nbatch_train=nbatch_train,
ent_coef=ent_coef,
vf_coef=vf_coef,
p_coef=p_coef,
max_grad_norm=max_grad_norm,
human=False,
robot=True)
if copeoperation == False:
model = Model(env=env,
nbatch_act=nenvs,
nbatch_train=nbatch_train,
ent_coef=ent_coef,
vf_coef=vf_coef,
p_coef=p_coef,
max_grad_norm=max_grad_norm)
initialize()
# Instantiate the runner object
if copeoperation == True:
runner = Runner(env=env,
model=None,
nsteps=nsteps,
gamma=gamma,
lam=lam,
human_model=human_model,
robot_model=robot_model)
if copeoperation == False:
runner = Runner(env=env,
model=model,
nsteps=nsteps,
gamma=gamma,
lam=lam)
epinfobuf = deque(maxlen=10) #recent 10 episode
pbar = tqdm(total=total_timesteps, dynamic_ncols=True)
tfirststart = time.perf_counter()
nupdates = total_timesteps // nbatch
for update in range(1, nupdates + 1):
assert nbatch % nminibatches == 0
# Start timer
frac = 1.0 - (update - 1.0) / nupdates
# Calculate the learning rate
lrnow = lr(frac)
# Calculate the cliprange
cliprangenow = cliprange(frac)
# Get minibatch
if copeoperation == False:
obs, returns, masks, actions, values, neglogpacs, epinfos = runner.run(
)
if copeoperation == True:
obs, human_returns, robot_returns, masks, human_actions, robot_actions, human_values, robot_values, human_neglogpacs, robot_neglogpacs, epinfos = runner.coop_run(
)
epinfobuf.extend(epinfos)
mblossvals = []
human_mblossvals = []
robot_mblossvals = []
inds = np.arange(nbatch)
for _ in range(noptepochs):
# Randomize the indexes
np.random.shuffle(inds)
for start in range(0, nbatch, nbatch_train):
end = start + nbatch_train
mbinds = inds[start:end]
if copeoperation == True:
human_slices = (arr[mbinds]
for arr in (obs[:, 24:], human_returns,
human_actions, human_values,
human_neglogpacs))
robot_slices = (arr[mbinds]
for arr in (obs[:, :24], robot_returns,
robot_actions, robot_values,
robot_neglogpacs))
human_mblossvals.append(
human_model.train(lrnow, cliprangenow, *human_slices))
robot_mblossvals.append(
robot_model.train(lrnow, cliprangenow, *robot_slices))
if copeoperation == False:
slices = (arr[mbinds] for arr in (obs, returns, actions,
values, neglogpacs))
mblossvals.append(model.train(lrnow, cliprangenow,
*slices)) #None
# Feedforward --> get losses --> update
if copeoperation == True:
human_lossvals = np.mean(human_mblossvals, axis=0)
robot_lossvals = np.mean(robot_mblossvals, axis=0)
if copeoperation == False:
lossvals = np.mean(mblossvals, axis=0)
summary = tf.Summary()
if copeoperation == True:
human_ev = explained_variance(human_values, human_returns)
robot_ev = explained_variance(robot_values, robot_returns)
if copeoperation == False:
ev = explained_variance(values, returns)
performance_r = np.mean([epinfo['r'] for epinfo in epinfobuf])
performance_len = np.mean([epinfo['l'] for epinfo in epinfobuf])
success_time = np.mean(
[epinfo['success_time'] for epinfo in epinfobuf])
fall_time = np.mean([epinfo['fall_time'] for epinfo in epinfobuf])
summary.value.add(tag='Perf/Reward', simple_value=performance_r)
summary.value.add(tag='Perf/episode_len', simple_value=performance_len)
summary.value.add(tag='Perf/success_time', simple_value=success_time)
summary.value.add(tag='Perf/fall_time', simple_value=fall_time)
if copeoperation == True:
summary.value.add(tag='Perf/human_explained_variance',
simple_value=float(human_ev))
summary.value.add(tag='Perf/robot_explained_variance',
simple_value=float(robot_ev))
if copeoperation == False:
summary.value.add(tag='Perf/explained_variance',
simple_value=float(ev))
if copeoperation == True:
for (human_lossval, human_lossname) in zip(human_lossvals,
human_model.loss_names):
if human_lossname == 'grad_norm':
summary.value.add(tag='grad/' + human_lossname,
simple_value=human_lossval)
else:
summary.value.add(tag='human_loss/' + human_lossname,
simple_value=human_lossval)
for (robot_lossval, robot_lossname) in zip(robot_lossvals,
robot_model.loss_names):
if robot_lossname == 'grad_norm':
summary.value.add(tag='grad/' + robot_lossname,
simple_value=robot_lossval)
else:
summary.value.add(tag='robot_loss/' + robot_lossname,
simple_value=robot_lossval)
if copeoperation == False:
for (lossval, lossname) in zip(lossvals, model.loss_names):
if lossname == 'grad_norm':
summary.value.add(tag='grad/' + lossname,
simple_value=lossval)
else:
summary.value.add(tag='loss/' + lossname,
simple_value=lossval)
global_summary.add_summary(summary, int(update * nbatch))
global_summary.flush()
print('finish one update')
if update % 10 == 0:
msg = 'step: {},episode reward: {},episode len: {},success_time: {},fall_time: {}'
pbar.update(update * nbatch)
pbar.set_description(
msg.format(update * nbatch, performance_r, performance_len,
success_time, fall_time))
if update % save_interval == 0:
tnow = time.perf_counter()
print('consume time', tnow - tfirststart)
if copeoperation == True:
savepath = osp.join("my_model_cop/", '%.5i' % update)
if copeoperation == False:
savepath = osp.join("my_model/", '%.5i' % update)
os.makedirs(savepath, exist_ok=True)
savepath = osp.join(savepath, 'ppomodel')
print('Saving to', savepath)
save_state(savepath)
pbar.close()
return model
vocabulary = Vocabulary.from_file(os.path.join(args.data, "1b_word_vocab.txt"))
ntokens = len(vocabulary)
nsampled = 16384
train_corpus = StreamGBWDataset(vocabulary, os.path.join(args.data, "training-monolingual.tokenized.shuffled/*"))
test_corpus = StreamGBWDataset(vocabulary, os.path.join(args.data, "heldout-monolingual.tokenized.shuffled/*"), deterministic=True)
print("load dataset - complete")
'''
###############################################################################
# Build the model
###############################################################################
embed = nn.Embedding(ntokens, args.emsize)
net = m.TransformerModel(m.DecoderPreprocessor(args, embed), m.TransformerDecoder(args, embed), nsampled)
util.initialize(embed.weight)
net.cuda()
print("Sampled Softmax:", nsampled, "Batch Size:", args.batch_size, "Initial LR:", args.lr)
#optimizer = Adam(net.parameters(), args.lr, betas=(0.0, 0.999))
#optimizer = optim.RMSprop(net.parameters(), args.lr)
optimizer = RMSprop(net.parameters(), args.lr)
scheduler = optim.lr_scheduler.CosineAnnealingLR(optimizer, T_max=train_corpus.batch_num*args.epochs, eta_min=1e-8)
net, optimizer = amp.initialize(net, optimizer, opt_level="O1")
###############################################################################
# Training code
###############################################################################
def get_batch(item, device_id=0):
data, target, wrd_cnt, batch_num = item
#!/usr/bin/env python3
import dotfiles
import util
import shutil
import os
from pathlib import Path
util.initialize(Path(__file__).parent)
modules = dotfiles.Module.__subclasses__()
for i, module in enumerate(modules):
progress = "({} out of {})".format(i + 1, len(modules))
module_name = util.color(module.__name__, util.Color.green)
module_source = util.color(module.__module__, util.Color.yellow)
print("Installing module", module_name, "from", module_source, progress)
instance = module()
for message in instance.install():
print(util.timestamp(message))
print("Copying over files from module", module_name)
default_path = Path(os.path.expanduser(module.default_dest))
module_path = Path(module.__module__)
valid_file_paths = [path for path in module_path.glob("*") if util.validate_path(path)]
for path in valid_file_paths:
