def test_bisect():
f = torch.sigmoid
targets = torch.linspace(0.1, 0.9, 10)
roots = bisect(f, targets, -6.0, 6.0)
assert torch.allclose(f(roots), targets, atol=1e-4)
f = lambda x: -torch.sigmoid(x)
targets = -torch.linspace(0.1, 0.9, 10)
roots = bisect(f, targets, -6.0, 6.0)
assert torch.allclose(f(roots), targets, atol=1e-4)
f = fn.tanhshrink
targets = torch.linspace(-0.4, 0.4, 10)
roots = bisect(f, targets, -6.0, 6.0, abort=100)
assert torch.allclose(f(roots), targets, atol=1e-4)
f = lambda x: -fn.tanhshrink(x)
targets = -torch.linspace(-0.4, 0.4, 10)
roots = bisect(f, targets, -6.0, 6.0, abort=100)
assert torch.allclose(f(roots), targets, atol=1e-4)
f = torch.tanh
targets = torch.linspace(-0.9, 0.9, 10)
roots = bisect(f, targets, -6.0, 6.0, abort=100)
assert torch.allclose(f(roots), targets, atol=1e-4)
f = lambda x: -torch.tanh(x)
targets = -torch.linspace(-0.9, 0.9, 10)
roots = bisect(f, targets, -6.0, 6.0, abort=100)
assert torch.allclose(f(roots), targets, atol=1e-4)
def forward(self, x):
# supported
x = self.conv2d(x)
x = x + x
# not supported
x = self.softshrink(x)
x = F.tanhshrink(x)
return x
def predict(self, h, t, r):
a = self.w[0] * h + self.w2[0] * r - t
b = self.w[1] * h + self.w2[1] * t - r
c = self.w[2] * t + self.w2[2] * r - h
d = self.w[3] * h - self.w2[3] * r * t
score_a = (torch.norm((a[0, :, :]), p=2, dim=1) + torch.norm(
(a[4, :, :]), p=2, dim=1)) / 2.0
score_b = (torch.norm((b[1, :, :]), p=2, dim=1) + torch.norm(
(b[5, :, :]), p=2, dim=1)) / 2.0
score_c = (torch.norm((c[2, :, :]), p=2, dim=1) + torch.norm(
(c[6, :, :]), p=2, dim=1)) / 2.0
score_d = (torch.norm((d[3, :, :]), p=2, dim=1) + torch.norm(
(d[7, :, :]), p=2, dim=1)) / 2.0
psi = 2.5
c = 4.0
amale = (torch.norm((self.w3[0]), p=2, dim=0) * score_a + torch.norm(
(self.w3[1]), p=2, dim=0) * score_b + torch.norm(
(self.w3[2]), p=2, dim=0) * score_c + torch.norm(
(self.w3[3]), p=2, dim=0) * score_d) - c * torch.norm(
(self.w3[4]), p=2, dim=0) - psi
score = F.tanhshrink(amale)
return score
def test_tanhshrink(self):
inp = torch.randn(1, 3, 32, 32, device='cuda', dtype=self.dtype)
output = F.tanhshrink(inp)
def main():
parser = argparse.ArgumentParser()
# decode setting
parser.add_argument("--model",
required=True,
type=str,
help="GRU_RNN model file")
parser.add_argument("--config",
required=True,
type=str,
help="GRU_RNN configure file")
parser.add_argument("--outdir",
required=True,
type=str,
help="directory to save generated samples")
# other setting
#parser.add_argument("--GPU_device", default=None,
# type=int, help="selection of GPU device")
#parser.add_argument("--GPU_device_str", default=None,
# type=str, help="selection of GPU device")
parser.add_argument("--verbose",
default=VERBOSE,
type=int,
help="log level")
args = parser.parse_args()
#if args.GPU_device is not None or args.GPU_device_str is not None:
# os.environ["CUDA_DEVICE_ORDER"] = "PCI_BUS_ID"
# if args.GPU_device_str is None:
# os.environ["CUDA_VISIBLE_DEVICES"] = str(args.GPU_device)
# else:
# os.environ["CUDA_VISIBLE_DEVICES"] = args.GPU_device_str
os.environ["CUDA_VISIBLE_DEVICES"] = ""
# check directory existence
if not os.path.exists(args.outdir):
os.makedirs(args.outdir)
# set log level
if args.verbose > 0:
logging.basicConfig(
level=logging.INFO,
format=
'%(asctime)s (%(module)s:%(lineno)d) %(levelname)s: %(message)s',
datefmt='%m/%d/%Y %I:%M:%S',
filemode='w',
filename=args.outdir + "/decode.log")
logging.getLogger().addHandler(logging.StreamHandler())
elif args.verbose > 1:
logging.basicConfig(
level=logging.DEBUG,
format=
'%(asctime)s (%(module)s:%(lineno)d) %(levelname)s: %(message)s',
datefmt='%m/%d/%Y %I:%M:%S',
filemode='w',
filename=args.outdir + "/decode.log")
logging.getLogger().addHandler(logging.StreamHandler())
else:
logging.basicConfig(
level=logging.WARN,
format=
'%(asctime)s (%(module)s:%(lineno)d) %(levelname)s: %(message)s',
datefmt='%m/%d/%Y %I:%M:%S',
filemode='w',
filename=args.outdir + "/decode.log")
logging.getLogger().addHandler(logging.StreamHandler())
logging.warn("logging is disabled.")
# load config
config = torch.load(args.config)
spk_list = config.spk_list.split('@')
n_spk = len(spk_list)
model_epoch = os.path.basename(args.model).split('.')[0].split('-')[1]
logging.info('epoch: ' + model_epoch)
device = torch.device("cpu")
#with torch.cuda.device(0):
# define model and load parameters
with torch.no_grad():
model_spkidtr = SPKID_TRANSFORM_LAYER(n_spk=n_spk,
emb_dim=config.emb_spk_dim,
n_weight_emb=config.n_weight_emb,
conv_emb_flag=True,
spkidtr_dim=config.spkidtr_dim)
logging.info(model_spkidtr)
model_spkidtr.load_state_dict(
torch.load(args.model, map_location=device)["model_spkidtr"])
model_spkidtr.eval()
for param in model_spkidtr.parameters():
param.requires_grad = False
#feat = torch.LongTensor(np.arange(n_spk)).cuda().unsqueeze(0)
feat = torch.LongTensor(np.arange(n_spk)).unsqueeze(0)
logging.info(feat)
logging.info(spk_list)
colormap = np.array(['b', 'r'])
#colormap = np.array(['b'])
male = ['bdl', 'p237', 'p245', 'p251', 'p252', 'p259', 'p274', 'p304', 'p311', 'p326', 'p345', 'p360', 'p363', \
'p226', 'p227', 'p232', 'p237', 'p241', 'p243', 'p245', 'p246', 'p247', 'p251', 'p252', 'p254', 'p255', 'p256', 'p258', 'p259', 'p260', 'p263', 'p270', 'p271', 'p272', 'p273', 'p274', 'p275', 'p278', 'p279', 'p281', 'p284', 'p285', 'p286', 'p287', 'p292', 'p298', 'p302', 'p304', 'p311', 'p315', 'p316', 'p326', \
'SEM1', 'SEM2', 'TFM1', 'TGM1', 'TMM1', 'TEM1', 'TEM2', \
'VCC2SM1', 'VCC2SM2', 'VCC2SM3', 'VCC2TM1', 'VCC2TM2', 'VCC2SM4', \
'okada', 'otake', 'uchino', 'morikawa', 'yamada', 'nuct'] #32
female = ['slt', 'p231', 'p238', 'p248', 'p253', 'p264', 'p265', 'p266', 'p276', 'p305', 'p308', 'p318', 'p335', \
'p225', 'p228', 'p229', 'p230', 'p231', 'p233', 'p234', 'p236', 'p238', 'p239', 'p240', 'p244', 'p248', 'p249', 'p250', 'p253', 'p257', 'p261', 'p262', 'p264', 'p265', 'p266', 'p267', 'p268', 'p269', 'p276', 'p277', 'p282', 'p283', 'p288', 'p293', 'p294', 'p295', 'p297', 'p299', 'p300', 'p301', 'p303', 'p305', \
'SEF1', 'SEF2', 'TEF1', 'TEF2', 'TFF1', 'TGF1', 'TMF1', \
'VCC2SF1', 'VCC2SF2', 'VCC2SF3', 'VCC2TF1', 'VCC2TF2', 'VCC2SF4', \
'taga', 'takada', 'mizokuchi', 'tts'] #30
#male = ['fake', 'morikawa', 'okada', 'otake', 'uchino', 'SEM1', 'SEM2', 'TFM1', 'TGM1', 'TMM1', 'p237', 'p245', 'p251', 'p252', 'p259', 'p274', 'p304', 'p311', 'p326', 'p345', 'p360', 'p363', 'yamada', 'TEM1', 'TEM2']
#female = ['SEF1', 'SEF2', 'naomi', 'mizokuchi', 'taga', 'takada', 'TEF1', 'TEF2', 'TFF1', 'TGF1', 'TMF1', 'p231', 'p238', 'p248', 'p253', 'p264', 'p265', 'p266', 'p276', 'p305', 'p308', 'p318', 'p335']
#male = ['VCC2SM1' , 'VCC2SM2' , 'VCC2SM3' , 'VCC2TM1' , 'VCC2TM2' , 'VCC2SM4']
#female = ['VCC2SF1' , 'VCC2SF2' , 'VCC2SF3' , 'VCC2TF1' , 'VCC2TF2' , 'VCC2SF4']
gender = []
for i in range(n_spk):
#gender.append(0)
if spk_list[i] in male:
gender.append(0)
elif spk_list[i] in female:
gender.append(1)
else:
logging.info('error %s not in gender list' % (spk_list[i]))
exit()
z = F.tanhshrink(
torch.clamp(model_spkidtr.conv(
model_spkidtr.conv_emb(
F.one_hot(feat, num_classes=n_spk).float().transpose(1,
2))),
min=MIN_CLAMP,
max=MAX_CLAMP)).transpose(
1, 2).cpu().data.numpy().astype(np.float64)
#z = F.tanhshrink(model_spkidtr.conv(F.one_hot(feat, num_classes=n_spk).float().transpose(1,2))).transpose(1,2).cpu().data.numpy().astype(np.float64)
#z = F.tanhshrink(torch.clamp(model_spkidtr.conv(F.one_hot(feat, num_classes=self.n_spk).float().transpose(1,2)), min=-32, max=32)).transpose(1,2)
logging.info(z)
logging.info(args.outdir)
#plt.rcParams["figure.figsize"] = (20,11.25) #1920x1080
plt.rcParams["figure.figsize"] = (14.229166667, 11.25) #1366x1080
#plt.rcParams["figure.figsize"] = (11.25,11.25) #1080x1080
#plt.rcParams["figure.figsize"] = (14.229166667,14.229166667) #1366x1366
logging.info("spk-id spk-name x-coord y-coord")
for i in range(n_spk):
logging.info("%d %s %lf %lf", i + 1, spk_list[i], z[0, i, 0],
z[0, i, 1])
#z = z.cpu().data.numpy()
#z = z.data.numpy()
logging.info(z.shape)
x = z[0, :, 0]
y = z[0, :, 1]
fig, ax = plt.subplots()
ax.scatter(x, y, s=40, c=colormap[gender])
#ax.scatter(x, y, s=80, c=colormap[gender])
#ax.scatter(x, y, s=100, c=colormap[gender])
#ax.scatter(x, y, s=120, c=colormap[gender])
#ax.scatter(x, y, s=160, c=colormap[gender])
for i, txt in enumerate(spk_list):
#ax.annotate(txt, (x[i], y[i]))
#ax.annotate(txt, (x[i], y[i]), weight='bold', size=16)
if txt == 'okada':
txt = 'JpnM1'
elif txt == 'otake':
txt = 'JpnM2'
elif txt == 'uchino':
txt = 'JpnM3'
elif txt == 'morikawa':
txt = 'JpnM4'
elif txt == 'yamada':
txt = 'JpnM5'
elif txt == 'nuct':
txt = 'JpnM6'
elif txt == 'taga':
txt = 'JpnF1'
elif txt == 'takada':
txt = 'JpnF2'
elif txt == 'mizokuchi':
txt = 'JpnF3'
elif txt == 'tts':
txt = 'ttsLJSpeech'
#ax.annotate(txt, (x[i], y[i]), weight='bold', size=16)
ax.annotate(txt, (x[i], y[i]), size=14)
#plt.xlim([-0.35, 0.05])
plt.savefig(os.path.join(args.outdir, 'spk_map.png'))
plt.close()
def forward(self, x):
x = F.tanhshrink(x)
x = x + x
x = F.tanhshrink(x)
return x
pic(x, y, (2,4,2), 'softplus') ''' tanh/ hardtanh/ softsign ''' x = torch.arange(-5,5,0.1).view(-1,1) y = F.tanh(x) pic(x, y, (2,4,3), 'tanh') y = F.hardtanh(x) pic(x, y, (2,4,3), 'hardtanh') y = F.softsign(x) pic(x, y, (2,4,3), 'softsign') ''' hardshrink/ tanhshrink/ tanhshrink ''' x = torch.arange(-3,3,0.1).view(-1,1) y = F.hardshrink(x) pic(x, y, (2,4,4), 'hardshrink') y = F.tanhshrink(x) pic(x, y, (2,4,4), 'tanhshrink') y = F.softshrink(x) pic(x, y, (2,4,4), 'tanhshrink') ''' sigmoid ''' x = torch.arange(-5,5,0.1).view(-1,1) y = F.sigmoid(x) pic(x, y, (2,4,5), 'sigmoid') ''' relu6 ''' x = torch.arange(-5,10,0.1).view(-1,1) y = F.relu6(x) pic(x, y, (2,4,6), 'relu6') ''' logsigmoid '''
def tanh_shrink(self, x):
return F.tanhshrink(x)
def tanhshrink(input, *args, **kwargs):
return _wrap_tensor(input, F.tanhshrink(input.F, *args, **kwargs))
def forward(self, x, y, z, w):
x = F.tanhshrink(x)
y = F.tanhshrink(y)
z = F.tanhshrink(z)
w = F.tanhshrink(w)
return x, y, z, w
def tanhshrink_annealing(model, iteration, max_iterations, start_tf=1.0):
scale = (iteration / max_iterations) * 4 - 2
value = start_tf * (-F.tanhshrink(torch.tensor(scale)).item() /
(2 * 1.0360) + 0.5)
model.teacher_forcing = min(1, max(value, 0))
