def model_test():
args = parse_args()
torch.set_grad_enabled(True)
args.dropout = 0
args.cuda = False
dataset = DatasetBase(args)
print("Data loaded.")
x, y = dataset.__getitem__(0)
model = Net(args)
result1 = model.forward(x.type(torch.int64))
result2 = model.forward(rearrange(x, 0, 1).type(torch.int64))
return result1, result2
def gaussian_correlation(self, x1, x2):
"""
计算高斯卷积
"""
if self._hog_features:
c = np.zeros((self.size_patch[0], self.size_patch[1]), np.float32)
for i in range(self.size_patch[2]):
x1aux = x1[i, :].reshape(
(self.size_patch[0], self.size_patch[1]))
x2aux = x2[i, :].reshape(
(self.size_patch[0], self.size_patch[1]))
caux = cv2.mulSpectrums(fftd(x1aux),
fftd(x2aux),
0,
conjB=True)
caux = real(fftd(caux, True))
c += caux
c = rearrange(c)
else:
c = cv2.mulSpectrums(fftd(x1), fftd(x2), 0,
conjB=True) # 'conjB=' 是必要的
c = fftd(c, True)
c = real(c)
c = rearrange(c)
if x1.ndim == 3 and x2.ndim == 3:
d = (np.sum(x1[:, :, 0] * x1[:, :, 0]) +
np.sum(x2[:, :, 0] * x2[:, :, 0]) -
2.0 * c) / (self.size_patch[0] * self.size_patch[1] *
self.size_patch[2])
elif x1.ndim == 2 and x2.ndim == 2:
d = (np.sum(x1 * x1) + np.sum(x2 * x2) - 2.0 * c) / (
self.size_patch[0] * self.size_patch[1] * self.size_patch[2])
d = d * (d >= 0)
d = np.exp(-d / (self.sigma * self.sigma))
return d
# make sure correct location was chosen
print XC[iy0c:iy1c + 1, ixc:ixc + 2]
print YC[iy0c:iy1c + 1, ixc]
print DXC[iy0c:iy1c + 1, ixc]
# loop over hours
for hour in range(start, end + 1):
# file to save record to
savefile = savedir + "rec_{:010d}.npz".format(hour)
# if this step is required
if (not os.path.isfile(savefile)) or force_overwrite:
step = hour * 3600 / dt
print hour, step
# read variables
U = rdmds(diagdir + 'trsp_3d_set1', step, rec=0, lev=0)
V = rdmds(diagdir + 'trsp_3d_set1', step, rec=1, lev=0)
U, V = utils.rearrange_velocities(U, V)
h = rdmds(diagdir + 'state_2d_set1', step, rec=0)
h = utils.rearrange(h)
# save to file
np.savez(savefile,
h=h[iy0c:iy1c + 1, ixc:ixc + 2],
u=U[iy0c:iy1c + 1, ixc:ixc + 3],
v=V[iy0c:iy1c + 2, ixc:ixc + 2])
f_fbank_train = data_dir + 'fbank/train.ark'
f_fbank_test = data_dir + 'fbank/test.ark'
f_mfcc_train = data_dir + 'mfcc/train.ark'
f_mfcc_test = data_dir + 'mfcc/test.ark'
f_train_label = data_dir + 'label/train.lab'
f_phone2phone = data_dir + 'phones/48_39.map'
f_phone2char = data_dir + '48phone_char.map'
# load map
phone2phone, phone2char, phone2idx = load_phone_map(f_phone2phone,
f_phone2char)
# load train
data_X, data_X_id = load_data(f_fbank_train, delimiter=' ', dtype='float32')
data_Y, data_Y_id = load_data(f_train_label, delimiter=',', dtype='str')
data_X, data_X_id = rearrange(data_X, data_X_id, data_Y_id)
# load test
test_X, test_X_id = load_data(f_fbank_test, delimiter=' ', dtype='float32')
# to 39 phone to idx to one-hot
for idx in range(len(data_Y)):
data_Y[idx] = np.vectorize(phone2phone.get)(data_Y[idx])
data_Y[idx] = np.vectorize(phone2idx.get)(data_Y[idx])
data_Y[idx] = np.eye(48)[data_Y[idx].reshape(-1)]
# padding
max_squ_len = np.array([len(d) for d in data_X] + [len(t)
for t in test_X]).max()
print('max_squ_len:{}'.format(max_squ_len))
data_X = np.array([pad(x, (max_squ_len, x.shape[1])) for x in data_X])