Python torch.tensordot方法代碼示例

# 需要導入模塊: import torch [as 別名]
# 或者: from torch import tensordot [as 別名]
def get_obs(Asymm, H, Sx, Sy, Sz, C, E ):
    # A(phy,u,l,d,r), C(d,r), E(u,r,d)
    
    Da = Asymm.size()
    Td = torch.einsum('mefgh,nabcd->eafbgchdmn',(Asymm,Asymm)).contiguous().view(Da[1]**2, Da[2]**2, Da[3]**2, Da[4]**2, Da[0], Da[0])
    #print( torch.dist( Td, Td.permute(0,3,2,1,4,5) ) )    # test left-right reflection symmetry of Td

    CE = torch.tensordot(C,E,([1],[0]))         # C(1d)E(dga)->CE(1ga)
    EL = torch.tensordot(E,CE,([2],[0]))        # E(2e1)CE(1ga)->EL(2ega)  use E(2e1) == E(1e2) 
    EL = torch.tensordot(EL,Td,([1,2],[1,0]))   # EL(2ega)T(gehbmn)->EL(2ahbmn)
    EL = torch.tensordot(EL,CE,([0,2],[0,1]))   # EL(2ahbmn)CE(2hc)->EL(abmnc), use CE(2hc) == CE(1ga) 
    Rho = torch.tensordot(EL,EL,([0,1,4],[0,1,4])).permute(0,2,1,3).contiguous().view(Da[0]**2,Da[0]**2)
    
    # print( (Rho-Rho.t()).norm() )
    Rho = 0.5*(Rho + Rho.t())
    
    Tnorm = Rho.trace()
    Energy = torch.mm(Rho,H).trace()/Tnorm
    Mx = torch.mm(Rho,Sx).trace()/Tnorm
    My = torch.mm(Rho,Sy).trace()/Tnorm
    Mz = torch.mm(Rho,Sz).trace()/Tnorm
   
    #print("Tnorm = %g, Energy = %g " % (Tnorm.item(), Energy.item()) )

    return Energy, Mx, My, Mz 

# 需要導入模塊: import torch [as 別名]
# 或者: from torch import tensordot [as 別名]
def idct_8x8(image):
  alpha = np.array([1. / np.sqrt(2)] + [1] * 7)
  alpha = torch.FloatTensor(np.outer(alpha, alpha)).cuda()
  image = image * alpha

  tensor = np.zeros((8, 8, 8, 8), dtype=np.float32)
  for x, y, u, v in itertools.product(range(8), repeat=4):
    tensor[x, y, u, v] = np.cos((2 * u + 1) * x * np.pi / 16) * np.cos(
        (2 * v + 1) * y * np.pi / 16)
  # result = 0.25 * torch.tensordot(image, torch.as_tensor(tensor, device="cuda"), dims=2) + 128
  result = 0.25 * tensordot_pytorch(image, torch.as_tensor(tensor, device="cuda"), dims=2) + 128
  result.view(image.size())
  return result


# -3. Block joining 

# 需要導入模塊: import torch [as 別名]
# 或者: from torch import tensordot [as 別名]
def forward(self, inputs):
        embeds_vec_list = inputs
        row = []
        col = []

        for r, c in itertools.combinations(embeds_vec_list, 2):
            row.append(r)
            col.append(c)

        p = torch.cat(row, dim=1)
        q = torch.cat(col, dim=1)
        inner_product = p * q

        bi_interaction = inner_product
        attention_temp = F.relu(torch.tensordot(
            bi_interaction, self.attention_W, dims=([-1], [0])) + self.attention_b)

        self.normalized_att_score = F.softmax(torch.tensordot(
            attention_temp, self.projection_h, dims=([-1], [0])), dim=1)
        attention_output = torch.sum(
            self.normalized_att_score * bi_interaction, dim=1)

        attention_output = self.dropout(attention_output)  # training

        afm_out = torch.tensordot(
            attention_output, self.projection_p, dims=([-1], [0]))
        return afm_out 

# 需要導入模塊: import torch [as 別名]
# 或者: from torch import tensordot [as 別名]
def tensordot(x, y, dims):
    """
    Wrapper around :func:`torch.tensordot` or :func:`np.tensordot`
    to operate on real-valued Funsors.

    Note this operates only on the ``output`` tensor. To perform sum-product
    contractions on named dimensions, instead use ``+`` and
    :class:`~funsor.terms.Reduce`.

    Arguments should satisfy::

        len(x.shape) >= dims
        len(y.shape) >= dims
        dims == 0 or x.shape[-dims:] == y.shape[:dims]

    :param Funsor x: A left hand argument.
    :param Funsor y: A y hand argument.
    :param int dims: The number of dimension of overlap of output shape.
    :rtype: Funsor
    """
    assert dims >= 0
    assert len(x.shape) >= dims
    assert len(y.shape) >= dims
    assert dims == 0 or x.shape[-dims:] == y.shape[:dims]
    x_start, x_end = 0, len(x.output.shape)
    y_start = x_end - dims
    y_end = y_start + len(y.output.shape)
    symbols = 'abcdefghijklmnopqrstuvwxyz'
    equation = '{},{}->{}'.format(symbols[x_start:x_end],
                                  symbols[y_start:y_end],
                                  symbols[x_start:y_start] + symbols[x_end:y_end])
    return Einsum(equation, (x, y)) 

# 需要導入模塊: import torch [as 別名]
# 或者: from torch import tensordot [as 別名]
def _numeric_tensordot(x, y, dim):
    if get_backend() == "torch":
        import torch

        return torch.tensordot(x, y, dim)
    else:
        return np.tensordot(x, y, axes=dim) 

# 需要導入模塊: import torch [as 別名]
# 或者: from torch import tensordot [as 別名]
def test_tensor_tensordot(x_shape, xy_shape, y_shape):
    x = randn(x_shape + xy_shape)
    y = randn(xy_shape + y_shape)
    dim = len(xy_shape)
    actual = tensordot(Tensor(x), Tensor(y), dim)
    expected = Tensor(_numeric_tensordot(x, y, dim))
    assert_close(actual, expected, atol=1e-5, rtol=None) 

# 需要導入模塊: import torch [as 別名]
# 或者: from torch import tensordot [as 別名]
def rgb_to_ycbcr(image):
  matrix = np.array(
      [[65.481, 128.553, 24.966],
       [-37.797, -74.203, 112.],
       [112., -93.786, -18.214]],
      dtype=np.float32).T / 255
  shift = torch.as_tensor([16., 128., 128.], device="cuda")

  # result = torch.tensordot(image, torch.as_tensor(matrix, device="cuda"), dims=1) + shift
  result = tensordot_pytorch(image, matrix, dims=1) + shift
  result.view(image.size())
  return result 

# 需要導入模塊: import torch [as 別名]
# 或者: from torch import tensordot [as 別名]
def rgb_to_ycbcr_jpeg(image):
  matrix = np.array(
      [[0.299, 0.587, 0.114],
       [-0.168736, -0.331264, 0.5],
       [0.5, -0.418688, -0.081312]],
      dtype=np.float32).T
  shift = torch.as_tensor([0., 128., 128.], device="cuda")

  # result = torch.tensordot(image, torch.as_tensor(matrix, device="cuda"), dims=1) + shift
  result = tensordot_pytorch(image, torch.as_tensor(matrix, device='cuda'), dims=1) + shift
  result.view(image.size())
  return result


# 2. Chroma subsampling 

# 需要導入模塊: import torch [as 別名]
# 或者: from torch import tensordot [as 別名]
def dct_8x8(image):
  image = image - 128
  tensor = np.zeros((8, 8, 8, 8), dtype=np.float32)
  for x, y, u, v in itertools.product(range(8), repeat=4):
    tensor[x, y, u, v] = np.cos((2 * x + 1) * u * np.pi / 16) * np.cos(
        (2 * y + 1) * v * np.pi / 16)
  alpha = np.array([1. / np.sqrt(2)] + [1] * 7)
  scale = torch.FloatTensor(np.outer(alpha, alpha) * 0.25).cuda()
  #result = scale * torch.tensordot(image, torch.as_tensor(tensor, device="cuda"), dims=2)
  result = scale * tensordot_pytorch(image, torch.as_tensor(tensor, device="cuda"), dims=2)
  result.view(image.size())
  return result 

# 需要導入模塊: import torch [as 別名]
# 或者: from torch import tensordot [as 別名]
def ycbcr_to_rgb(image):
  matrix = np.array(
      [[298.082, 0, 408.583],
       [298.082, -100.291, -208.120],
       [298.082, 516.412, 0]],
      dtype=np.float32).T / 256
  shift = torch.as_tensor([-222.921, 135.576, -276.836], device="cuda")

  # result = torch.tensordot(image, torch.tensor(matrix, device="cuda"), dims=1) + shift
  result = tensordot_pytorch(image, torch.tensor(matrix, device="cuda"), dims=1) + shift
  result.view(image.size())
  return result 

# 需要導入模塊: import torch [as 別名]
# 或者: from torch import tensordot [as 別名]
def test_tensordot():
  backend = pytorch_backend.PyTorchBackend()
  a = backend.convert_to_tensor(2 * np.ones((2, 3, 4)))
  b = backend.convert_to_tensor(np.ones((2, 3, 4)))
  actual = backend.tensordot(a, b, ((1, 2), (1, 2)))
  expected = np.array([[24.0, 24.0], [24.0, 24.0]])
  np.testing.assert_allclose(expected, actual) 

# 需要導入模塊: import torch [as 別名]
# 或者: from torch import tensordot [as 別名]
def test_eigsh_lanczos_0():
  #this test should just not crash
  dtype = torch.float64
  backend = pytorch_backend.PyTorchBackend()
  D = 4
  init = backend.randn((2, 2, 2), dtype=dtype)
  tmp = backend.randn((8, 8), dtype=dtype)
  H = tmp + backend.transpose(backend.conj(tmp), (1, 0))
  H = H.reshape([2, 2, 2, 2, 2, 2])

  def mv(x, mat):
    return torch.tensordot(mat, x, ([0, 3, 5], [2, 0, 1])).permute([2, 0, 1])

  backend.eigsh_lanczos(mv, [H], init, num_krylov_vecs=D) 

# 需要導入模塊: import torch [as 別名]
# 或者: from torch import tensordot [as 別名]
def forward(self, inputs, embed=True):
    if embed:
      return torch.nn.functional.embedding(inputs, self.w)
    else:
      return torch.tensordot(inputs, self.w.t(), 1) + self.b 

# 需要導入模塊: import torch [as 別名]
# 或者: from torch import tensordot [as 別名]
def sample(self, p, z=None):
        """Input p to be shaped [T,B,A,P] or [B,A,P], A: number of actions, P:
        number of atoms.  Optional input z is domain of atom-values, shaped
        [P].  Vector epsilon of lenght B will apply across Batch dimension."""
        q = torch.tensordot(p, z or self.z, dims=1)
        return super().sample(q) 

# 需要導入模塊: import torch [as 別名]
# 或者: from torch import tensordot [as 別名]
def reweight(self, msa1hot):
        # Reweight
        seqlen = msa1hot.size(1)
        id_min = seqlen * self.msa_cutoff
        id_mtx = torch.tensordot(msa1hot, msa1hot, [[1, 2], [1, 2]])
        id_mask = id_mtx > id_min
        weights = 1.0 / id_mask.float().sum(-1)
        return weights