mindscience.models.neural_operator.FFNO2D

class mindscience.models.neural_operator.FFNO2D(in_channels, out_channels, n_modes, resolutions, hidden_channels=20, lifting_channels=None, projection_channels=128, factor=1, n_layers=4, n_ff_layers=2, ff_weight_norm=False, layer_norm=True, share_weight=False, r_padding=0, data_format='channels_last', positional_embedding=True, dft_compute_dtype=mstype.float32, ffno_compute_dtype=mstype.float16)[source]

The 2D Factorized Fourier Neural Operator, which usually contains a Lifting Layer, a Factorized Fourier Block Layer and a Projection Layer. The details can be found in A. Tran, A. Mathews, et. al: FACTORIZED FOURIER NEURAL OPERATORS.

Parameters

  • in_channels (int) – The number of channels in the input space.

  • out_channels (int) – The number of channels in the output space.

  • n_modes (Union[int, list(int)]) – The number of modes reserved after linear transformation in Fourier Layer.

  • resolutions (Union[int, list(int)]) – The resolutions of the input tensor.

  • hidden_channels (int, optional) – The number of channels of the FNOBlock input and output. Default: 20.

  • lifting_channels (int, optional) – The number of channels of the lifting layer mid channels. Default: None.

  • projection_channels (int, optional) – The number of channels of the projection layer mid channels. Default: 128.

  • factor (int, optional) – The number of neurons in the hidden layer of a feedforward network. Default: 1.

  • n_layers (int, optional) – The number that Fourier Layer nests. Default: 4.

  • n_ff_layers (int, optional) – The number of layers (hidden layers) in the feedforward neural network. Default: 2.

  • ff_weight_norm (bool, optional) – Whether to do weight normalization in feedforward or not. Used as a reserved function interface, the weight normalization is not supported in feedforward. Default: False.

  • layer_norm (bool, optional) – Whether to do layer normalization in feedforward or not. Default: True.

  • share_weight (bool, optional) – Whether to share weights between SpectralConv layers or not. Default: False.

  • r_padding (int, optional) – The number used to pad a tensor on the right in a certain dimension. Default: 0.

  • data_format (str, optional) – The input data channel sequence. Default: "channels_last".

  • positional_embedding (bool, optional) – Whether to embed positional information or not. Default: True.

  • dft_compute_dtype (dtype.Number, optional) – The computation type of DFT in SpectralConvDft. Default: mstype.float32.

  • ffno_compute_dtype (dtype.Number, optional) – The computation type of MLP in fno skip. Should be mstype.float32 or mstype.float16. mstype.float32 is recommended for the GPU backend, mstype.float16 is recommended for the Ascend backend. Default: mstype.float16.

Inputs:

  • x (Tensor) - Tensor of shape \((batch\_size, resolution, in\_channels)\).

Outputs:

  • output (Tensor) -Tensor of shape \((batch\_size, resolution, out\_channels)\).

Raises

ValueError – If ff_weight_norm is not False.

Examples

>>> import numpy as np
>>> import mindspore
>>> from mindspore import Tensor
>>> import mindspore.common.dtype as mstype
>>> from mindscience.models.neural_operator.ffno import FFNO2D
>>> data = Tensor(np.ones([2, 128, 128, 3]), mstype.float32)
>>> net = FFNO2D(in_channels=3, out_channels=3, n_modes=[20, 20], resolutions=[128, 128])
>>> out = net(data)
>>> print(data.shape, out.shape)
(2, 128, 128, 3) (2, 128, 128, 3)