Source code for pyshred.models.decoder_models.unet_model
import torch
import torch.nn as nn
from .abstract_decoder import AbstractDecoder
import torch.nn as nn
[docs]
class UNET(AbstractDecoder):
"""
1D U-Net style convolutional decoder for SHRED.
Uses a series of 1D convolutions to decode latent representations
back to the physical state space. Particularly suitable for
spatially-structured data.
Parameters
----------
conv1 : int, optional
Number of channels in first convolutional layer. Defaults to 256.
conv2 : int, optional
Number of channels in second convolutional layer. Defaults to 1024.
Attributes
----------
c1 : int
First convolution layer channel count.
c2 : int
Second convolution layer channel count.
"""
def __init__(self, conv1: int = 256, conv2: int = 1024):
"""
Initialize the UNET decoder.
Parameters
----------
conv1 : int, optional
Number of channels in first convolutional layer. Defaults to 256.
conv2 : int, optional
Number of channels in second convolutional layer. Defaults to 1024.
"""
super().__init__()
self.c1 = conv1
self.c2 = conv2
# self.dropout = dropout
[docs]
def initialize(self, input_size, output_size):
"""
Initialize the UNET decoder with input and output sizes.
Parameters
----------
input_size : int
Size of the input latent features.
output_size : int
Size of the output physical state.
"""
super().initialize(input_size)
self.conv1 = nn.Conv1d(input_size, self.c1, kernel_size=2, padding=1)
self.conv2 = nn.Conv1d(self.c1, self.c2, kernel_size=4, padding=1)
self.conv3 = nn.Conv1d(self.c2, output_size, kernel_size=2, padding=1)
self.gelu = nn.LeakyReLU()
[docs]
def forward(self, x: torch.Tensor) -> torch.Tensor:
"""
Forward pass through the UNET decoder.
Parameters
----------
x : torch.Tensor
Input latent tensor of shape (batch_size, input_size, sequence_length).
Returns
-------
torch.Tensor
Decoded output tensor of shape (batch_size, output_size).
"""
x = self.gelu(self.conv1(x))
x = self.gelu(self.conv2(x))
x = self.gelu(self.conv3(x))
x = x.permute(0, 2, 1) # Change shape back to [batch_size, sequence_length, d_model]
x = torch.mean(x, dim=1)
return x
@property
def model_name(self):
"""
Name of the decoder model.
Returns
-------
str
Returns "UNET".
"""
return "UNET"