Source code for pyshred.models.sequence_models.lstm_model

import torch.nn as nn
import torch
from .abstract_sequence import AbstractSequence
from ..decoder_models.mlp_model import MLP
from ..decoder_models.unet_model import UNET

[docs] class LSTM(AbstractSequence): """ LSTM sequence model for encoding temporal sensor dynamics. Parameters ---------- hidden_size : int, optional Size of the hidden state. Defaults to 64. num_layers : int, optional Number of LSTM layers. Defaults to 2. layer_norm : bool, optional Whether to apply layer normalization. Defaults to False. Attributes ---------- hidden_size : int Size of the hidden state. num_layers : int Number of LSTM layers. use_layer_norm : bool Whether layer normalization is applied. """ def __init__(self, hidden_size:int =64, num_layers:int =2, layer_norm: bool = False): """ Initialize the LSTM model. Parameters ---------- hidden_size : int, optional Size of the hidden state. Defaults to 64. num_layers : int, optional Number of LSTM layers. Defaults to 2. layer_norm : bool, optional Whether to apply layer normalization. Defaults to False. """ super().__init__() self.hidden_size = hidden_size self.num_layers = num_layers self.lstm = None self.decoder_type = None self.output_size = hidden_size self.use_layer_norm = layer_norm if self.use_layer_norm: self.layer_norm = nn.LayerNorm(self.hidden_size)
[docs] def initialize(self, input_size:int, decoder_type, **kwargs): """ Initialize the LSTM with input size and decoder. Parameters ---------- input_size : int Number of input features. decoder_type : str Decoder model type. **kwargs Additional keyword arguments. """ super().initialize(input_size) self.lstm = nn.LSTM( input_size=self.input_size, hidden_size=self.hidden_size, num_layers=self.num_layers, batch_first=True ) self.decoder_type = decoder_type
[docs] def forward(self, x): """ Forward pass through the LSTM model. Parameters ---------- x : torch.Tensor Input tensor of shape (batch_size, sequence_length, input_size). Returns ------- torch.Tensor Output tensor with latent representations. """ super().forward(x) device = next(self.parameters()).device # Initialize hidden and cell h_0 = torch.zeros((self.num_layers, x.size(0), self.hidden_size), device=device) c_0 = torch.zeros((self.num_layers, x.size(0), self.hidden_size), device=device) out, (h_out, c_out) = self.lstm(x, (h_0, c_0)) # Apply layer normalization if enabled if self.use_layer_norm: # Normalize per time-step across hidden dimension out = self.layer_norm(out) # Normalize the final hidden state from last layer h_last = self.layer_norm(h_out[-1]) else: h_last = h_out[-1] # Decode if self.decoder_type == "MLP": return h_last.view(-1, self.hidden_size) elif self.decoder_type == "UNET": return out.permute(0, 2, 1) else: raise TypeError( f"Unsupported decoder type: {self.decoder_type}." )
@property def model_name(self): """ Name of the sequence model. Returns ------- str Returns "LSTM". """ return "LSTM"