You can not select more than 25 topics
Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
65 lines
2.0 KiB
65 lines
2.0 KiB
# -*- coding: utf-8 -*-
|
|
"""
|
|
Created on 2024/11/4 11:07
|
|
@author: Whenxuan Wang
|
|
@email: wwhenxuan@gmail.com
|
|
@url: https://github.com/wwhenxuan/SymTime
|
|
"""
|
|
import torch
|
|
from torch import nn
|
|
from torch import Tensor
|
|
from typing import Optional
|
|
|
|
|
|
class Flatten_Heads(nn.Module):
|
|
"""Integrate the final output of the time series encoder"""
|
|
|
|
def __init__(
|
|
self,
|
|
individual: bool,
|
|
n_vars: int,
|
|
nf: int,
|
|
patch_num: int,
|
|
targets_window: int,
|
|
head_dropout: int = 0,
|
|
cls_token: Optional[bool] = True,
|
|
) -> None:
|
|
super().__init__()
|
|
# Whether to output in a channel-independent manner
|
|
self.individual = individual
|
|
self.n_vars = n_vars
|
|
self.patch_num = patch_num
|
|
|
|
# Whether to take the [CLS] Token
|
|
if cls_token is True:
|
|
self.patch_num += 1
|
|
|
|
if self.individual is True:
|
|
self.linears = nn.ModuleList()
|
|
self.dropouts = nn.ModuleList()
|
|
self.flattens = nn.ModuleList()
|
|
for i in range(self.n_vars):
|
|
self.linears.append(nn.Linear(nf * self.patch_num, targets_window))
|
|
self.dropouts.append(nn.Dropout(head_dropout))
|
|
self.flattens.append(nn.Flatten(start_dim=-2))
|
|
else:
|
|
self.linear = nn.Linear(nf * self.patch_num, targets_window)
|
|
self.dropout = nn.Dropout(head_dropout)
|
|
self.flatten = nn.Flatten(start_dim=-2)
|
|
|
|
def forward(self, x: Tensor) -> Tensor: # [batch_size, n_vars, d_model, patch_num]
|
|
if self.individual is True:
|
|
x_out = []
|
|
for i in range(self.n_vars):
|
|
# Flatten the dimensions of a specific channel.
|
|
z = self.flattens[i](x[:, i, :, :])
|
|
z = self.linears[i](z)
|
|
z = self.dropouts[i](z)
|
|
x_out.append(z)
|
|
x = torch.stack(x_out, dim=1)
|
|
else:
|
|
x = self.flatten(x)
|
|
x = self.linear(x)
|
|
x = self.dropout(x)
|
|
return x
|