WenjieDu/PyPOTS

"""
 
"""
 
# Created by Wenjie Du <wenjay.du@gmail.com>
# License: BSD-3-Clause
 
 
import torch.nn as nn
from torch.nn.utils import weight_norm
 
 
class Chomp1d(nn.Module):
    def __init__(self, chomp_size):
        super().__init__()
        self.chomp_size = chomp_size
 
    def forward(self, x):
        return x[:, :, : -self.chomp_size].contiguous()
 
 
class TemporalBlock(nn.Module):
Function `__init__` has 7 arguments (exceeds 4 allowed). Consider refactoring.
    def __init__(
        self,
        n_inputs,
        n_outputs,
        kernel_size,
        stride,
        dilation,
        padding,
        dropout=0.2,
    ):
        super().__init__()
        self.conv1 = weight_norm(
            nn.Conv1d(
                n_inputs,
                n_outputs,
                kernel_size,
                stride=stride,
                padding=padding,
                dilation=dilation,
            )
        )
        self.chomp1 = Chomp1d(padding)
        self.relu1 = nn.ReLU()
        self.dropout1 = nn.Dropout(dropout)
 
        self.conv2 = weight_norm(
            nn.Conv1d(
                n_outputs,
                n_outputs,
                kernel_size,
                stride=stride,
                padding=padding,
                dilation=dilation,
            )
        )
        self.chomp2 = Chomp1d(padding)
        self.relu2 = nn.ReLU()
        self.dropout2 = nn.Dropout(dropout)
 
        self.net = nn.Sequential(
            self.conv1,
            self.chomp1,
            self.relu1,
            self.dropout1,
            self.conv2,
            self.chomp2,
            self.relu2,
            self.dropout2,
        )
        self.downsample = nn.Conv1d(n_inputs, n_outputs, 1) if n_inputs != n_outputs else None
        self.relu = nn.ReLU()
        self.init_weights()
 
    def init_weights(self):
        self.conv1.weight.data.normal_(0, 0.01)
        self.conv2.weight.data.normal_(0, 0.01)
        if self.downsample is not None:
            self.downsample.weight.data.normal_(0, 0.01)
 
    def forward(self, x):
        out = self.net(x)
        res = x if self.downsample is None else self.downsample(x)
        return self.relu(out + res)