PyTorch torch.nn.Conv1d 関数

import torch
import torch.nn as nn

# 一次元畳み込み：入力チャンネル=3、出力チャンネル=64、カーネルサイズ=3
conv1d = nn.Conv1d(in_channels=3, out_channels=64, kernel_size=3, padding=1)

# 入力：batch=4、チャンネル=3、シーケンス長=100
x = torch.randn(4, 3, 100)

output = conv1d(x)

print("入力形状:", x.shape)
print("出力形状:", output.shape)

import torch
import torch.nn as nn

class TextCNN(nn.Module):
    def __init__(self, vocab_size, embed_dim, num_classes):
        super(TextCNN, self).__init__()
        # 埋め込み層: (batch, seq) -> (batch, embed, seq)
        self.embedding = nn.Embedding(vocab_size, embed_dim)

        # 異なるサイズの複数のカーネル
        self.convs = nn.ModuleList([
            nn.Conv1d(embed_dim, 128, kernel_size=k)
            for k in [2, 3, 4, 5]
        ])
        self.fc = nn.Linear(128 * 4, num_classes)

    def forward(self, x):
        # x: (batch, seq)
        x = self.embedding(x)  # (batch, seq, embed)
        x = x.permute(0, 2, 1)  # (batch, embed, seq)

        # 畳み込み + ReLU + GlobalMaxPool
        pooled = []
        for conv in self.convs:
            c = conv(x)  # (batch, 128, seq')
            c = nn.functional.relu(c)
            c = c.max(dim=2)[0]  # Global max pooling
            pooled.append(c)

        x = torch.cat(pooled, dim=1)
        return self.fc(x)

# テスト
model = TextCNN(vocab_size=10000, embed_dim=128, num_classes=2)
x = torch.randint(0, 10000, (4, 50))  # batch=4, seq=50
output = model(x)

print("入力形状:", x.shape)
print("出力形状:", output.shape)

import torch
import torch.nn as nn

# 音声特徴抽出
conv_audio = nn.Conv1d(in_channels=1, out_channels=32, kernel_size=5, stride=2)

# モノラル音声：batch=4、チャンネル=1、サンプリングポイント=16000
audio = torch.randn(4, 1, 16000)

output = conv_audio(audio)

print("入力形状:", audio.shape)
print("出力形状:", output.shape)
print("出力長:", output.shape[2])