PyTorch torch.nn.GRU 関数

import torch
import torch.nn as nn

# GRU: 入力256次元、隠れ256次元、2層
gru = nn.GRU(input_size=256, hidden_size=256, num_layers=2, batch_first=True)

# 入力: batch=4、系列=10、特徴量=256
x = torch.randn(4, 10, 256)

output, hidden = gru(x)

print("入力形状:", x.shape)
print("出力形状:", output.shape)
print("隠れ状態形状:", hidden.shape)

import torch
import torch.nn as nn
import time

# 同じ構成のLSTMとGRU
lstm = nn.LSTM(256, 256, 1, batch_first=True)
gru = nn.GRU(256, 256, 1, batch_first=True)

x = torch.randn(32, 100, 256)

# 性能比較
for model, name in [(lstm, "LSTM"), (gru, "GRU")]:
    start = time.time()
    for _ in range(100):
        _ = model(x)
    print(f"{name} 時間: {time.time()-start:.3f}s")

import torch
import torch.nn as nn

class GRUClassifier(nn.Module):
    def __init__(self, vocab_size, embed_dim, hidden_dim, num_classes):
        super(GRUClassifier, self).__init__()
        self.embedding = nn.Embedding(vocab_size, embed_dim)
        self.gru = nn.GRU(embed_dim, hidden_dim, batch_first=True, bidirectional=True)
        self.fc = nn.Linear(hidden_dim * 2, num_classes)

    def forward(self, x):
        embedded = self.embedding(x)
        _, hidden = self.gru(embedded)
        # 双方向の最終隠れ状態を連結
        hidden = torch.cat([hidden[-2], hidden[-1]], dim=1)
        return self.fc(hidden)

model = GRUClassifier(10000, 128, 128, 2)
x = torch.randint(0, 10000, (8, 50))
output = model(x)

print("入力:", x.shape, "-> 出力:", output.shape)