{"id":86118,"date":"2026-07-28T05:40:34","date_gmt":"2026-07-27T21:40:34","guid":{"rendered":"https:\/\/www.wsisp.com\/helps\/86118.html"},"modified":"2026-07-28T05:40:34","modified_gmt":"2026-07-27T21:40:34","slug":"%e9%87%91%e8%9e%8d%e6%97%b6%e5%ba%8f%e6%95%b0%e6%8d%ae%e8%ae%ad%e7%bb%83%ef%bc%9a%e9%95%bf%e5%ba%8f%e5%88%97%e5%bb%ba%e6%a8%a1%e7%9a%84%e8%ae%ad%e7%bb%83%e7%a8%b3%e5%ae%9a%e6%80%a7%e9%97%ae%e9%a2%98","status":"publish","type":"post","link":"https:\/\/www.wsisp.com\/helps\/86118.html","title":{"rendered":"\u91d1\u878d\u65f6\u5e8f\u6570\u636e\u8bad\u7ec3\uff1a\u957f\u5e8f\u5217\u5efa\u6a21\u7684\u8bad\u7ec3\u7a33\u5b9a\u6027\u95ee\u9898"},"content":{"rendered":"

\u91d1\u878d\u65f6\u5e8f\u6570\u636e\u8bad\u7ec3：\u957f\u5e8f\u5217\u5efa\u6a21\u7684\u8bad\u7ec3\u7a33\u5b9a\u6027\u95ee\u9898<\/h2>\n

\u4e00\u3001\u4e2a\u6027\u5316\u6df1\u5ea6\u5f15\u8a00<\/h3>\n

\u5f53\u4f60\u5728\u8bad\u7ec3\u4e00\u4e2a LSTM \u9884\u6d4b\u80a1\u7968\u4ef7\u683c\u65f6，\u7a81\u7136\u53d1\u73b0 loss \u5728\u7b2c 347 \u4e2a epoch \u7206\u70b8\u6210 NaN\u2014\u2014\u8fd9\u4e0d\u662f bug，\u8fd9\u662f\u91d1\u878d\u65f6\u5e8f\u5efa\u6a21\u7684\u65e5\u5e38\u3002<\/p>\n

\u91d1\u878d\u65f6\u95f4\u5e8f\u5217\u6709\u4e24\u4e2a\u8ba9\u6a21\u578b\u5934\u75bc\u7684\u7279\u6027：\u4e00\u662f\u957f\u7a0b\u4f9d\u8d56（3 \u5e74\u524d\u7684\u4e8b\u4ef6\u53ef\u80fd\u5f71\u54cd\u4eca\u5929\u7684\u8d70\u52bf），\u4e8c\u662f\u5206\u5e03\u6f02\u79fb（\u5e02\u573a\u7684\u7edf\u8ba1\u7279\u6027\u968f\u65f6\u95f4\u6539\u53d8）\u3002\u8fd9\u4e24\u4e2a\u7279\u6027\u53e0\u52a0\u8d77\u6765，\u8ba9\u8bad\u7ec3\u7a33\u5b9a\u6027\u6210\u4e3a\u91d1\u878d\u65f6\u5e8f\u6a21\u578b\u7684\u7b2c\u4e00\u9053\u95e8\u69db\u3002\u89c1\u8bc1\u5947\u8ff9\u7684\u65f6\u523b\u5728\u4e8e：\u5f53\u6211\u4eec\u5728\u635f\u5931\u51fd\u6570\u4e2d\u5f15\u5165\u201c\u5bf9\u7a81\u53d8\u70b9\u7684\u60e9\u7f5a\u9879\u201d\u540e，\u90a3\u4e2a\u5728 347 \u4e2a epoch \u5c31\u5d29\u6389\u7684\u6a21\u578b，\u7a33\u5b9a\u8dd1\u4e86 2000 \u4e2a epoch\u3002<\/p>\n

\u4e8c\u3001\u4e2a\u6027\u5316\u539f\u7406\u5256\u6790<\/h3>\n

\u957f\u5e8f\u5217\u5efa\u6a21\u7684\u4e09\u5927\u4e0d\u7a33\u5b9a\u6e90<\/h4>\n

\u7a33\u5b9a\u6027\u5206\u6790<\/h4>\n

1. \u68af\u5ea6\u95ee\u9898<\/h5>\n

\u4f20\u7edf RNN \u5728\u53cd\u5411\u4f20\u64ad\u65f6，\u68af\u5ea6\u4f1a\u7ecf\u5386\u8fde\u4e58\u64cd\u4f5c\u3002\u5bf9\u4e8e\u957f\u5e8f\u5217，\u8fd9\u4e2a\u4e58\u79ef\u8981\u4e48\u8d8b\u8fd1\u4e8e 0（\u68af\u5ea6\u6d88\u5931），\u8981\u4e48\u8d8b\u8fd1\u65e0\u7a77（\u68af\u5ea6\u7206\u70b8）\u3002<\/p>\n

LSTM \u901a\u8fc7\u95e8\u63a7\u673a\u5236\u7f13\u89e3\u4e86\u8fd9\u4e2a\u95ee\u9898，\u4f46\u5e76\u6ca1\u6709\u5f7b\u5e95\u89e3\u51b3\u3002\u5f53\u5e8f\u5217\u957f\u5ea6\u8d85\u8fc7 1000 \u6b65\u65f6，\u5373\u4f7f\u662f LSTM \u4e5f\u4f1a\u51fa\u73b0\u68af\u5ea6\u95ee\u9898\u3002<\/p>\n

2. \u5206\u5e03\u6f02\u79fb<\/h5>\n

\u91d1\u878d\u6570\u636e\u7684\u5206\u5e03\u4e0d\u662f\u56fa\u5b9a\u7684\u3002\u5e02\u573a\u7684\u6ce2\u52a8\u7387\u3001\u76f8\u5173\u6027\u3001\u8d8b\u52bf\u7279\u5f81\u90fd\u4f1a\u968f\u65f6\u95f4\u53d8\u5316\u3002\u7528 2020 \u5e74\u7684\u6570\u636e\u8bad\u7ec3\u7684\u6a21\u578b，\u653e\u5230 2022 \u5e74\u5c31\u4f1a\u5931\u6548\u2014\u2014\u56e0\u4e3a\u5e02\u573a\u57fa\u672c\u9762\u5df2\u7ecf\u5b8c\u5168\u6539\u53d8\u3002<\/p>\n

\u8fd9\u79cd\u201c\u975e\u5e73\u7a33\u6027\u201d\u662f\u91d1\u878d\u65f6\u5e8f\u5efa\u6a21\u7279\u6709\u7684\u6311\u6218\u3002\u5176\u4ed6\u9886\u57df\u7684\u65f6\u5e8f\u6570\u636e（\u5982\u5929\u6c14\u3001\u7535\u529b\u8d1f\u8377）\u867d\u7136\u4e5f\u6709\u5b63\u8282\u6027\u53d8\u5316，\u4f46\u5e95\u5c42\u7269\u7406\u89c4\u5f8b\u4e0d\u53d8\u3002\u91d1\u878d\u6ca1\u6709\u4e0d\u53d8\u7684\u7269\u7406\u89c4\u5f8b\u3002<\/p>\n

3. \u5f02\u5e38\u503c\u51b2\u51fb<\/h5>\n

2020 \u5e74 3 \u6708\u7684\u7f8e\u80a1\u7194\u65ad\u30012015 \u5e74 A \u80a1\u5f02\u5e38\u6ce2\u52a8\u2014\u2014\u8fd9\u4e9b\u6781\u7aef\u4e8b\u4ef6\u4f1a\u5728\u8bad\u7ec3\u6570\u636e\u4e2d\u4ea7\u751f\u5de8\u5927\u7684\u5f02\u5e38\u503c\u3002\u5982\u679c\u7528 MSE \u635f\u5931，\u8fd9\u4e9b\u5f02\u5e38\u503c\u4f1a\u4e3b\u5bfc\u68af\u5ea6\u66f4\u65b0\u65b9\u5411，\u5bfc\u81f4\u6a21\u578b\u53c2\u6570\u5411\u9519\u8bef\u65b9\u5411\u5927\u5e45\u8df3\u52a8\u3002<\/p>\n

\u4e09\u3001\u4e2a\u6027\u5316\u4ee3\u7801\u5b9e\u8df5<\/h3>\n

import torch
\nimport torch.nn as nn
\nimport numpy as np
\nfrom typing import Tuple<\/p>\n

class StableFinancialLSTM(nn.Module):
\n """\u91d1\u878d\u65f6\u5e8f\u7a33\u5b9a\u8bad\u7ec3 LSTM"""<\/p>\n

def __init__(
\n self,
\n input_dim: int,
\n hidden_dim: int = 128,
\n num_layers: int = 2,
\n dropout: float = 0.3,
\n ):
\n super().__init__()
\n # \u8bbe\u8ba1\u539f\u56e0：\u591a\u5c42 LSTM \u6355\u6349\u4e0d\u540c\u65f6\u95f4\u5c3a\u5ea6\u7684\u6a21\u5f0f
\n # \u7b2c\u4e00\u5c42：\u77ed\u671f\u6ce2\u52a8（\u65e5\u7ea7\u522b）
\n # \u7b2c\u4e8c\u5c42：\u4e2d\u671f\u8d8b\u52bf（\u5468\/\u6708\u7ea7\u522b）
\n self.lstm = nn.LSTM(
\n input_dim,
\n hidden_dim,
\n num_layers,
\n batch_first=True,
\n dropout=dropout,
\n )
\n # \u8bbe\u8ba1\u539f\u56e0：\u68af\u5ea6\u88c1\u526a + \u6279\u5f52\u4e00\u5316\u5728 LSTM \u5916\u90e8
\n self.batch_norm = nn.BatchNorm1d(hidden_dim)
\n self.fc = nn.Linear(hidden_dim, 1)
\n self.dropout = nn.Dropout(dropout)<\/p>\n

def forward(self, x: torch.Tensor) -> torch.Tensor:
\n """
\n x: (batch, seq_len, input_dim)
\n """
\n # LSTM \u524d\u5411\u4f20\u64ad
\n lstm_out, (h_n, c_n) = self.lstm(x)<\/p>\n

# \u8bbe\u8ba1\u539f\u56e0：\u53d6\u6700\u540e\u65f6\u523b\u7684\u9690\u72b6\u6001
\n last_out = lstm_out[:, -1, :]<\/p>\n

# \u6279\u5f52\u4e00\u5316\u7a33\u5b9a\u5206\u5e03
\n last_out = self.batch_norm(last_out.unsqueeze(0)).squeeze(0)<\/p>\n

# Dropout \u6b63\u5219\u5316
\n last_out = self.dropout(last_out)<\/p>\n

# \u8f93\u51fa\u5c42
\n out = self.fc(last_out)
\n return out<\/p>\n

class StableTrainingConfig:
\n """\u7a33\u5b9a\u8bad\u7ec3\u914d\u7f6e"""<\/p>\n

def __init__(self):
\n # \u8bbe\u8ba1\u539f\u56e0：\u6bcf\u4e2a\u53c2\u6570\u90fd\u6709\u660e\u786e\u7684\u7a33\u5b9a\u5316\u76ee\u7684
\n self.grad_clip_value = 1.0 # \u68af\u5ea6\u88c1\u526a\u9608\u503c
\n self.use_huber_loss = True # \u662f\u5426\u4f7f\u7528 Huber Loss
\n self.huber_delta = 1.0 # Huber Loss \u7684 \u03b4 \u53c2\u6570
\n self.use_grad_norm = True # \u662f\u5426\u4f7f\u7528\u68af\u5ea6\u8303\u6570\u88c1\u526a
\n self.max_grad_norm = 5.0 # \u6700\u5927\u68af\u5ea6\u8303\u6570<\/p>\n

@staticmethod
\n def huber_loss(y_pred: torch.Tensor, y_true: torch.Tensor, delta: float = 1.0) -> torch.Tensor:
\n """
\n Huber \u635f\u5931\u51fd\u6570：\u5bf9\u5f02\u5e38\u503c\u9c81\u68d2<\/p>\n

\u8bbe\u8ba1\u539f\u56e0：MSE \u5bf9\u5f02\u5e38\u503c\u8fc7\u4e8e\u654f\u611f
\n Huber Loss \u5728 |error| < delta \u65f6\u7528 MSE，\u5728 |error| > delta \u65f6\u7528 MAE
\n """
\n error = y_pred – y_true
\n abs_error = torch.abs(error)<\/p>\n

# \u5c0f\u8bef\u5dee\u7528 MSE（\u5e73\u6ed1），\u5927\u8bef\u5dee\u7528 MAE（\u9c81\u68d2）
\n quadratic = 0.5 * error ** 2
\n linear = delta * (abs_error – 0.5 * delta)<\/p>\n

loss = torch.where(abs_error <= delta, quadratic, linear)
\n return loss.mean()<\/p>\n

@staticmethod
\n def quantile_loss(
\n y_pred: torch.Tensor,
\n y_true: torch.Tensor,
\n quantile: float = 0.5,
\n ) -> torch.Tensor:
\n """
\n \u5206\u4f4d\u6570\u635f\u5931：\u9002\u5408\u9884\u6d4b\u533a\u95f4<\/p>\n

\u8bbe\u8ba1\u539f\u56e0：\u91d1\u878d\u573a\u666f\u4e0d\u4ec5\u9700\u8981\u70b9\u9884\u6d4b，\u66f4\u9700\u8981\u9884\u6d4b\u533a\u95f4
\n \u5206\u4f4d\u6570\u635f\u5931\u53ef\u4ee5\u76f4\u63a5\u8f93\u51fa\u7f6e\u4fe1\u533a\u95f4
\n """
\n error = y_true – y_pred
\n loss = torch.max(
\n quantile * error,
\n (quantile – 1) * error,
\n )
\n return loss.mean()<\/p>\n

@staticmethod
\n def detect_distribution_shift(
\n old_data: np.ndarray,
\n new_data: np.ndarray,
\n threshold: float = 0.05,
\n ) -> bool:
\n """
\n \u5206\u5e03\u6f02\u79fb\u68c0\u6d4b<\/p>\n

\u8bbe\u8ba1\u539f\u56e0：\u7528 KL \u6563\u5ea6\u68c0\u6d4b\u8bad\u7ec3\u6570\u636e\u548c\u5f53\u524d\u6570\u636e\u7684\u5206\u5e03\u5dee\u5f02
\n \u5982\u679c\u6f02\u79fb\u8d85\u8fc7\u9608\u503c，\u89e6\u53d1\u91cd\u65b0\u8bad\u7ec3
\n """
\n # \u7b80\u5316\u5b9e\u73b0：\u6bd4\u8f83\u5747\u503c\u548c\u6807\u51c6\u5dee\u7684\u76f8\u5bf9\u53d8\u5316
\n old_mean, old_std = old_data.mean(), old_data.std()
\n new_mean, new_std = new_data.mean(), new_data.std()<\/p>\n

mean_shift = abs(new_mean – old_mean) \/ max(abs(old_mean), 1e-8)
\n std_shift = abs(new_std – old_std) \/ max(old_std, 1e-8)<\/p>\n

return mean_shift > threshold or std_shift > threshold<\/p>\n

# \u8bad\u7ec3\u5faa\u73af\u793a\u4f8b
\ndef train_stable_epoch(
\n model: StableFinancialLSTM,
\n dataloader: torch.utils.data.DataLoader,
\n optimizer: torch.optim.Optimizer,
\n config: StableTrainingConfig,
\n) -> float:
\n """\u4e00\u4e2a\u7a33\u5b9a\u8bad\u7ec3 epoch"""
\n model.train()
\n total_loss = 0.0<\/p>\n

for batch_x, batch_y in dataloader:
\n optimizer.zero_grad()<\/p>\n

# \u524d\u5411\u4f20\u64ad
\n y_pred = model(batch_x)<\/p>\n

# \u8ba1\u7b97\u635f\u5931（Huber Loss \u5bf9\u5f02\u5e38\u503c\u9c81\u68d2）
\n loss = config.huber_loss(
\n y_pred.squeeze(), batch_y, delta=config.huber_delta
\n )<\/p>\n

# \u53cd\u5411\u4f20\u64ad
\n loss.backward()<\/p>\n

# \u8bbe\u8ba1\u539f\u56e0：\u68af\u5ea6\u88c1\u526a\u662f\u9632\u6b62\u68af\u5ea6\u7206\u70b8\u7684\u6700\u6709\u6548\u624b\u6bb5
\n if config.use_grad_norm:
\n torch.nn.utils.clip_grad_norm_(
\n model.parameters(), config.max_grad_norm
\n )
\n else:
\n torch.nn.utils.clip_grad_value_(
\n model.parameters(), config.grad_clip_value
\n )<\/p>\n

optimizer.step()
\n total_loss += loss.item()<\/p>\n

return total_loss \/ len(dataloader)<\/p>\n

\u56db\u3001\u4e2a\u6027\u5316\u8fb9\u754c\u6743\u8861<\/h3>\n\n\u7a33\u5b9a\u7b56\u7565\u8bad\u7ec3\u901f\u5ea6\u6536\u655b\u901f\u5ea6\u5bf9\u5f02\u5e38\u503c\u9c81\u68d2\u6027\u4ee3\u4ef7<\/tr>\n\n\n\n\n\n\n\n
\u65e0\u5904\u7406<\/td>\n\u5feb<\/td>\n\u5feb（\u4f46\u53ef\u80fd\u5d29\u6e83）<\/td>\n\u6781\u5dee<\/td>\n\u8bad\u7ec3\u4e0d\u7a33\u5b9a<\/td>\n<\/tr>\n
\u4ec5\u68af\u5ea6\u88c1\u526a<\/td>\n\u4e2d<\/td>\n\u4e2d<\/td>\n\u5dee<\/td>\n\u53ef\u80fd\u9519\u8fc7\u91cd\u8981\u4fe1\u53f7<\/td>\n<\/tr>\n
Huber Loss<\/td>\n\u4e2d<\/td>\n\u4e2d<\/td>\n\u597d<\/td>\n\u8d85\u53c2\u6570 \u03b4 \u9700\u8981\u8c03\u4f18<\/td>\n<\/tr>\n
\u68af\u5ea6\u88c1\u526a + Huber<\/td>\n\u4e2d<\/td>\n\u4e2d<\/td>\n\u597d<\/td>\n\u4e24\u4e2a\u8d85\u53c2\u6570<\/td>\n<\/tr>\n
Quantile Loss<\/td>\n\u6162<\/td>\n\u6162<\/td>\n\u6781\u597d<\/td>\n\u9700\u8981\u9884\u5b9a\u4e49\u5206\u4f4d\u6570<\/td>\n<\/tr>\n
\u6279\u5f52\u4e00\u5316 + Dropout<\/td>\n\u4e2d<\/td>\n\u4e2d<\/td>\n\u4e2d<\/td>\n\u5c0f\u6279\u91cf\u65f6\u4e0d\u7a33\u5b9a<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

\u5173\u952e\u6743\u8861：<\/p>\n

  • \u9c81\u68d2\u6027 vs \u654f\u611f\u6027：Huber Loss \u727a\u7272\u4e86\u5bf9\u6b63\u5e38\u6837\u672c\u7684\u62df\u5408\u7cbe\u5ea6，\u6362\u53d6\u4e86\u5728\u5f02\u5e38\u503c\u9762\u524d\u4e0d\u5d29\u6e83\u3002\u5728\u91d1\u878d\u573a\u666f\u4e2d，\u8fd9\u4e2a\u4ee3\u4ef7\u662f\u503c\u5f97\u7684\u3002<\/li>\n
  • \u5728\u7ebf\u66f4\u65b0 vs \u6279\u91cf\u91cd\u8bad：\u5206\u5e03\u6f02\u79fb\u68c0\u6d4b\u5230\u540e，\u6709\u4e24\u79cd\u5904\u7406\u65b9\u5f0f\u2014\u2014\u5728\u7ebf\u589e\u91cf\u66f4\u65b0（\u5feb\u4f46\u53ef\u80fd\u9057\u5fd8\u65e7\u6a21\u5f0f）\u548c\u5168\u91cf\u91cd\u8bad\u7ec3（\u6162\u4f46\u66f4\u5168\u9762）\u3002\u901a\u5e38\u5efa\u8bae\u4e24\u8005\u7ed3\u5408：\u5728\u7ebf\u66f4\u65b0\u7528\u4e8e\u77ed\u671f\u9002\u914d，\u5b9a\u671f\u5168\u91cf\u91cd\u8bad\u7528\u4e8e\u957f\u671f\u7a33\u5b9a\u6027\u3002<\/li>\n
  • \u7a97\u53e3\u957f\u5ea6\u7684\u9009\u62e9：\u6ed1\u52a8\u7a97\u53e3\u592a\u957f，\u5305\u542b\u8fc7\u65f6\u7684\u6a21\u5f0f；\u7a97\u53e3\u592a\u77ed，\u65e0\u6cd5\u6355\u6349\u957f\u5468\u671f\u89c4\u5f8b\u3002\u91d1\u878d\u573a\u666f\u4e2d 2-5 \u5e74\u7684\u7a97\u53e3\u662f\u5e38\u89c1\u9009\u62e9\u3002<\/li>\n

    \u4e94\u3001\u603b\u7ed3<\/h3>\n

    \u91d1\u878d\u65f6\u5e8f\u6570\u636e\u7684\u8bad\u7ec3\u7a33\u5b9a\u6027\u95ee\u9898\u6765\u6e90\u4e8e\u4e09\u4e2a\u6839\u56e0：\u957f\u5e8f\u5217\u5bfc\u81f4\u7684\u68af\u5ea6\u6d88\u5931\/\u7206\u70b8\u3001\u5e02\u573a\u5206\u5e03\u7684\u975e\u5e73\u7a33\u6f02\u79fb\u3001\u4f4e\u9891\u4f46\u9ad8\u5f3a\u5ea6\u7684\u5f02\u5e38\u503c\u51b2\u51fb\u3002\u89e3\u51b3\u65b9\u6848\u9700\u8981\u4e09\u7ba1\u9f50\u4e0b：\u68af\u5ea6\u88c1\u526a（\u9608\u503c 1.0-5.0）\u89e3\u51b3\u68af\u5ea6\u95ee\u9898，Huber\/Quantile Loss \u66ff\u4ee3 MSE \u89e3\u51b3\u5f02\u5e38\u503c\u9c81\u68d2\u6027，\u5206\u5e03\u6f02\u79fb\u68c0\u6d4b + \u5728\u7ebf\u5b66\u4e60\u89e3\u51b3\u975e\u5e73\u7a33\u6027\u3002\u5de5\u7a0b\u4e0a，\u5efa\u8bae\u5728\u8bad\u7ec3\u5faa\u73af\u4e2d\u52a0\u5165 NaN \u68c0\u6d4b\u548c\u81ea\u52a8\u6062\u590d\u673a\u5236；\u5c06\u5206\u5e03\u6f02\u79fb\u68c0\u6d4b\u4f5c\u4e3a\u751f\u4ea7\u73af\u5883\u7684\u5e38\u9a7b\u76d1\u63a7；\u6ed1\u52a8\u7a97\u53e3\u957f\u5ea6\u5728 2-5 \u5e74\u4e4b\u95f4\u6839\u636e\u8d44\u4ea7\u7c7b\u522b\u8c03\u6574\u3002\u7a33\u5b9a\u8bad\u7ec3\u4e0d\u662f\u4e00\u6b21\u6027\u5de5\u4f5c，\u800c\u662f\u9700\u8981\u6301\u7eed\u76d1\u63a7\u7684\u7cfb\u7edf\u6027\u5de5\u7a0b\u3002<\/p>\n","protected":false},"excerpt":{"rendered":"

    \u91d1\u878d\u65f6\u5e8f\u6570\u636e\u8bad\u7ec3：\u957f\u5e8f\u5217\u5efa\u6a21\u7684\u8bad\u7ec3\u7a33\u5b9a\u6027\u95ee\u9898
    \n\u4e00\u3001\u4e2a\u6027\u5316\u6df1\u5ea6\u5f15\u8a00
    \n\u5f53\u4f60\u5728\u8bad\u7ec3\u4e00\u4e2a LSTM \u9884\u6d4b\u80a1\u7968\u4ef7\u683c\u65f6，\u7a81\u7136\u53d1\u73b0 loss \u5728\u7b2c 347 \u4e2a epoch \u7206\u70b8\u6210 NaN\u2014\u2014\u8fd9\u4e0d\u662f bug，\u8fd9\u662f\u91d1\u878d\u65f6\u5e8f\u5efa\u6a21\u7684\u65e5\u5e38\u3002
    \n\u91d1\u878d\u65f6\u95f4\u5e8f\u5217\u6709\u4e24\u4e2a\u8ba9\u6a21\u578b\u5934\u75bc\u7684\u7279\u6027：\u4e00\u662f\u957f\u7a0b\u4f9d\u8d56（3 \u5e74\u524d\u7684\u4e8b\u4ef6\u53ef\u80fd\u5f71\u54cd\u4eca\u5929\u7684\u8d70\u52bf），\u4e8c\u662f\u5206\u5e03\u6f02\u79fb（\u5e02\u573a\u7684\u7edf\u8ba1\u7279\u6027\u968f\u65f6\u95f4\u6539\u53d8）\u3002\u8fd9\u4e24\u4e2a\u7279<\/p>\n","protected":false},"author":2,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[1],"tags":[50],"topic":[],"class_list":["post-86118","post","type-post","status-publish","format-standard","hentry","category-server","tag-50"],"yoast_head":"\n\u91d1\u878d\u65f6\u5e8f\u6570\u636e\u8bad\u7ec3\uff1a\u957f\u5e8f\u5217\u5efa\u6a21\u7684\u8bad\u7ec3\u7a33\u5b9a\u6027\u95ee\u9898 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3<\/title>\n<meta name=\"robots\" content=\"index, follow, max-snippet:-1, max-image-preview:large, max-video-preview:-1\" \/>\n<link rel=\"canonical\" href=\"https:\/\/www.wsisp.com\/helps\/86118.html\" \/>\n<meta property=\"og:locale\" content=\"zh_CN\" \/>\n<meta property=\"og:type\" content=\"article\" \/>\n<meta property=\"og:title\" content=\"\u91d1\u878d\u65f6\u5e8f\u6570\u636e\u8bad\u7ec3\uff1a\u957f\u5e8f\u5217\u5efa\u6a21\u7684\u8bad\u7ec3\u7a33\u5b9a\u6027\u95ee\u9898 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\" \/>\n<meta property=\"og:description\" content=\"\u91d1\u878d\u65f6\u5e8f\u6570\u636e\u8bad\u7ec3：\u957f\u5e8f\u5217\u5efa\u6a21\u7684\u8bad\u7ec3\u7a33\u5b9a\u6027\u95ee\u9898 \u4e00\u3001\u4e2a\u6027\u5316\u6df1\u5ea6\u5f15\u8a00 \u5f53\u4f60\u5728\u8bad\u7ec3\u4e00\u4e2a LSTM \u9884\u6d4b\u80a1\u7968\u4ef7\u683c\u65f6，\u7a81\u7136\u53d1\u73b0 loss \u5728\u7b2c 347 \u4e2a epoch \u7206\u70b8\u6210 NaN\u2014\u2014\u8fd9\u4e0d\u662f bug，\u8fd9\u662f\u91d1\u878d\u65f6\u5e8f\u5efa\u6a21\u7684\u65e5\u5e38\u3002 \u91d1\u878d\u65f6\u95f4\u5e8f\u5217\u6709\u4e24\u4e2a\u8ba9\u6a21\u578b\u5934\u75bc\u7684\u7279\u6027：\u4e00\u662f\u957f\u7a0b\u4f9d\u8d56（3 \u5e74\u524d\u7684\u4e8b\u4ef6\u53ef\u80fd\u5f71\u54cd\u4eca\u5929\u7684\u8d70\u52bf），\u4e8c\u662f\u5206\u5e03\u6f02\u79fb（\u5e02\u573a\u7684\u7edf\u8ba1\u7279\u6027\u968f\u65f6\u95f4\u6539\u53d8）\u3002\u8fd9\u4e24\u4e2a\u7279\" \/>\n<meta property=\"og:url\" content=\"https:\/\/www.wsisp.com\/helps\/86118.html\" \/>\n<meta property=\"og:site_name\" content=\"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\" \/>\n<meta property=\"article:published_time\" content=\"2026-07-27T21:40:34+00:00\" \/>\n<meta name=\"author\" content=\"admin\" \/>\n<meta name=\"twitter:card\" content=\"summary_large_image\" \/>\n<meta name=\"twitter:label1\" content=\"\u4f5c\u8005\" \/>\n\t<meta name=\"twitter:data1\" content=\"admin\" \/>\n\t<meta name=\"twitter:label2\" content=\"\u9884\u8ba1\u9605\u8bfb\u65f6\u95f4\" \/>\n\t<meta name=\"twitter:data2\" content=\"3 \u5206\" \/>\n<script type=\"application\/ld+json\" class=\"yoast-schema-graph\">{\"@context\":\"https:\/\/schema.org\",\"@graph\":[{\"@type\":\"WebPage\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/86118.html\",\"url\":\"https:\/\/www.wsisp.com\/helps\/86118.html\",\"name\":\"\u91d1\u878d\u65f6\u5e8f\u6570\u636e\u8bad\u7ec3\uff1a\u957f\u5e8f\u5217\u5efa\u6a21\u7684\u8bad\u7ec3\u7a33\u5b9a\u6027\u95ee\u9898 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\",\"isPartOf\":{\"@id\":\"https:\/\/www.wsisp.com\/helps\/#website\"},\"datePublished\":\"2026-07-27T21:40:34+00:00\",\"dateModified\":\"2026-07-27T21:40:34+00:00\",\"author\":{\"@id\":\"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41\"},\"breadcrumb\":{\"@id\":\"https:\/\/www.wsisp.com\/helps\/86118.html#breadcrumb\"},\"inLanguage\":\"zh-Hans\",\"potentialAction\":[{\"@type\":\"ReadAction\",\"target\":[\"https:\/\/www.wsisp.com\/helps\/86118.html\"]}]},{\"@type\":\"BreadcrumbList\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/86118.html#breadcrumb\",\"itemListElement\":[{\"@type\":\"ListItem\",\"position\":1,\"name\":\"\u9996\u9875\",\"item\":\"https:\/\/www.wsisp.com\/helps\"},{\"@type\":\"ListItem\",\"position\":2,\"name\":\"\u91d1\u878d\u65f6\u5e8f\u6570\u636e\u8bad\u7ec3\uff1a\u957f\u5e8f\u5217\u5efa\u6a21\u7684\u8bad\u7ec3\u7a33\u5b9a\u6027\u95ee\u9898\"}]},{\"@type\":\"WebSite\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/#website\",\"url\":\"https:\/\/www.wsisp.com\/helps\/\",\"name\":\"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\",\"description\":\"\u9999\u6e2f\u670d\u52a1\u5668_\u9999\u6e2f\u4e91\u670d\u52a1\u5668\u8d44\u8baf_\u670d\u52a1\u5668\u5e2e\u52a9\u6587\u6863_\u670d\u52a1\u5668\u6559\u7a0b\",\"potentialAction\":[{\"@type\":\"SearchAction\",\"target\":{\"@type\":\"EntryPoint\",\"urlTemplate\":\"https:\/\/www.wsisp.com\/helps\/?s={search_term_string}\"},\"query-input\":\"required name=search_term_string\"}],\"inLanguage\":\"zh-Hans\"},{\"@type\":\"Person\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41\",\"name\":\"admin\",\"image\":{\"@type\":\"ImageObject\",\"inLanguage\":\"zh-Hans\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/image\/\",\"url\":\"https:\/\/gravatar.wp-china-yes.net\/avatar\/?s=96&d=mystery\",\"contentUrl\":\"https:\/\/gravatar.wp-china-yes.net\/avatar\/?s=96&d=mystery\",\"caption\":\"admin\"},\"sameAs\":[\"http:\/\/wp.wsisp.com\"],\"url\":\"https:\/\/www.wsisp.com\/helps\/author\/admin\"}]}<\/script>\n<!-- \/ Yoast SEO plugin. -->","yoast_head_json":{"title":"\u91d1\u878d\u65f6\u5e8f\u6570\u636e\u8bad\u7ec3\uff1a\u957f\u5e8f\u5217\u5efa\u6a21\u7684\u8bad\u7ec3\u7a33\u5b9a\u6027\u95ee\u9898 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","robots":{"index":"index","follow":"follow","max-snippet":"max-snippet:-1","max-image-preview":"max-image-preview:large","max-video-preview":"max-video-preview:-1"},"canonical":"https:\/\/www.wsisp.com\/helps\/86118.html","og_locale":"zh_CN","og_type":"article","og_title":"\u91d1\u878d\u65f6\u5e8f\u6570\u636e\u8bad\u7ec3\uff1a\u957f\u5e8f\u5217\u5efa\u6a21\u7684\u8bad\u7ec3\u7a33\u5b9a\u6027\u95ee\u9898 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","og_description":"\u91d1\u878d\u65f6\u5e8f\u6570\u636e\u8bad\u7ec3：\u957f\u5e8f\u5217\u5efa\u6a21\u7684\u8bad\u7ec3\u7a33\u5b9a\u6027\u95ee\u9898 \u4e00\u3001\u4e2a\u6027\u5316\u6df1\u5ea6\u5f15\u8a00 \u5f53\u4f60\u5728\u8bad\u7ec3\u4e00\u4e2a LSTM \u9884\u6d4b\u80a1\u7968\u4ef7\u683c\u65f6，\u7a81\u7136\u53d1\u73b0 loss \u5728\u7b2c 347 \u4e2a epoch \u7206\u70b8\u6210 NaN\u2014\u2014\u8fd9\u4e0d\u662f bug，\u8fd9\u662f\u91d1\u878d\u65f6\u5e8f\u5efa\u6a21\u7684\u65e5\u5e38\u3002 \u91d1\u878d\u65f6\u95f4\u5e8f\u5217\u6709\u4e24\u4e2a\u8ba9\u6a21\u578b\u5934\u75bc\u7684\u7279\u6027：\u4e00\u662f\u957f\u7a0b\u4f9d\u8d56（3 \u5e74\u524d\u7684\u4e8b\u4ef6\u53ef\u80fd\u5f71\u54cd\u4eca\u5929\u7684\u8d70\u52bf），\u4e8c\u662f\u5206\u5e03\u6f02\u79fb（\u5e02\u573a\u7684\u7edf\u8ba1\u7279\u6027\u968f\u65f6\u95f4\u6539\u53d8）\u3002\u8fd9\u4e24\u4e2a\u7279","og_url":"https:\/\/www.wsisp.com\/helps\/86118.html","og_site_name":"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","article_published_time":"2026-07-27T21:40:34+00:00","author":"admin","twitter_card":"summary_large_image","twitter_misc":{"\u4f5c\u8005":"admin","\u9884\u8ba1\u9605\u8bfb\u65f6\u95f4":"3 \u5206"},"schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":"WebPage","@id":"https:\/\/www.wsisp.com\/helps\/86118.html","url":"https:\/\/www.wsisp.com\/helps\/86118.html","name":"\u91d1\u878d\u65f6\u5e8f\u6570\u636e\u8bad\u7ec3\uff1a\u957f\u5e8f\u5217\u5efa\u6a21\u7684\u8bad\u7ec3\u7a33\u5b9a\u6027\u95ee\u9898 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","isPartOf":{"@id":"https:\/\/www.wsisp.com\/helps\/#website"},"datePublished":"2026-07-27T21:40:34+00:00","dateModified":"2026-07-27T21:40:34+00:00","author":{"@id":"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41"},"breadcrumb":{"@id":"https:\/\/www.wsisp.com\/helps\/86118.html#breadcrumb"},"inLanguage":"zh-Hans","potentialAction":[{"@type":"ReadAction","target":["https:\/\/www.wsisp.com\/helps\/86118.html"]}]},{"@type":"BreadcrumbList","@id":"https:\/\/www.wsisp.com\/helps\/86118.html#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"\u9996\u9875","item":"https:\/\/www.wsisp.com\/helps"},{"@type":"ListItem","position":2,"name":"\u91d1\u878d\u65f6\u5e8f\u6570\u636e\u8bad\u7ec3\uff1a\u957f\u5e8f\u5217\u5efa\u6a21\u7684\u8bad\u7ec3\u7a33\u5b9a\u6027\u95ee\u9898"}]},{"@type":"WebSite","@id":"https:\/\/www.wsisp.com\/helps\/#website","url":"https:\/\/www.wsisp.com\/helps\/","name":"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","description":"\u9999\u6e2f\u670d\u52a1\u5668_\u9999\u6e2f\u4e91\u670d\u52a1\u5668\u8d44\u8baf_\u670d\u52a1\u5668\u5e2e\u52a9\u6587\u6863_\u670d\u52a1\u5668\u6559\u7a0b","potentialAction":[{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https:\/\/www.wsisp.com\/helps\/?s={search_term_string}"},"query-input":"required name=search_term_string"}],"inLanguage":"zh-Hans"},{"@type":"Person","@id":"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41","name":"admin","image":{"@type":"ImageObject","inLanguage":"zh-Hans","@id":"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/image\/","url":"https:\/\/gravatar.wp-china-yes.net\/avatar\/?s=96&d=mystery","contentUrl":"https:\/\/gravatar.wp-china-yes.net\/avatar\/?s=96&d=mystery","caption":"admin"},"sameAs":["http:\/\/wp.wsisp.com"],"url":"https:\/\/www.wsisp.com\/helps\/author\/admin"}]}},"_links":{"self":[{"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/posts\/86118","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/comments?post=86118"}],"version-history":[{"count":0,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/posts\/86118\/revisions"}],"wp:attachment":[{"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/media?parent=86118"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/categories?post=86118"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/tags?post=86118"},{"taxonomy":"topic","embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/topic?post=86118"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}