python打卡训练营打卡记录day37

知识点回顾：
  1. 过拟合的判断：测试集和训练集同步打印指标
  2. 模型的保存和加载
    1. 仅保存权重
    2. 保存权重和模型
    3. 保存全部信息checkpoint，还包含训练状态
  3. 早停策略

作业：对信贷数据集训练后保存权重，加载权重后继续训练50轮，并采取早停策略

import pandas as pd    #用于数据处理和分析，可处理表格数据。
import numpy as np     #用于数值计算，提供了高效的数组操作。
import matplotlib.pyplot as plt    #用于绘制各种类型的图表
import seaborn as sns   #基于matplotlib的高级绘图库，能绘制更美观的统计图形。
import warnings
warnings.filterwarnings("ignore")import torch
import torch.nn as nn
import torch.optim as optim
from sklearn.datasets import load_iris
from sklearn.model_selection import train_test_split
from sklearn.preprocessing import MinMaxScaler
import time
from tqdm import tqdm  # 导入tqdm库用于进度条显示
import os
from sklearn.metrics import accuracy_score# 设置中文字体（解决中文显示问题）
plt.rcParams['font.sans-serif'] = ['SimHei']  # Windows系统常用黑体字体
plt.rcParams['axes.unicode_minus'] = False    # 正常显示负号
data = pd.read_csv('data.csv')    #读取数据# 设置GPU设备
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
print(f"使用设备: {device}")# 先筛选字符串变量 
discrete_features = data.select_dtypes(include=['object']).columns.tolist()
# Home Ownership 标签编码
home_ownership_mapping = {'Own Home': 1,'Rent': 2,'Have Mortgage': 3,'Home Mortgage': 4
}
data['Home Ownership'] = data['Home Ownership'].map(home_ownership_mapping)# Years in current job 标签编码
years_in_job_mapping = {'< 1 year': 1,'1 year': 2,'2 years': 3,'3 years': 4,'4 years': 5,'5 years': 6,'6 years': 7,'7 years': 8,'8 years': 9,'9 years': 10,'10+ years': 11
}
data['Years in current job'] = data['Years in current job'].map(years_in_job_mapping)# Purpose 独热编码，记得需要将bool类型转换为数值
data = pd.get_dummies(data, columns=['Purpose'])
data2 = pd.read_csv("data.csv") # 重新读取数据，用来做列名对比
list_final = [] # 新建一个空列表，用于存放独热编码后新增的特征名
for i in data.columns:if i not in data2.columns:list_final.append(i) # 这里打印出来的就是独热编码后的特征名
for i in list_final:data[i] = data[i].astype(int) # 这里的i就是独热编码后的特征名# Term 0 - 1 映射
term_mapping = {'Short Term': 0,'Long Term': 1
}
data['Term'] = data['Term'].map(term_mapping)
data.rename(columns={'Term': 'Long Term'}, inplace=True) # 重命名列
continuous_features = data.select_dtypes(include=['int64', 'float64']).columns.tolist()  #把筛选出来的列名转换成列表# 连续特征用中位数补全
for feature in continuous_features:     mode_value = data[feature].mode()[0]            #获取该列的众数。data[feature].fillna(mode_value, inplace=True)          #用众数填充该列的缺失值，inplace=True表示直接在原数据上修改。# 最开始也说了 很多调参函数自带交叉验证，甚至是必选的参数，你如果想要不交叉反而实现起来会麻烦很多
# 所以这里我们还是只划分一次数据集
from sklearn.model_selection import train_test_split
X = data.drop(['Credit Default'], axis=1)  # 特征，axis=1表示按列删除
y = data['Credit Default'] # 标签
# 按照8:2划分训练集和测试集
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=42)  # 80%训练集，20%测试集
# 划分训练集和测试集
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=42)# 归一化数据
scaler = MinMaxScaler()
X_train = scaler.fit_transform(X_train)
X_test = scaler.transform(X_test)
X_train = torch.FloatTensor(X_train).to(device)
y_train = torch.LongTensor(y_train.values).to(device) 
X_test = torch.FloatTensor(X_test).to(device)
y_test = torch.LongTensor(y_test.values).to(device) # 模型定义
class MLP(nn.Module):def __init__(self, input_size):super(MLP, self).__init__()self.fc1 = nn.Linear(input_size, 64)self.relu = nn.ReLU()self.fc2 = nn.Linear(64, 2)def forward(self, x):x = self.relu(self.fc1(x))x = self.fc2(x)return x# -------------------- 初始训练阶段（20000轮） --------------------
def initial_training():# 实例化模型并移至GPUmodel = MLP(input_size=X_train.shape[1]).to(device)criterion = nn.CrossEntropyLoss()optimizer = optim.SGD(model.parameters(), lr=0.01)# 初始化记录列表train_losses = []test_losses = []test_accuracies = []eval_epochs = []# 训练参数num_epochs = 20000best_test_loss = float('inf')patience = 50counter = 0# 训练循环with tqdm(total=num_epochs, desc="初始训练进度") as pbar:for epoch in range(num_epochs):# 训练步骤model.train()outputs = model(X_train)train_loss = criterion(outputs, y_train)# 反向传播optimizer.zero_grad()train_loss.backward()optimizer.step()# 每200轮评估测试损失if (epoch + 1) % 200 == 0:model.eval()with torch.no_grad():test_outputs = model(X_test)test_loss = criterion(test_outputs, y_test)# 准确率计算_, predicted = torch.max(test_outputs, 1)accuracy = accuracy_score(y_test.cpu(), predicted.cpu())model.train()# 打印准确率print(f"Epoch [{epoch+1}/{num_epochs}], Test Accuracy: {accuracy:.4f}")# 记录指标train_losses.append(train_loss.item())test_losses.append(test_loss.item())test_accuracies.append(accuracy)eval_epochs.append(epoch + 1)# 保存最佳检查点if test_loss < best_test_loss:best_test_loss = test_loss.item()torch.save({'model_state_dict': model.state_dict(),'optimizer_state_dict': optimizer.state_dict(),'best_test_loss': best_test_loss,'epoch': epoch}, 'best_model.pth')counter = 0else:counter += 1if counter >= patience:print(f"\n初始训练早停于第{epoch+1}轮")break# 更新进度条pbar.set_postfix({'Train Loss': f'{train_loss.item():.4f}', 'Test Loss': f'{test_loss.item():.4f}'})# 更新进度if (epoch + 1) % 1000 == 0:pbar.update(1000)print("初始训练完成，最佳模型已保存")plot_metrics(train_losses, test_losses, test_accuracies, eval_epochs, title_prefix="Initial Training")# -------------------- 绘图函数 --------------------
def plot_metrics(train_losses, test_losses, test_accuracies, eval_epochs, title_prefix=""):plt.figure(figsize=(12, 5))# Loss 曲线plt.subplot(1, 2, 1)plt.plot(eval_epochs, train_losses, label='Train Loss')plt.plot(eval_epochs, test_losses, label='Test Loss')plt.xlabel('Epoch')plt.ylabel('Loss')plt.title(f'{title_prefix} Loss Curve')plt.legend()# Accuracy 曲线plt.subplot(1, 2, 2)plt.plot(eval_epochs, test_accuracies, label='Test Accuracy')plt.xlabel('Epoch')plt.ylabel('Accuracy')plt.title(f'{title_prefix} Accuracy Curve')plt.legend()plt.tight_layout()plt.show()# -------------------- 继续训练阶段（50轮） --------------------  
def continue_training():# 重新实例化模型model = MLP(input_size=X_train.shape[1]).to(device)criterion = nn.CrossEntropyLoss()optimizer = optim.SGD(model.parameters(), lr=0.01)# 初始化记录列表train_losses = []test_losses = []test_accuracies = []eval_epochs = []# 加载检查点if not os.path.exists('best_model.pth'):raise FileNotFoundError("找不到检查点文件，请先运行初始训练")checkpoint = torch.load('best_model.pth')try:model.load_state_dict(checkpoint['model_state_dict'])optimizer.load_state_dict(checkpoint['optimizer_state_dict'])except KeyError:# 兼容旧版检查点model.load_state_dict(torch.load('best_model.pth'))# 重置训练参数num_epochs = 50best_test_loss = float('inf')  # 独立重置patience = 10counter = 0# 继续训练循环with tqdm(total=num_epochs, desc="继续训练进度") as pbar:for epoch in range(num_epochs):# 训练步骤model.train()outputs = model(X_train)train_loss = criterion(outputs, y_train)# 反向传播optimizer.zero_grad()train_loss.backward()optimizer.step()# 评估测试损失和准确率model.eval()with torch.no_grad():test_outputs = model(X_test)test_loss = criterion(test_outputs, y_test)_, predicted = torch.max(test_outputs, 1)accuracy = accuracy_score(y_test.cpu(), predicted.cpu())print(f"Epoch [{epoch+1}/{num_epochs}], Test Accuracy: {accuracy:.4f}")# 记录train_losses.append(train_loss.item())test_losses.append(test_loss.item())test_accuracies.append(accuracy)eval_epochs.append(epoch + 1)model.train()# 早停判断（仅用新参数）if test_loss < best_test_loss:best_test_loss = test_loss.item()counter = 0else:counter += 1if counter >= patience:print(f"\n继续训练早停于第{epoch+1}轮")break# 更新进度pbar.set_postfix({'Train Loss': f'{train_loss.item():.4f}','Test Loss': f'{test_loss.item():.4f}','Test Acc': f'{accuracy:.4f}' })pbar.update(1)print("继续训练完成")plot_metrics(train_losses, test_losses, test_accuracies, eval_epochs, title_prefix="Continue Training")if __name__ == "__main__":initial_training()
初始训练进度:   0%|          | 0/20000 [00:00<?, ?it/s, Train Loss=0.5753, Test Loss=0.5885]
Epoch [200/20000], Test Accuracy: 0.7060
Epoch [400/20000], Test Accuracy: 0.7060
初始训练进度:   0%|          | 0/20000 [00:00<?, ?it/s, Train Loss=0.5611, Test Loss=0.5740]
Epoch [600/20000], Test Accuracy: 0.7060
Epoch [800/20000], Test Accuracy: 0.7060
初始训练进度:   5%|▌         | 1000/20000 [00:01<00:16, 1168.82it/s, Train Loss=0.5458, Test Loss=0.5581]
Epoch [1000/20000], Test Accuracy: 0.7060
Epoch [1200/20000], Test Accuracy: 0.7060
初始训练进度:   5%|▌         | 1000/20000 [00:01<00:16, 1168.82it/s, Train Loss=0.5312, Test Loss=0.5427]
Epoch [1400/20000], Test Accuracy: 0.7140
Epoch [1600/20000], Test Accuracy: 0.7207
初始训练进度:  10%|█         | 2000/20000 [00:01<00:13, 1303.08it/s, Train Loss=0.5185, Test Loss=0.5291]
Epoch [1800/20000], Test Accuracy: 0.7267
Epoch [2000/20000], Test Accuracy: 0.7407
初始训练进度:  10%|█         | 2000/20000 [00:01<00:13, 1303.08it/s, Train Loss=0.5080, Test Loss=0.5179]
Epoch [2200/20000], Test Accuracy: 0.7473
Epoch [2400/20000], Test Accuracy: 0.7593
初始训练进度:  10%|█         | 2000/20000 [00:02<00:13, 1303.08it/s, Train Loss=0.4996, Test Loss=0.5090]
Epoch [2600/20000], Test Accuracy: 0.7640
Epoch [2800/20000], Test Accuracy: 0.7640
初始训练进度:  15%|█▌        | 3000/20000 [00:02<00:12, 1378.59it/s, Train Loss=0.4930, Test Loss=0.5021]
Epoch [3000/20000], Test Accuracy: 0.7653
Epoch [3200/20000], Test Accuracy: 0.7647
初始训练进度:  15%|█▌        | 3000/20000 [00:02<00:12, 1378.59it/s, Train Loss=0.4879, Test Loss=0.4968]
Epoch [3400/20000], Test Accuracy: 0.7653
Epoch [3600/20000], Test Accuracy: 0.7660
初始训练进度:  20%|██        | 4000/20000 [00:02<00:11, 1411.22it/s, Train Loss=0.4840, Test Loss=0.4927]
Epoch [3800/20000], Test Accuracy: 0.7667
Epoch [4000/20000], Test Accuracy: 0.7667
初始训练进度:  20%|██        | 4000/20000 [00:03<00:11, 1411.22it/s, Train Loss=0.4810, Test Loss=0.4895]
Epoch [4200/20000], Test Accuracy: 0.7667
Epoch [4400/20000], Test Accuracy: 0.7667
初始训练进度:  20%|██        | 4000/20000 [00:03<00:11, 1411.22it/s, Train Loss=0.4787, Test Loss=0.4871]
Epoch [4600/20000], Test Accuracy: 0.7667
Epoch [4800/20000], Test Accuracy: 0.7667
初始训练进度:  25%|██▌       | 5000/20000 [00:03<00:10, 1454.26it/s, Train Loss=0.4768, Test Loss=0.4852]
Epoch [5000/20000], Test Accuracy: 0.7673
Epoch [5200/20000], Test Accuracy: 0.7673
初始训练进度:  25%|██▌       | 5000/20000 [00:03<00:10, 1454.26it/s, Train Loss=0.4753, Test Loss=0.4836]
Epoch [5400/20000], Test Accuracy: 0.7673
Epoch [5600/20000], Test Accuracy: 0.7673
初始训练进度:  30%|███       | 6000/20000 [00:04<00:09, 1504.37it/s, Train Loss=0.4740, Test Loss=0.4822]
Epoch [5800/20000], Test Accuracy: 0.7673
Epoch [6000/20000], Test Accuracy: 0.7673
初始训练进度:  30%|███       | 6000/20000 [00:04<00:09, 1504.37it/s, Train Loss=0.4730, Test Loss=0.4811]
Epoch [6200/20000], Test Accuracy: 0.7673
Epoch [6400/20000], Test Accuracy: 0.7673
初始训练进度:  30%|███       | 6000/20000 [00:04<00:09, 1504.37it/s, Train Loss=0.4720, Test Loss=0.4802]
Epoch [6600/20000], Test Accuracy: 0.7673
Epoch [6800/20000], Test Accuracy: 0.7660
初始训练进度:  35%|███▌      | 7000/20000 [00:05<00:08, 1451.45it/s, Train Loss=0.4712, Test Loss=0.4794]
Epoch [7000/20000], Test Accuracy: 0.7660
Epoch [7200/20000], Test Accuracy: 0.7660
初始训练进度:  35%|███▌      | 7000/20000 [00:05<00:08, 1451.45it/s, Train Loss=0.4705, Test Loss=0.4787]
Epoch [7400/20000], Test Accuracy: 0.7660
Epoch [7600/20000], Test Accuracy: 0.7660
初始训练进度:  40%|████      | 8000/20000 [00:05<00:08, 1448.11it/s, Train Loss=0.4699, Test Loss=0.4780]
Epoch [7800/20000], Test Accuracy: 0.7660
Epoch [8000/20000], Test Accuracy: 0.7660
初始训练进度:  40%|████      | 8000/20000 [00:05<00:08, 1448.11it/s, Train Loss=0.4693, Test Loss=0.4775]
Epoch [8200/20000], Test Accuracy: 0.7660
Epoch [8400/20000], Test Accuracy: 0.7660
初始训练进度:  40%|████      | 8000/20000 [00:06<00:08, 1448.11it/s, Train Loss=0.4688, Test Loss=0.4769]
Epoch [8600/20000], Test Accuracy: 0.7660
Epoch [8800/20000], Test Accuracy: 0.7667
初始训练进度:  45%|████▌     | 9000/20000 [00:06<00:07, 1468.30it/s, Train Loss=0.4683, Test Loss=0.4765]
Epoch [9000/20000], Test Accuracy: 0.7667
Epoch [9200/20000], Test Accuracy: 0.7667
初始训练进度:  45%|████▌     | 9000/20000 [00:06<00:07, 1468.30it/s, Train Loss=0.4679, Test Loss=0.4761]
Epoch [9400/20000], Test Accuracy: 0.7667
Epoch [9600/20000], Test Accuracy: 0.7667
初始训练进度:  50%|█████     | 10000/20000 [00:06<00:06, 1505.67it/s, Train Loss=0.4675, Test Loss=0.4757]
Epoch [9800/20000], Test Accuracy: 0.7660
Epoch [10000/20000], Test Accuracy: 0.7667
初始训练进度:  50%|█████     | 10000/20000 [00:07<00:06, 1505.67it/s, Train Loss=0.4671, Test Loss=0.4753]
Epoch [10200/20000], Test Accuracy: 0.7667
Epoch [10400/20000], Test Accuracy: 0.7667
初始训练进度:  50%|█████     | 10000/20000 [00:07<00:06, 1505.67it/s, Train Loss=0.4667, Test Loss=0.4750]
Epoch [10600/20000], Test Accuracy: 0.7673
Epoch [10800/20000], Test Accuracy: 0.7680
初始训练进度:  55%|█████▌    | 11000/20000 [00:07<00:05, 1539.84it/s, Train Loss=0.4664, Test Loss=0.4747]
Epoch [11000/20000], Test Accuracy: 0.7680
Epoch [11200/20000], Test Accuracy: 0.7680
初始训练进度:  55%|█████▌    | 11000/20000 [00:07<00:05, 1539.84it/s, Train Loss=0.4661, Test Loss=0.4744]
Epoch [11400/20000], Test Accuracy: 0.7680
Epoch [11600/20000], Test Accuracy: 0.7680
初始训练进度:  60%|██████    | 12000/20000 [00:08<00:05, 1584.28it/s, Train Loss=0.4658, Test Loss=0.4742]
Epoch [11800/20000], Test Accuracy: 0.7687
Epoch [12000/20000], Test Accuracy: 0.7680
初始训练进度:  60%|██████    | 12000/20000 [00:08<00:05, 1584.28it/s, Train Loss=0.4655, Test Loss=0.4739]
Epoch [12200/20000], Test Accuracy: 0.7680
Epoch [12400/20000], Test Accuracy: 0.7680
初始训练进度:  60%|██████    | 12000/20000 [00:08<00:05, 1584.28it/s, Train Loss=0.4652, Test Loss=0.4737]
Epoch [12600/20000], Test Accuracy: 0.7680
Epoch [12800/20000], Test Accuracy: 0.7687
初始训练进度:  65%|██████▌   | 13000/20000 [00:08<00:04, 1618.84it/s, Train Loss=0.4650, Test Loss=0.4735]
Epoch [13000/20000], Test Accuracy: 0.7687
Epoch [13200/20000], Test Accuracy: 0.7687
初始训练进度:  65%|██████▌   | 13000/20000 [00:09<00:04, 1618.84it/s, Train Loss=0.4647, Test Loss=0.4733]
Epoch [13400/20000], Test Accuracy: 0.7687
Epoch [13600/20000], Test Accuracy: 0.7687
初始训练进度:  70%|███████   | 14000/20000 [00:09<00:03, 1635.55it/s, Train Loss=0.4645, Test Loss=0.4731]
Epoch [13800/20000], Test Accuracy: 0.7687
Epoch [14000/20000], Test Accuracy: 0.7693
初始训练进度:  70%|███████   | 14000/20000 [00:09<00:03, 1635.55it/s, Train Loss=0.4643, Test Loss=0.4730]
Epoch [14200/20000], Test Accuracy: 0.7687
Epoch [14400/20000], Test Accuracy: 0.7687
初始训练进度:  70%|███████   | 14000/20000 [00:09<00:03, 1635.55it/s, Train Loss=0.4641, Test Loss=0.4728]
Epoch [14600/20000], Test Accuracy: 0.7687
Epoch [14800/20000], Test Accuracy: 0.7687
初始训练进度:  75%|███████▌  | 15000/20000 [00:10<00:03, 1641.97it/s, Train Loss=0.4639, Test Loss=0.4727]
Epoch [15000/20000], Test Accuracy: 0.7687
Epoch [15200/20000], Test Accuracy: 0.7687
初始训练进度:  75%|███████▌  | 15000/20000 [00:10<00:03, 1641.97it/s, Train Loss=0.4637, Test Loss=0.4725]
Epoch [15400/20000], Test Accuracy: 0.7700
Epoch [15600/20000], Test Accuracy: 0.7707
初始训练进度:  80%|████████  | 16000/20000 [00:10<00:02, 1630.00it/s, Train Loss=0.4635, Test Loss=0.4724]
Epoch [15800/20000], Test Accuracy: 0.7707
Epoch [16000/20000], Test Accuracy: 0.7707
初始训练进度:  80%|████████  | 16000/20000 [00:10<00:02, 1630.00it/s, Train Loss=0.4633, Test Loss=0.4723]
Epoch [16200/20000], Test Accuracy: 0.7707
Epoch [16400/20000], Test Accuracy: 0.7707
初始训练进度:  80%|████████  | 16000/20000 [00:11<00:02, 1630.00it/s, Train Loss=0.4631, Test Loss=0.4722]
Epoch [16600/20000], Test Accuracy: 0.7707
Epoch [16800/20000], Test Accuracy: 0.7707
初始训练进度:  85%|████████▌ | 17000/20000 [00:11<00:01, 1629.38it/s, Train Loss=0.4629, Test Loss=0.4721]
Epoch [17000/20000], Test Accuracy: 0.7707
Epoch [17200/20000], Test Accuracy: 0.7707
初始训练进度:  85%|████████▌ | 17000/20000 [00:11<00:01, 1629.38it/s, Train Loss=0.4628, Test Loss=0.4720]
Epoch [17400/20000], Test Accuracy: 0.7707
Epoch [17600/20000], Test Accuracy: 0.7707
初始训练进度:  90%|█████████ | 18000/20000 [00:11<00:01, 1594.18it/s, Train Loss=0.4626, Test Loss=0.4719]
Epoch [17800/20000], Test Accuracy: 0.7707
Epoch [18000/20000], Test Accuracy: 0.7700
初始训练进度:  90%|█████████ | 18000/20000 [00:12<00:01, 1594.18it/s, Train Loss=0.4624, Test Loss=0.4718]
Epoch [18200/20000], Test Accuracy: 0.7700
Epoch [18400/20000], Test Accuracy: 0.7700
初始训练进度:  90%|█████████ | 18000/20000 [00:12<00:01, 1594.18it/s, Train Loss=0.4623, Test Loss=0.4717]
Epoch [18600/20000], Test Accuracy: 0.7693
Epoch [18800/20000], Test Accuracy: 0.7693
初始训练进度:  95%|█████████▌| 19000/20000 [00:12<00:00, 1597.16it/s, Train Loss=0.4621, Test Loss=0.4717]
Epoch [19000/20000], Test Accuracy: 0.7693
Epoch [19200/20000], Test Accuracy: 0.7693
初始训练进度:  95%|█████████▌| 19000/20000 [00:12<00:00, 1597.16it/s, Train Loss=0.4620, Test Loss=0.4716]
Epoch [19400/20000], Test Accuracy: 0.7693
Epoch [19600/20000], Test Accuracy: 0.7693
初始训练进度: 100%|██████████| 20000/20000 [00:13<00:00, 1534.45it/s, Train Loss=0.4619, Test Loss=0.4715]
Epoch [19800/20000], Test Accuracy: 0.7693
Epoch [20000/20000], Test Accuracy: 0.7693
初始训练完成，最佳模型已保存

continue_training()
Epoch [29/50], Test Accuracy: 0.7693
Epoch [30/50], Test Accuracy: 0.7693
Epoch [31/50], Test Accuracy: 0.7693
Epoch [32/50], Test Accuracy: 0.7693
Epoch [33/50], Test Accuracy: 0.7693
Epoch [34/50], Test Accuracy: 0.7693
Epoch [35/50], Test Accuracy: 0.7693
Epoch [36/50], Test Accuracy: 0.7693
Epoch [37/50], Test Accuracy: 0.7693
Epoch [38/50], Test Accuracy: 0.7693
Epoch [39/50], Test Accuracy: 0.7693
Epoch [40/50], Test Accuracy: 0.7693
Epoch [41/50], Test Accuracy: 0.7693
Epoch [42/50], Test Accuracy: 0.7693
Epoch [43/50], Test Accuracy: 0.7693
Epoch [44/50], Test Accuracy: 0.7693
Epoch [45/50], Test Accuracy: 0.7693
Epoch [46/50], Test Accuracy: 0.7693
Epoch [47/50], Test Accuracy: 0.7693
Epoch [48/50], Test Accuracy: 0.7693
Epoch [49/50], Test Accuracy: 0.7693
Epoch [50/50], Test Accuracy: 0.7693
继续训练完成

总结

一、初始训练阶段分析（20000轮）

1. 损失曲线表现

  • 训练损失：从约0.60稳步下降至0.46，说明模型在训练集上持续优化。

  • 测试损失：从0.58下降至0.47，但下降幅度较小，且在训练后期趋于稳定。

  • 关键问题：

    • 训练损失和测试损失的差距较小（0.46 vs 0.47），表明模型未明显过拟合，但学习能力可能不足。

    • 测试损失未进一步下降，可能是模型复杂度过低或数据特征难以捕捉。

2. 准确率曲线表现

  • 测试准确率：最终稳定在 76.93%，未突破77%。

  • 关键问题：

    • 准确率曲线在训练后期完全平坦，表明模型已收敛到局部最优，但性能瓶颈明显。

    • 可能原因：模型结构简单（仅两层全连接）、特征工程不足或类别不均衡。

二、继续训练阶段分析（50轮）

1. 损失曲线表现

  • 训练/测试损失：完全无变化（保持0.4619和0.4715）。

  • 关键问题：

    • 加载初始训练的最佳模型后，继续训练未带来任何优化，说明模型参数已完全收敛，优化器无法找到更优方向。

2. 准确率曲线表现

  • 测试准确率：全程保持在76.93%，与初始训练结果一致。

  • 关键问题：

    • 继续训练未能提升性能，表明模型能力已达上限，或需要调整训练策略。

 

本文来自互联网用户投稿，该文观点仅代表作者本人，不代表本站立场。本站仅提供信息存储空间服务，不拥有所有权，不承担相关法律责任。如若转载，请注明出处：http://www.pswp.cn/bicheng/82517.shtml

如若内容造成侵权/违法违规/事实不符，请联系多彩编程网进行投诉反馈email:809451989@qq.com，一经查实，立即删除！

相关文章

卷积神经网络(CNN)深度讲解

卷积神经网络(CNN)深度讲解

卷积神经网络&#xff08;CNN&#xff09; 本篇博客参考自大佬的开源书籍&#xff0c;帮助大家从头开始学习卷积神经网络&#xff0c;谢谢各位的支持了&#xff0c;在此期待各位能与我共同进步​ 卷积神经网络&#xff08;CNN&#xff09;是一种特殊的深度学习网络结构&#x…
阅读更多...
深度体验：海螺 AI，开启智能创作新时代

深度体验：海螺 AI，开启智能创作新时代

人工智能 AI 工具如雨后春笋般涌现&#xff0c;而海螺 AI 以其独特的魅力与卓越的性能&#xff0c;迅速在众多产品中崭露头角&#xff0c;成为了无数创作者、办公族以及各行业人士的得力助手。近期&#xff0c;我对海螺 AI 进行了深入的使用体验&#xff0c;接下来就为大家详细…
阅读更多...
哈希表day5

哈希表day5

242 有效的字母异位词 思路就是转为ASCII码&#xff0c;然后用一个数组记录26位字母出现的次数 #include <string> class Solution{ public:bool isAnagram(string s,string t){int record[26]{0};for (int i0;i<s.size();i){record[s[i]-a];}for (int i0;i<t.si…
阅读更多...
【Python数据库全栈指南】从SQL到ORM深度实践

【Python数据库全栈指南】从SQL到ORM深度实践

目录 &#x1f31f; 前言&#x1f3d7;️ 技术背景与价值&#x1fa79; 当前技术痛点&#x1f6e0;️ 解决方案概述&#x1f465; 目标读者说明 &#x1f9e0; 一、技术原理剖析&#x1f4ca; 核心概念图解&#x1f4a1; 核心作用讲解&#x1f527; 关键技术模块说明⚖️ 技术选…
阅读更多...
Android磁盘占用优化全解析：从监控到治理的存储效率革命

Android磁盘占用优化全解析：从监控到治理的存储效率革命

引言 随着移动应用功能的复杂化&#xff0c;磁盘占用问题日益突出。据统计&#xff0c;国内头部应用的平均安装包大小已超100MB&#xff0c;运行时缓存、日志、图片等数据更可能使磁盘占用突破GB级。过度的磁盘消耗不仅影响用户设备空间&#xff0c;还可能触发系统的“应用数据…
阅读更多...
AJAX-让数据活起来(一)：入门

AJAX-让数据活起来(一)：入门

目录 一、AJAX概念和axios使用 1.1 什么是AJAX ? 1.2 怎么用AJAX ? 1.3 axios使用 二、认识URL 2.1 什么是URL? 2.2 URL的组成 组成 协议 域名 资源路径 获取-新闻列表 三、URL查询参数 URL查询参数 axios - 查询参数 四、常用请求方法和数据提交 常用请求…
阅读更多...
【C++篇】list模拟实现

【C++篇】list模拟实现

实现接口&#xff1a; list的无参构造、n个val构造、拷贝构造 operator重载 实现迭代器 push_back() push_front() erase() insert() 头尾删 #pragma once #include<iostream> #include<assert.h> using namespace std;namespace liu {//定义list节点temp…
阅读更多...
Go 语言范围循环变量重用问题与 VSCode 调试解决方法

Go 语言范围循环变量重用问题与 VSCode 调试解决方法

文章目录 问题描述问题原因1. Go 1.21 及更早版本的范围循环行为2. Go 1.22 的改进3. VSCode 调试中的问题4. 命令行 dlv debug 的正确输出 三种解决方法1. 启用 Go 模块2. 优化 VSCode 调试配置3. 修改代码以确保兼容性4. 清理缓存5. 验证环境 验证结果结论 在 Go 编程中&…
阅读更多...
快速创建 Vue 3 项目

快速创建 Vue 3 项目

安装 Node.js 和 Vue CL 安装 Node.js&#xff1a;访问 https://nodejs.org/ 下载并安装 LTS 版本。 安装完后&#xff0c;在终端检查版本&#xff1a; node -v npm -v安装 Vue CLI&#xff08;全局&#xff09;&#xff1a; npm install -g vue/cli创建 Vue 3 项目 vue cr…
阅读更多...
java学习日志——Spring Security介绍

java学习日志——Spring Security介绍

使用Spring Security要重写UserDetailsService的loadUserByUsername方法&#xff08;相当于自定了认证逻辑&#xff09;
阅读更多...
【C++进阶篇】初识哈希

【C++进阶篇】初识哈希

哈希表深度剖析&#xff1a;原理、冲突解决与C容器实战 一. 哈希1.1 哈希概念1.2 哈希思想1.3 常见的哈希函数1.3.1 直接定址法1.3.2 除留余数法1.3.3 乘法散列法&#xff08;了解&#xff09;1.3.4 平方取中法&#xff08;了解&#xff09; 1.4 哈希冲突1.4.1 冲突原因1.4.2 解…
阅读更多...
单机Kafka配置ssl并在springboot使用

单机Kafka配置ssl并在springboot使用

目录 SSL证书生成根证书生成服务端和客户端证书生成keystore.jks和truststore.jks辅助脚本单独生成truststore.jks 环境配置hosts文件kafka server.properties配置ssl 启动kafkakafka基础操作springboot集成准备工作需要配置的文件开始消费 SSL证书 证书主要包含两大类&#x…
阅读更多...
PCB设计教程【入门篇】——电路分析基础-元件数据手册

PCB设计教程【入门篇】——电路分析基础-元件数据手册

前言 本教程基于B站Expert电子实验室的PCB设计教学的整理&#xff0c;为个人学习记录&#xff0c;旨在帮助PCB设计新手入门。所有内容仅作学习交流使用&#xff0c;无任何商业目的。若涉及侵权&#xff0c;请随时联系&#xff0c;将会立即处理 目录 前言 一、数据手册的重要…
阅读更多...
Vue2实现Office文档（docx、xlsx、pdf）在线预览

Vue2实现Office文档（docx、xlsx、pdf）在线预览

&#x1f31f; 前言 欢迎来到我的技术小宇宙&#xff01;&#x1f30c; 这里不仅是我记录技术点滴的后花园&#xff0c;也是我分享学习心得和项目经验的乐园。&#x1f4da; 无论你是技术小白还是资深大牛&#xff0c;这里总有一些内容能触动你的好奇心。&#x1f50d; &#x…
阅读更多...
【辰辉创聚生物】JAK-STAT信号通路相关蛋白：细胞信号传导的核心枢纽

【辰辉创聚生物】JAK-STAT信号通路相关蛋白：细胞信号传导的核心枢纽

在细胞间复杂的信号传递网络中&#xff0c;Janus 激酶 - 信号转导和转录激活因子&#xff08;JAK-STAT&#xff09;信号通路犹如一条高速信息公路&#xff0c;承担着传递细胞外信号、调控基因表达的重要使命。JAK-STAT 信号通路相关蛋白作为这条信息公路上的 “关键节点” 和 “…
阅读更多...
OceanBase数据库从入门到精通（运维监控篇）

OceanBase数据库从入门到精通（运维监控篇）

文章目录 一、OceanBase 运维监控体系概述二、OceanBase 系统表与元数据查询2.1 元数据查询基础2.2 核心系统表详解2.3 分区元数据查询实战三、OceanBase 性能监控SQL详解3.1 关键性能指标监控3.2 SQL性能分析实战四、OceanBase 空间使用监控4.1 表空间监控体系4.2 空间使用趋势…
阅读更多...
linux 进程间通信_共享内存

linux 进程间通信_共享内存

目录 一、什么是共享内存&#xff1f; 二、共享内存的特点 优点 缺点 三、使用共享内存的基本函数 1、创建共享内存shmget() 2、挂接共享内存shmat 3、脱离挂接shmdt 4、共享内存控制shmctl 5.查看和删除共享内存 comm.hpp server.cc Client.cc Makefile 一、什么…
阅读更多...
Spring Boot 登录实现：JWT 与 Session 全面对比与实战讲解

Spring Boot 登录实现：JWT 与 Session 全面对比与实战讲解

Spring Boot 登录实现&#xff1a;JWT 与 Session 全面对比与实战讲解 2025.5.21-23:11今天在学习黑马点评时突然发现用的是与苍穹外卖jwt不一样的登录方式-Session&#xff0c;于是就想记录一下这两种方式有什么不同 在实际开发中&#xff0c;登录认证是后端最基础也是最重要…
阅读更多...
Vue中的 VueComponent

Vue中的 VueComponent

VueComponent 组件的本质 Vue 组件是一个可复用的 Vue 实例。每个组件本质上就是通过 Vue.extend() 创建的构造函数&#xff0c;或者在 Vue 3 中是由函数式 API&#xff08;Composition API&#xff09;创建的。 // Vue 2 const MyComponent Vue.extend({template: <div…
阅读更多...
使用 FFmpeg 将视频转换为高质量 GIF（保留原始尺寸和帧率）

使用 FFmpeg 将视频转换为高质量 GIF（保留原始尺寸和帧率）

在制作教程动图、产品展示、前端 UI 演示等场景中，我们经常需要将视频转换为体积合适且清晰的 GIF 动图。本文将详细介绍如何使用 FFmpeg 工具将视频转为高质量 GIF，包括： ✅ 保留原视频尺寸或自定义缩放✅ 保留原始帧率或自定义帧率✅ 使用调色板优化色彩质量✅ 降低体积同…
阅读更多...
最新文章

Copyright @ 2022~2023