各类深度学习模型示例

system

深度学习方法利用神经网络模型进行高级模式识别和自动特征提取，近年来在数据挖掘领域成果显著。常用模型有除了基础的DNN，还有RNN、LSTM、GRU、CNN、Attention和Mix混合模型等。

system

DNN（深度神经网络）

import tensorflow as tf
from tensorflow.keras import layers, models

# 数据加载与预处理
(x_train, y_train), (x_test, y_test) = tf.keras.datasets.mnist.load_data()
x_train = x_train.reshape(-1, 28*28).astype('float32') / 255.0  # 展平为784维向量
x_test = x_test.reshape(-1, 28*28).astype('float32') / 255.0

# 构建DNN模型
model = models.Sequential([
    layers.Dense(512, activation='relu', input_shape=(784,)),  # 全连接层
    layers.Dropout(0.2),
    layers.Dense(256, activation='relu'),
    layers.Dense(10, activation='softmax')  # 输出层
])

# 编译与训练
model.compile(optimizer='adam',
              loss='sparse_categorical_crossentropy',
              metrics=['accuracy'])
model.fit(x_train, y_train, epochs=5, batch_size=64, validation_split=0.2)

# 评估
test_loss, test_acc = model.evaluate(x_test, y_test)
print(f"DNN测试准确率: {test_acc:.4f}")

system

CNN（卷积神经网络）

import tensorflow as tf
from tensorflow.keras import layers, models

# 数据加载与预处理（保留空间结构）
(x_train, y_train), (x_test, y_test) = tf.keras.datasets.mnist.load_data()
x_train = x_train.reshape(-1, 28, 28, 1).astype('float32') / 255.0  # 输入形状 (28,28,1)
x_test = x_test.reshape(-1, 28, 28, 1).astype('float32') / 255.0
# 构建CNN模型
model = models.Sequential([
    layers.Conv2D(32, (3,3), activation='relu', input_shape=(28,28,1)),  # 卷积层
    layers.MaxPooling2D((2,2)),  # 池化层
    layers.Conv2D(64, (3,3), activation='relu'),
    layers.MaxPooling2D((2,2)),
    layers.Flatten(),  # 展平为向量
    layers.Dense(128, activation='relu'),
    layers.Dense(10, activation='softmax')
])
# 编译与训练
model.compile(optimizer='adam',
              loss='sparse_categorical_crossentropy',
              metrics=['accuracy'])
model.fit(x_train, y_train, epochs=5, batch_size=64, validation_split=0.2)
# 评估
test_loss, test_acc = model.evaluate(x_test, y_test)
print(f"CNN测试准确率: {test_acc:.4f}")

system

LSTM（长短期记忆网络）

import tensorflow as tf
from tensorflow.keras import layers, models

#数据加载与预处理（文本序列）
max_features = 10000  # 词汇表大小
max_len = 200         # 序列截断/填充长度
(x_train, y_train), (x_test, y_test) = tf.keras.datasets.imdb.load_data(num_words=max_features)
x_train = tf.keras.preprocessing.sequence.pad_sequences(x_train, maxlen=max_len)  # 填充为固定长度
x_test = tf.keras.preprocessing.sequence.pad_sequences(x_test, maxlen=max_len)
#构建LSTM模型
model = models.Sequential([
    layers.Embedding(max_features, 128),      # 嵌入层（将整数索引转为密集向量）
    layers.LSTM(64, dropout=0.2, recurrent_dropout=0.2),  # LSTM层
    layers.Dense(1, activation='sigmoid')     # 二分类输出
])
#编译与训练
model.compile(optimizer='adam',
              loss='binary_crossentropy',
              metrics=['accuracy'])
model.fit(x_train, y_train, epochs=3, batch_size=128, validation_split=0.2)
#评估
test_loss, test_acc = model.evaluate(x_test, y_test)
print(f"LSTM测试准确率: {test_acc:.4f}")

system

RNN（循环神经网络）

import numpy as np
import matplotlib.pyplot as plt
import tensorflow as tf
from tensorflow.keras import layers, models
# 生成合成数据（正弦波）
def generate_sine_wave(time_steps, seq_length=1000):
    x = np.linspace(0, 2 * np.pi, seq_length)
    y = np.sin(x)  # 正弦波
    # 将序列分割为多个样本，每个样本包含 time_steps 个时间步
    samples = []
    targets = []
    for i in range(len(y) - time_steps):
        samples.append(y[i:i+time_steps])
        targets.append(y[i+time_steps])  # 预测下一个时间步的值
    return np.array(samples), np.array(targets)
# 超参数
time_steps = 20  # 输入序列长度
n_features = 1   # 每个时间步的特征数（单变量）

# 生成数据
X, y = generate_sine_wave(time_steps)
X = X.reshape(-1, time_steps, n_features)  # 形状：(样本数, 时间步数, 特征数)
y = y.reshape(-1, 1)
# 划分训练集和测试集
split = int(0.8 * len(X))
X_train, y_train = X[:split], y[:split]
X_test, y_test = X[split:], y[split:]
# 构建RNN模型
model = models.Sequential([
    layers.SimpleRNN(units=64, activation='tanh', input_shape=(time_steps, n_features)),  # RNN层
    layers.Dense(1)  # 输出层（预测一个值）
])
# 编译模型
model.compile(optimizer='adam', loss='mse')  # 均方误差损失
# 训练
history = model.fit(X_train, y_train, epochs=50, batch_size=32, validation_split=0.1)
# 预测
predictions = model.predict(X_test)

# 可视化结果
plt.figure(figsize=(12, 6))
plt.plot(y_test, label='真实值')
plt.plot(predictions, label='预测值')
plt.legend()
plt.title("RNN时间序列预测结果")
plt.show()

system

模型	数据特点	核心层	典型准确率/误差
DNN	扁平化结构化数据	Dense	Fashion-MNIST ₈₈%
CNN	空间数据（图像）	Conv2D + MaxPooling	CIFAR-10 ₇₀%
LSTM	长序列文本	LSTM + Embedding	IMDB ₈₅%
RNN	短/中期时间序列	SimpleRNN	MSE 可视化趋势一致