【Python深度学习系列】基于tensorflow原生代码搭建BP神经网络实现分类任务（附源码和实现效果）

数据杂坛

已于 2024-01-25 11:30:23 修改

阅读量1.4k

点赞数 1

分类专栏：深度学习文章标签： python 深度学习 tensorflow

于 2023-10-26 16:51:51 首次发布

本文链接：https://blog.csdn.net/sinat_41858359/article/details/134060088

版权

深度学习专栏收录该内容

14 篇文章 1 订阅

订阅专栏

一、实现功能

前面两篇文章分别介绍了两种搭建神经网络模型的方法，一种是基于tensorflow的keras框架，另一种是继承父类自定义class类，本篇文章将编写原生代码搭建BP神经网络。

二、实现代码

import tensorflow as tf
from sklearn.datasets import load_iris
from sklearn.model_selection import train_test_split
from sklearn.preprocessing import StandardScaler

# 加载鸢尾花数据集
iris = load_iris()
X = iris.data
y = iris.target

# 数据预处理
scaler = StandardScaler()
X = scaler.fit_transform(X)

# 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)

# 设置超参数
learning_rate = 0.001
num_epochs = 100
batch_size = 32

# 定义输入和输出的维度
input_dim = X.shape[1]
output_dim = len(set(y))

# 定义权重和偏置项
W1 = tf.Variable(tf.random.normal(shape=(input_dim, 64), dtype=tf.float64))
b1 = tf.Variable(tf.zeros(shape=(64,), dtype=tf.float64))
W2 = tf.Variable(tf.random.normal(shape=(64, 64), dtype=tf.float64))
b2 = tf.Variable(tf.zeros(shape=(64,), dtype=tf.float64))
W3 = tf.Variable(tf.random.normal(shape=(64, output_dim), dtype=tf.float64))
b3 = tf.Variable(tf.zeros(shape=(output_dim,), dtype=tf.float64))


# 定义前向传播函数
def forward_pass(X):
    X = tf.cast(X, tf.float64)
    h1 = tf.nn.relu(tf.matmul(X, W1) + b1)
    h2 = tf.nn.relu(tf.matmul(h1, W2) + b2)
    logits = tf.matmul(h2, W3) + b3
    return logits

# 定义损失函数
def loss_fn(logits, labels):
    return tf.reduce_mean(tf.nn.sparse_softmax_cross_entropy_with_logits(labels=labels, logits=logits))

# 定义优化器
optimizer = tf.optimizers.Adam(learning_rate)

# 定义准确率指标
accuracy_metric = tf.metrics.SparseCategoricalAccuracy()

# 定义训练步骤
def train_step(inputs, labels):
    with tf.GradientTape() as tape:
        logits = forward_pass(inputs)
        loss_value = loss_fn(logits, labels)
    gradients = tape.gradient(loss_value, [W1, b1, W2, b2, W3, b3])
    optimizer.apply_gradients(zip(gradients, [W1, b1, W2, b2, W3, b3]))
    accuracy_metric(labels, logits)
    return loss_value

# 进行训练
for epoch in range(num_epochs):
    epoch_loss = 0.0
    accuracy_metric.reset_states()

    for batch_start in range(0, len(X_train), batch_size):
        batch_end = batch_start + batch_size
        batch_X = X_train[batch_start:batch_end]
        batch_y = y_train[batch_start:batch_end]

        loss = train_step(batch_X, batch_y)
        epoch_loss += loss

    train_loss = epoch_loss / (len(X_train) // batch_size)
    train_accuracy = accuracy_metric.result()

    print(f"Epoch {epoch+1}/{num_epochs}, Loss: {train_loss:.4f}, Accuracy: {train_accuracy:.4f}")

# 进行评估
logits = forward_pass(X_test)
test_loss = loss_fn(logits, y_test)
test_accuracy = accuracy_metric(y_test, logits)

print(f"Test Loss: {test_loss:.4f}, Test Accuracy: {test_accuracy:.4f}")

三、实现效果

本人读研期间发表5篇SCI数据挖掘相关论文，现在某研究院从事数据挖掘相关科研工作，对数据挖掘有一定认知和理解，会结合自身科研实践经历不定期分享关于python、机器学习、深度学习基础知识与案例。致力于只做原创，以最简单的方式理解和学习，关注我一起交流成长。关注底部公众号：数据杂坛，即可在后台联系我获取相关数据集和源码，送有关数据分析、数据挖掘、机器学习、深度学习相关的电子书籍。