Tensorflow入门到实战四(识别手写数字集mnist)

最新推荐文章于 2021-02-26 18:48:33 发布

黑洲非人lyf

最新推荐文章于 2021-02-26 18:48:33 发布

阅读量242

点赞数

分类专栏： Tensorflow Tensorflow从入门到实战文章标签： tensorflow

本文链接：https://blog.csdn.net/lyf52010/article/details/88026964

版权

Tensorflow 同时被 2 个专栏收录

7 篇文章 0 订阅

订阅专栏

Tensorflow从入门到实战

7 篇文章 2 订阅

订阅专栏

本文介绍了一种使用TensorFlow实现的手写数字识别模型。该模型利用MNIST数据集进行训练，通过添加dropout层来防止过拟合，采用softmax作为激活函数进行多分类预测。在训练过程中，使用交叉熵损失函数和梯度下降优化器进行参数更新，最终模型在测试集上的准确率达到84.26%。

摘要由CSDN通过智能技术生成

手写数字集mnist

任然是集合作为特征数*样本数，特征数代表了某层神经元数量

wx_plus_b = tf.nn.dropout(wx_plus_b,keep_prob)

dropout可以解决过拟合

def add_layer(inputs,in_size,out_size,activation_function=None):
    Weights = tf.Variable(tf.random_normal([out_size,in_size]))
    bias = tf.Variable(tf.zeros([out_size,1]))
    wx_plus_b = tf.matmul(Weights,inputs)+bias
    
    if activation_function is None:
        return wx_plus_b
    else :
        return activation_function(wx_plus_b)
    


xs = tf.placeholder(tf.float32,[784,None])
ys = tf.placeholder(tf.float32,[10,None])


prediction = add_layer(xs,784,10,activation_function=tf.nn.softmax)

cross_entropy = tf.reduce_mean(-tf.reduce_sum(ys*tf.log(prediction),reduction_indices=[1]))

train_step = tf.train.GradientDescentOptimizer(0.1).minimize(cross_entropy)

init = tf.global_variables_initializer()
session = tf.Session()
session.run(init)

def compute_accuracy(v_xs, v_ys):
    global prediction
    y_pre = session.run(prediction, feed_dict={xs: v_xs})
    # tf.argmax (y_pre,1 ) 返回每一行 下标最大的元素，1表示按行
    correct_prediction = tf.equal(tf.argmax(y_pre,0), tf.argmax(v_ys,0))
    accuracy = tf.reduce_mean(tf.cast(correct_prediction, tf.float32))
    result = session.run(accuracy, feed_dict={xs: v_xs, ys: v_ys})
    return result
for i in range(1000):
    batch_xs,batch_ys = mnist.train.next_batch(100)
    session.run(train_step,feed_dict={xs:np.transpose(batch_xs),ys:np.transpose(batch_ys)})
    if i%50==0:
        print("loss",session.run(cross_entropy,feed_dict={xs:np.transpose(batch_xs),ys:np.transpose(batch_ys)}))
        print(compute_accuracy(np.transpose(mnist.test.images), np.transpose(mnist.test.labels)))

输出

loss 148.512
0.1313
loss 48.9171
0.6285
loss 45.432
0.6736
loss 36.675
0.6946
loss 35.2471
0.7188
loss 33.0088
0.7398
loss 33.254
0.7498
loss 33.4405
0.7662
loss 30.5252
0.7718
loss 35.2746
0.7744
loss 33.5173
0.7945
loss 33.7244
0.7903
loss 30.1716
0.8109
loss 30.2718
0.8003
loss 29.682
0.8288
loss 31.7652
0.8243
loss 29.8357
0.8333
loss 30.9754
0.8324
loss 30.0272
0.835
loss 29.3583
0.8426