python - Tensorflow tf.nn.softmax() 函数比手写的 softmax 性能好很多

Question

我正在使用 tensorflow 编写一个简单的逻辑回归。我发现当使用 tf.nn.softmax 时，算法收敛得更快，最终精度更高。如果切换到我自己的 softmax 实现，网络收敛速度较慢，最终精度也没有那么好。

代码如下:

SEED = 1025
W = tf.Variable(tf.truncated_normal([image_size * image_size, num_labels], seed=SEED))
b = tf.Variable(tf.zeros([num_labels]))
logits = tf.matmul(train_dataset, W) + b

# My softmax:
y_ = tf.exp(logits) / tf.reduce_sum(tf.exp(logits), axis=0)
# Tensorflow softmax: 
y_ = tf.nn.softmax(logits)

y_clipped = tf.clip_by_value(y_, 1e-10, 0.9999999)
loss = -tf.reduce_mean(tf.reduce_sum(train_labels * tf.log(y_clipped), axis=1))

使用我的 softmax:

Loss at step 0: 22.213934
Training accuracy: 12.7%
Validation accuracy: 13.2%
Loss at step 100: 12.777291
Training accuracy: 45.3%
Validation accuracy: 45.5%
Loss at step 200: 11.361242
Training accuracy: 48.2%
Validation accuracy: 47.4%
Loss at step 300: 10.658278
Training accuracy: 51.4%
Validation accuracy: 49.7%
Loss at step 400: 9.297832
Training accuracy: 59.2%
Validation accuracy: 56.8%
Loss at step 500: 8.902699
Training accuracy: 62.0%
Validation accuracy: 59.2%
Loss at step 600: 8.681184
Training accuracy: 64.2%
Validation accuracy: 61.0%
Loss at step 700: 8.529438
Training accuracy: 65.8%
Validation accuracy: 62.3%
Loss at step 800: 8.416442
Training accuracy: 66.8%
Validation accuracy: 63.3%
Test accuracy: 70.4%

使用tensorflow的softmax:

Loss at step 0: 13.555875
Training accuracy: 12.7%
Validation accuracy: 14.5%
Loss at step 100: 2.194562
Training accuracy: 72.5%
Validation accuracy: 72.0%
Loss at step 200: 1.808641
Training accuracy: 75.5%
Validation accuracy: 74.5%
Loss at step 300: 1.593390
Training accuracy: 76.8%
Validation accuracy: 75.0%
Loss at step 400: 1.442661
Training accuracy: 77.7%
Validation accuracy: 75.2%
Loss at step 500: 1.327751
Training accuracy: 78.2%
Validation accuracy: 75.4%
Loss at step 600: 1.236314
Training accuracy: 78.5%
Validation accuracy: 75.6%
Loss at step 700: 1.161479
Training accuracy: 78.9%
Validation accuracy: 75.6%
Loss at step 800: 1.098717
Training accuracy: 79.4%
Validation accuracy: 75.8%
Test accuracy: 83.3%

从 documentation 来看，理论上 tensorflow 的 softmax 应该和我实现的完全一样，不是吗？

This function performs the equivalent of

softmax = tf.exp(logits) / tf.reduce_sum(tf.exp(logits), axis)

编辑: 我在从正态分布初始化时添加了一个种子，现在我可以重现准确度结果。在“My softmax”行中设置轴值时，只有 axis=0 不会导致错误。设置 axis=1 或 axis=-1 都会导致此错误:

tensorflow.python.framework.errors_impl.InvalidArgumentError: Dimensions must be equal, but are 10 and 10000 for 'truediv' (op: 'RealDiv') with input shapes: [10000,10], [10000].

Answer 1

• 假设您的 softmax 实现是正确的
• 首先，将 tensorflow softmax 与手写 softmax 进行比较是不公平的，因为您的程序中包含随机性
• 我的意思是 W = tf.Variable(tf.truncated_normal([image_size * image_size, num_labels])) 这行在你的程序中引入了随机性，因为权重最初是随机设置的，所以每次运行你的程序你都会得到不同的结果
• 如果你有某种种子(某种起点)，你只能比较两个 softmax
• 现在，如果您多次执行上述实验并且每次 tensorflow softmax 都击败手写 softmax，那么您的问题是有效的
• tf.truncated_normal 函数确实接受了一个种子参数……您可以使用该参数并查看结果是什么
• 无论如何，如果您手写的 softmax 是正确的，那么使用种子的 tensorflow softmax 和您的 softmax 应该输出相同的结果
• 甚至我认为在你的情况下你的轴应该是 1，这是最后一个轴，因为 softmax 应该沿着有类的轴