Tensorboard Histogram of Activations

Question

How can I plot histograms of the activations of tf.layers.dense in tensorboard? Consider the following situation:

with tf.variable_scope('hidden'):
hidden_layer = tf.layers.dense(input, 100, activation=tf.nn.relu)

I can plot the weights and biases from hidden with the following code:

for v in tf.get_collection(tf.GraphKeys.TRAINABLE_VARIABLES, scope='hidden'):
    tf.summary.histogram("{}/weight_histogram".format(v.name), v)

However, I can't seem to get a collection of the activations as well. How could I modify this to also plot histograms of the activations?

Answer 1

在模型范围内，您需要调用以下内容： tf.contrib.layers.summarize_activation(hidden_layer)