CODEWITHSUNDEEP
pythontensorflowneural-network
sandboxj
sandboxj

Reputation: 1254

Is tf.layers.dense a single layer?

If I just use a single layer like this:

layer = tf.layers.dense(tf_x, 1, tf.nn.relu)

Is this just a single layer with a single node?

Or is it actually a set of layers (input, hidden, output) with 1 node? My network seemed to work properly with just 1 layer, so I was curious about the setup.

Consequently, does this setup below have 2 hidden layers (are layer1 and layer2 here both hidden layers)? Or actually just 1 (just layer 1)?

layer1 = tf.layers.dense(tf_x, 10, tf.nn.relu)
layer2 = tf.layers.dense(layer1, 1, tf.nn.relu)

tf_x is my input features tensor.

Upvotes: 20

Views: 24986

Answers (2)

Christian Frei
Christian Frei

Reputation: 481

tf.layers.dense (tf.compat.v1.layers.dense) is only one layer with a amount of nodes. You can check on TensorFlow web site about tf.layers.dense (tf.compat.v1.layers.dense)

layer1 = tf.layers.dense(inputs=pool2_flat, units=1024, activation=tf.nn.relu)
layer2 = tf.layers.dense(inputs=layer1, units=1024, activation=tf.nn.relu)

Upvotes: 3

GeertH
GeertH

Reputation: 1768

tf.layers.dense adds a single layer to your network. The second argument is the number of neurons/nodes of the layer. For example:

# no hidden layers, dimension output layer = 1
output = tf.layers.dense(tf_x, 1, tf.nn.relu)

# one hidden layer, dimension hidden layer = 10,  dimension output layer = 1
hidden = tf.layers.dense(tf_x, 10, tf.nn.relu)
output = tf.layers.dense(hidden, 1, tf.nn.relu)

My network seemed to work properly with just 1 layer, so I was curious about the setup.

That is possible, for some tasks you will get decent results without hidden layers.

Upvotes: 18

Related Questions