1# TensorFlow contrib layers.
2
3## initializers.py
4
5Functions that produce variable initializer functions with signature:
6
7`foo(shape, dtype) : Tensor`
8
9These are typically consumed by functions in [layers.py](#layers.py).
10
11## layers.py {#.py}
12
13Functions that produce layer operations and associated weight & bias variables. Signatures will vary for different functions, but they will often take many of
14these arguments.
15
16`foo(x,
17     num_outputs,
18     …,
19     weight_init=<DEFAULT>,
20     bias_init=<DEFAULT>,
21     weight_regularizer=None,
22     bias_regularizer=None,
23     name=None) : Tensor`
24
25`x` is the input tensor.
26
27Weights and biases are added to `tf.GraphKeys.GLOBAL_VARIABLES` and
28`tf.GraphKeys.TRAINABLE_VARIABLES`.
29
30## optimizers.py
31
32Functions that add optimization ops given `loss` and `global_step` tensors.
33
34## regularizers.py
35
36Functions that produce weight regularization functions with signature
37
38`foo(weight_vars, name=None) : Operation`
39
40These are typically consumed by functions in [layers.py](#layers.py).
41
42## summaries.py
43
44Functions that add summary ops to the standard `tf.GraphKeys.SUMMARIES`
45collection. They also avoid name conflicts in the summary key.
46