mt.tf.utils

Useful subroutines dealing with GPU devices.

Functions

mt.tf.utils.gpus_in_tf_format(gpus)

Converts a gpu list or a gpu count into a list of GPUs in TF format.

mt.tf.utils.as_floatx(x)

Ensures that a tensor is of dtype floatx.

mt.tf.utils.sigmoid(x)

Stable sigmoid, taken from tfp.

mt.tf.utils.asigmoid(y)

Inverse of sigmoid, taken from tfp.

Classes

  • NameScope: An iterator that generates name scope prefixes, mostly for Keras layers.

class mt.tf.utils.NameScope(name: str, parent_scope=None)

An iterator that generates name scope prefixes, mostly for Keras layers.

Parameters:
  • name (str) – the name of the scope

  • parent_scope (NameScope, optional) – the parent name scope

__call__()

Gets the full name of a base name, with prefix generated from the name scope.

Examples

>>> from mt import tf
>>> name_scope = tf.NameScope("myblock")
>>> name_scope("a")
'myblock_0/a'
>>> name_scope("b")
'myblock_0/b'
>>> next(name_scope)
>>> name_scope("c")
'myblock_1/c'
>>> child_scope = tf.NameScope("childblock", parent_scope=name_scope)
>>> child_scope("d")
'myblock_1/childblock_0/d'

Inheritance

digraph inheritanced6f0c93074 { bgcolor=transparent; rankdir=LR; size="8.0, 12.0"; "NameScope" [URL="#mt.tf.utils.NameScope",fillcolor=white,fontname="Vera Sans, DejaVu Sans, Liberation Sans, Arial, Helvetica, sans",fontsize=10,height=0.25,shape=box,style="setlinewidth(0.5),filled",target="_top",tooltip="An iterator that generates name scope prefixes, mostly for Keras layers."]; }
prefix(full: bool = False)

Returns the current prefix, with or without any parent prefix.