keraflow.layers.base.Kensor Class Reference

Wrapper for a backend tensor. More...

Inheritance diagram for keraflow.layers.base.Kensor:

Public Member Functions
def	__init__

Detailed Description

Wrapper for a backend tensor.

A Kensor wraps a backend tensor and record meta data such as the tensor's shape and path from previous kensors to this kensor. A Kensor is genrated when:

1. An Input layer is instantiated
2. A kensor is feed to a layer (Layer.__call__).

Constructor & Destructor Documentation

def keraflow.layers.base.Kensor.__init__	(		self,
			layer,
			name,
			shape,
			tensor,
			ancestor_info = {},
			path = []
	)

Parameters

layer	Layer. The layer that outputs the kensor.
name	str. The name of the kensor.
shape	tuple. The shape of the tensor.
tensor	backend tensor. The backend tensor of the kensor.
ancestor_info	dict. Info of ancestor kensors of the kensor. Used by Model to determine trainable parameters, regularizers, etc... See Layer.__call__, Model.compile
path	list. The path from ancestors to the kensor. Used by Model to reconstruct the tensor graph. See Layer.__call__, Model.unpack_init_param