Package crino :: Module module :: Class Linear

[frames] | no frames]

Class Linear

A Linear module computes its outputs as a linear transformation of its inputs. It has two params : W ∈ ℳ_{n_out × n_in}(ℝ) and b ∈ ℝ^n_out.

The outputs expression can be written as follows : ŷ = W × x + b

Instance Methods

__init__(self, nOutputs, nInputs=None, W_init=None, b_init=None)
Constructs a new Linear module.

source code

prepareParams(self, W=None, b=None)
Initializes the W and b params of the Linear module.

source code

prepareOutput(self)
Computes the linear relation ŷ = W × x + b

source code

Inherited from Standalone: prepareGeometry

Inherited from Module: criterionFunction, forward, forwardFunction, holdFunction, linkInputs, linkModule, prepare, prepareBackup, restoreFunction, save, trainFunction

Instance Variables

ndarray W_init
The initialization matrix for W.

ndarray b_init
The initialization vector for b.

TensorVariable W
The symbolic linear transformation matrix.

TensorVariable b
The symbolic offset vector.

Inherited from Module: backupParams, inputs, nInputs, nOutputs, outputs, params, prepared

Method Details

init(self, nOutputs, nInputs=None, W_init=None, b_init=None)
(Constructor)

Constructs a new Linear module.

Parameters:

nOutputs (int) - The outputs size.
nInputs (int) - The inputs size.
W_init (ndarray) - The initialization matrix for W.
b_init (ndarray) - The initialization vector for b.

Overrides: Module.__init__

prepareParams(self, W=None, b=None)

Initializes the W and b params of the Linear module.

Parameters:

W (TensorVariable) - If provided, the Linear module will use W as a shared parameter from another module.
b (TensorVariable) - If provided, the Linear module will use b as a shared parameter from another module.

Overrides: Module.prepareParams

Attention: W\_init and b\_init values will be ignored if existing W and b are passed, since they have already been initialized in another module.

prepareOutput(self)

Computes the linear relation ŷ = W × x + b

Overrides: Module.prepareOutput