Linear¶

Description¶

Info

Parent class: Module

Derived classes: -

This module performs a linear transformation to the input data:

$y = xW^T + b$

where

$x$ - input feature vector;
$y$ - transformed feature vector;
$W$ - layer weights matrix;
$b$ - layer bias vector.

def __init__(self,  insize,  outsize,  wscale=1.0,  useBias=True,  initscheme=None,  name=None,  empty=False,  transpose=False):

Parameters

Parameter	Allowed types	Description	Default
insize	int	Size of the input vector	-
outsize	int	Size of the input vector	-
wscale	float	Dispersion of random layer weights	1.0
useBias	bool	Whether to use the bias vector	True
initscheme	Union[tuple, str]	Specifies the layer weights initialization scheme. (see createTensorWithScheme).	None -> ("xavier_uniform", "in")
name	str	Layer name	None
empty	bool	Whether to initialize the matrix of weights and biases	False
transpose	bool	Use of a transposed matrix	False

Explanations

-

Necessary imports.

import numpy as np
from PuzzleLib.Backend import gpuarray
from PuzzleLib.Modules import Linear

Info

gpuarray is required to properly place the tensor in the GPU

batchsize, insize, outsize = 10, 32, 16
data = gpuarray.to_gpu(np.random.randn(batchsize, insize).astype(np.float32))
print(data.shape)

(10, 32)

linear = Linear(insize, outsize)
print(linear.W.shape)

(32, 16)

linear(data)
print(linear.data.shape)

(10, 16)