API Reference

vulkpy Package

vulkpy: GPGPU array on Vulkan

vulkpy provides GPGPU computations.

See also

vulkpy.random

Random Module

vulkpy.nn

Neural Network Module

vulkpy.util

Utility Module

Examples

>>> import vulkpy as vk

>>> gpu = vk.GPU()
>>> a = vk.Array(gpu, data=[1, 2, 3])
>>> b = vk.Array(gpu, data=[3, 3, 3])

>>> c = a + b
>>> print(c)
[4., 5., 6.]

Functions

zeros(gpu, shape)

Zero initialized Array

Classes

Array	GPU Array for float (32bit)
GPU	GPU instance
Shape	GPU Array of uint (32bit) for shape
U32Array	GPU Array of uint (32bit) for shape or indices

Class Inheritance Diagram

vulkpy.random Module

Random Module (vulkpy.random)

GPU-based Pseudo Random Number Generator (PRNG)

Examples

>>> import vulkpy as vk
>>> gpu = vk.GPU()
>>> r = vk.random.Xoshiro128pp(gpu, seed=0)

[0, 1) uniform random numbers can be generated by random(shape=None, buffer=None).

>>> print(r.random(shape=(3,)))
[0.42977667 0.8235899  0.90622926]

Gaussian random numbers can be generated by normal(shape=None, buffer=None, mean=0.0, stddev=1.0).

>>> print(r.normal(shape=(3,)))
[-2.3403292  0.7247794  0.7118352]

Classes

Xoshiro128pp

xoshiro128++: Pseudo Random Number Generator

Class Inheritance Diagram

vulkpy.nn Package

Neural Network Module (vulkpy.nn)

Examples

>>> import vulkpy as vk
>>> from vulkpy import nn
>>> gpu = vk.GPU()
>>> x = vk.Array(gpu, data=[ ... ]) # Features
>>> y = vk.Array(gpu, data=[ ... ]) # Labels

Create Optimizer and Model

>>> opt = nn.Adam(gpu, lr=1e-4)
>>> net = nn.Sequence(
...   [
...     nn.Dense(gpu, 3, 32, w_opt=opt, b_opt=opt),
...     nn.ReLU(),
...     nn.Dense(gpu, 32, 4, w_opt=opt, b_opt=opt),
...     nn.Softmax(),
...   ],
...   nn.CrossEntropy()
... )

Training Model

>>> pred_y, loss = net.train(x, y)

Predict with Model

>>> pred_y = net.predict(x)

Classes

AdaGrad	AdaGrad Optimizer
AdaGradState	Optimizer State for AdaGrad
Adam	Adam Optimizer
AdamState	Optimizer State for Adam
Constant	Constant Initializer
CrossEntropyLoss	Cross Entropy Loss
Dense	Fully connected Dense Layer
Elastic	Elastic (L1 + L2) Regularization
HeNormal	He Normal Initializer
HuberLoss	Huber Loss
Lasso	Lasso (L1) Regularization
Loss	Abstract base class for Loss
MSELoss	Mean Squared Loss
Module	Abstract base class for Module
Optimizer	Abstract base class for Optimizer
OptimizerState	Abstract base class for Optimizer State
ReLU	Rectified Linear Unit (ReLU)
Regularizer	Abstract base class for Regularizer
Ridge	Ridge (L2) Regularization
SGD	SGD Optimizer
SGDState	Optimizer State for SGD
Sequence	Sequential Model
Sigmoid	Sigmoid
Softmax	SoftMax
SoftmaxCrossEntropyLoss	Softmax Cross Entropy Loss

Class Inheritance Diagram

vulkpy.util Module

Utility Module (vulkpy.util)

Examples

>>> from vulkpy.util import enable_debug
>>> enable_debug(api_dump=False)

Functions

enable_debug(*[, validation, api_dump])	Enable debug message
getShader(name)	Get Shader Path