pydpmtest 0.1.1

A python library focuses on constructing deep probabilistic models on GPU.

Pydpm

A python package focuses on constructing deep probabilistic models on GPU. Pydpm provides efficient distribution sampling functions and has included lots of probabilistic models.

Install

Temporarily support Linux system only and the Windows version will be launched soon.

pip install pydpm

Requirements

pycuda
scipy
numpy

Create Probabilistic Model

Model list

Model list is as following:

Whole name	Model	Paper
Latent Dirichlet Allocation	LDA	Link
Poisson Factor Analysis	PFA	Link
Poisson Gamma Belief Network	PGBN	Link
Convolutional Poisson Factor Analysis	CPFA	Link
Convolutional Poisson Gamma Belief Network	CPGBN	Link
Poisson Gamma Dynamical Systems	PGDS	Link
Deep Poisson Gamma Dynamical Systems	DPGDS	Link

More probabilistic models will be included further...

Demo

Create a PGBN model:

from pydpm.model import PGBN
test_model = PGBN([128, 64, 32], device='gpu')
test_model.initial(train_data)
test_model.train(100)

More complete demos can be found in pydpm/examples/...

Layer construction

Construct your own probabilistic model as you wish.

from pydpm.layer import data_base,prob_layer,model
data = data_base('./mnist_gray')
layer1 = prob_layer(128)
layer2 = prob_layer(64)
layer3 = prob_layer(32)
pgbn = model([data, layer1, layer2, layer3],'gpu')
pgbn.train(iter=100)

This module is under construction and will be launched soon...

Sample on GPU

Function list

The parameters of partial distribution functions are as following:

Function	Parameters List
Normal	mean,std,times
Multinomial	count,prob,times
Poisson	lambda,times
Gamma	shape,scale,times
Beta	alpha,beta,times
F	n1,n2,times
StudentT	n,times
Dirichlet	alpha,times
Crt	point,p,times
Weibull	shape,scale,times
Chisquare	n,times
Geometric	p,times
...	...

Example

import pydpm.distribution as dsg
a=dsg.gamma(1.5,1,100000)
b=dsg.gamma([1.5,2.5],[1,2],100000)

More complete demos can be found in pydpm/distribution/...

Compare

Compare the sampling speed of distribution functions with numpy: The compared code can be found in pydpm/example/Sample_Demo.py

Compare the sampling speed of distribution functions with tensorflow and torch: The compared code can be found in pydpm/example/Sample_Demo2.py

Contact

License: Apache License Version 2.0

Contact: Chaojie Wang [e-mail], Wei Zhao [e-mail], Jiawen Wu [e-mail]

Copyright (c), 2020, Chaojie Wang, Wei Zhao, Jiawen Wu, Bo Chen and Mingyuan Zhou