Skip to main content

Probabilistic programming using pytorch.

Project description

Borch

pipeline status coverage report lifecycle Code style: black docs arXiv

Getting Started | Documentation | Contributing

Borch is a universal probabilistic programming language (PPL) framework developed by Desupervised, that uses and integrates with PyTorch. Borch was designed with special attention to support Bayesian neural networks in a native fashion. Further, it's designed to

  • Flexible and scalable framework
  • Support neural networks out of the box.
  • Have bells and whistles a universal PPL needs.

It can be installed with

pip install borch

Usage

See our full tutorials here.

As a quick example let's look into how the neural network interface looks. The module borch.nn provides implementations of neural network modules that are used for deep probabilistic programming and provides an interface almost identical to the torch.nn modules. In many cases it is possible to just switch

import torch.nn as nn

to

import borch.nn as nn

and a network defined in torch is now probabilistic, without any other changes in the model specification, one also need to change the loss function to infer.vi.vi_loss.

For example, a convolutional neural network can be written as

import torch
import torch.nn.functional as F
from borch import nn

class Net(nn.Module):
    def __init__(self):
        super(Net, self).__init__()
        self.conv1 = nn.Conv2d(1, 6, 5)
        self.conv2 = nn.Conv2d(6, 16, 5)
        self.fc1 = nn.Linear(16 * 5 * 5, 120)
        self.fc2 = nn.Linear(120, 84)
        self.fc3 = nn.Linear(84, 10)

    def forward(self, x):
        x = F.max_pool2d(F.relu(self.conv1(x)), (2, 2))
        x = F.max_pool2d(F.relu(self.conv2(x)), 2)
        x = x.view(-1, self.num_flat_features(x))
        x = F.relu(self.fc1(x))
        x = F.relu(self.fc2(x))
        x = self.fc3(x)
        return x

    def num_flat_features(self, x):
        size = x.size()[1:]
        num_features = 1
        for s in size:
            num_features *= s
        return num_features

Installation

Borch can be installed using

pip install borch

Docker

The Borch Docker images are available as both CPU and GPU versions at gitlab.com/desupervised/borch/container_registry. The latest CPU images can be used as

docker run registry.gitlab.com/desupervised/borch/cpu:master

Contributing

Please read the contribution guidelines in CONTRIBUTING.md.

Citation

If you use this software for your research or business please cite us and help the package grow!

@misc{belcher2022borch,
	title        = {Borch: A Deep Universal Probabilistic Programming Language},
	author       = {Belcher, Lewis and Gudmundsson, Johan and Green, Michael},
	year         = 2022,
	publisher    = {arXiv},
	doi          = {10.48550/ARXIV.2209.06168},
	url          = {https://arxiv.org/abs/2209.06168},
	copyright    = {Creative Commons Attribution 4.0 International},
	keywords     = {Artificial Intelligence (cs.AI), Machine Learning (cs.LG), Programming Languages (cs.PL), FOS: Computer and information sciences, FOS: Computer and information sciences}
}

Project details


Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

borch-0.1.6.tar.gz (67.7 kB view details)

Uploaded Source

Built Distribution

borch-0.1.6-py3-none-any.whl (81.3 kB view details)

Uploaded Python 3

File details

Details for the file borch-0.1.6.tar.gz.

File metadata

  • Download URL: borch-0.1.6.tar.gz
  • Upload date:
  • Size: 67.7 kB
  • Tags: Source
  • Uploaded using Trusted Publishing? No
  • Uploaded via: twine/6.0.1 CPython/3.12.8

File hashes

Hashes for borch-0.1.6.tar.gz
Algorithm Hash digest
SHA256 5cc4873a82a6b9af8e2777c673850a21f16b9af380f2f44695566c0c69477ef8
MD5 87bf1de93df96f63e1e065ceecbdadd7
BLAKE2b-256 df79c65599f33fdae10a06555668dbbeeb862627463f540bcae3bcf75fa39992

See more details on using hashes here.

File details

Details for the file borch-0.1.6-py3-none-any.whl.

File metadata

  • Download URL: borch-0.1.6-py3-none-any.whl
  • Upload date:
  • Size: 81.3 kB
  • Tags: Python 3
  • Uploaded using Trusted Publishing? No
  • Uploaded via: twine/6.0.1 CPython/3.12.8

File hashes

Hashes for borch-0.1.6-py3-none-any.whl
Algorithm Hash digest
SHA256 3adbd05a57684e9f030f3852e6ba22983ca8b3b4db83134374b8ff13a8c13747
MD5 c7a86e1e0909c2022c9ff912146cf027
BLAKE2b-256 c7ef88c8ba4fe25a352891d17cf89c1570523df7affc1b4b78b9c366751156c3

See more details on using hashes here.

Supported by

AWS Cloud computing and Security Sponsor Datadog Monitoring Fastly CDN Google Download Analytics Pingdom Monitoring Sentry Error logging StatusPage Status page