https://github.com/omni-us/narchi

A neural network architecture definition package
https://github.com/omni-us/narchi

ai artificial-intelligence configuration data-science deep-learning machine-learning neural-networks pytorch

Last synced: 3 months ago
JSON representation

A neural network architecture definition package

• Host: GitHub
• URL: https://github.com/omni-us/narchi
• Owner: omni-us
• License: mit
• Created: 2020-05-16T08:43:04.000Z (over 4 years ago)
• Default Branch: master
• Last Pushed: 2024-10-07T07:42:04.000Z (4 months ago)
• Last Synced: 2024-10-07T08:13:59.932Z (4 months ago)
• Topics: ai, artificial-intelligence, configuration, data-science, deep-learning, machine-learning, neural-networks, pytorch
• Language: HTML
• Homepage:
• Size: 6.84 MB
• Stars: 4
• Watchers: 6
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.rst
  • License: LICENSE.rst

Awesome Lists containing this project

README

        
.. image:: https://circleci.com/gh/omni-us/narchi.svg?style=svg

    :target: https://circleci.com/gh/omni-us/narchi

.. image:: https://codecov.io/gh/omni-us/narchi/branch/master/graph/badge.svg

    :target: https://codecov.io/gh/omni-us/narchi

.. image:: https://sonarcloud.io/api/project_badges/measure?project=omni-us_narchi&metric=alert_status

    :target: https://sonarcloud.io/dashboard?id=omni-us_narchi

.. image:: https://badge.fury.io/py/narchi.svg

    :target: https://badge.fury.io/py/narchi

.. image:: https://img.shields.io/badge/contributions-welcome-brightgreen.svg

    :target: https://github.com/omni-us/narchi

narchi - A neural network architecture definition package

=========================================================

`narchi` is as python package that provides functionalities for defining neural

network architectures in an implementation independent way. It is intended

to make network architectures highly configurable while also making the task

easier.

Main features

-------------

- Network architectures are written in `jsonnet format

  `__, which provides useful features like input

  parameters and functions to define repeated blocks.

- The shapes of the tensors internal to the networks are automatically deduced

  by propagating the shapes of the inputs, thus requiring less effort and being

  less error prone.

- Propagation of shapes is done using symbolic arithmetic which makes it

  simple to understand relationships between inputs and the derived shapes.

- Architecture files can reference other architecture files, thus making this

  approach modular.

- A command line tool is included to validate jsonnet architecture files and to

  create detailed diagrams of the respective network architectures.

- Several `examples `_

  intended to illustrate different features supported.

- Includes basic implementations that allows to instantiate pytorch modules:

  * Instantiation only requires a jsonnet architecture file.

  * No need to write module classes or forward function for each new architecture.

  * One basic implementation that supports instatiating several of the examples.

  * A second example that supports packed 1d and 2d sequences which illustrates the implementation independent nature of the architecture files.

Teaser example

--------------

Here you can see an example that illustrates what `narchi` provides. The example

is for resnet18 as implemented in torchvision, though bare in mind that the

potential of `narchi` is the ease of configurability of network architectures, not

the reimplementation of existing architectures.

Instantiating a pytorch module from the architecture file can be easily done as

follows.

.. code-block:: PYTHON

   from narchi.instantiators.pytorch import StandardModule

   module = StandardModule('resnet.jsonnet',

                           state_dict='resnet18-5c106cde.pth',

                           cfg={'ext_vars': {"num_blocks": [2, 2, 2, 2]}})

Creating a diagram of the architecture requires a single command like the following.

.. code-block:: BASH

   narchi_cli.py render \

     --ext_vars '{"num_blocks": [2, 2, 2, 2]}' \

     --nested_depth 4 \

     resnet.jsonnet \

     resnet18.pdf

Below you can see a small part of the rendered diagram of the start of the

first downsample layer of resnet18.

.. image:: imgs/resnet_crop.png

   :width: 500

The part of the json that generated the previous crop of the architecture

diagram can be seen below. Note that information of the shapes is not included,

since these are derived automatically.

.. code-block:: JSON

   {

     "_class": "Sequential",

     "_id": "layer2",

     "blocks": [

       {

         "_class": "Group",

         "_name": "ResBlock",

         "blocks": [

           {

             "_class": "Identity",

             "_id": "ident"

           },

           {

             "_class": "Conv2d",

             "_id": "conv1",

             "bias": false,

             "dilation": 1,

             "kernel_size": 3,

             "output_size": 128,

             "padding": 1,

             "stride": 2

           },

           {

             "_class": "BatchNorm2d",

             "_id": "bn1"

           },

           {

             "_class": "Sequential",

             "_id": "downsample",

             "blocks": [

               {

                 "_class": "Conv2d",

                 "bias": false,

                 "kernel_size": 1,

                 "output_size": 128,

                 "padding": 0,

                 "stride": 2

               },

               {

                 "_class": "BatchNorm2d"

               }

             ]

           },

           {"...": "..."}

         ],

         "graph": [

           "ident -> conv1 -> bn1 -> relu1 -> conv2 -> bn2 -> add -> relu2",

           "ident -> downsample -> add"

         ],

         "input": "ident",

         "output": "relu2"

       }

     ]

   }