nnlib 0.1

Welcome to nnlib’s documentation!

nnlib is a Pytorch-based, type-annotated utilities for neural networks in Python 3.6+. It is designed with the following criteria in mind:

• Type-Annotated: Though useful, dynamic typing can be error prone. nnlib uses type annotations anywhere possible so that type mismatch errors can be spotted before running.
• PyCharm-Friendly: nnlib was developed using PyCharm, so compatibility with the IDE is a high priority.
• Generalize Boilerplates: Everyone hates writing the same thing over and over again, and that’s basically why we use libraries.

Installation

Since nnlib is still in its early stages, PyPI packages are not provided. It is recommended to include nnlib as a submodule in your git repository:

cd /path/to/your/repo
git submodule add https://github.com/huzecong/nnlib.git

and just use away!

For code autocompletion in PyCharm to work, you would also need to mark the nnlib submodule root as sources root, which can be done by right-clicking the folder in the Project panel, and selecting “Mark Directory as… > Sources Root”.

To update the submodule to the newest version:

git submodule update --remote --recursive nnlib

Contents

• Argument Parsing - nnlib.arguments
  • The Arguments Class
  • Type Resolution
  • API
  • Validators
• Data Preprocessing Utilities - nnlib.data
  • Datasets
  • Data Loader Interface
  • Data Iterators
  • Preprocessing
• Neural Network Modules - nnlib.modules
  • Embedding
  • Linear
  • Softmax
• PyTorch Utilities - nnlib.torchutils
  • CUDA
  • Parameter
• Training Procedure - nnlib.train
  • The Trainer Class
  • Hooks
• Utiliies - nnlib.utils
  • Batching
  • File System
  • Functional
  • IO
  • Iterable
  • Logging
  • Math
  • Miscellaneous
  • NER
  • String
  • Timing
  • Values
  • Black Magic
• Workaround for PyTorch Autocompletion - nnlib.torch

Indices and tables

• Index
• Module Index
• Search Page