nlp-architect 0.3.1

Intel AI Lab's open-source NLP and NLU research library

NLP Architect by Intel® AI LAB

NLP Architect is an open-source Python library for exploring state-of-the-art deep learning topologies and techniques for natural language processing and natural language understanding. It is intended to be a platform for future research and collaboration.

Overview

The current version of NLP Architect includes these features that we found interesting from both research perspectives and practical applications:

• NLP core models and NLU modules that provide best in class performance: Intent Extraction (IE), Name Entity Recognition (NER), Word Chunker, Dependency parser (BIST)
• Modules that address semantic understanding: co-locations, most common word sense, noun phrase embedding representation (NP2Vec), relation identification and cross document coreference.
• Components instrumental for conversational AI: ChatBot applications (Memory Networks for Dialog, Key-Value Memory Networks), Intent Extraction.
• End-to-end DL applications using and new topologies: Q&A, Machine Reading Comprehension, Language modeling using Temporal Convolution Networks (TCN), Unsupervised Cross-lingual embeddings, Sparse and quantized GNMT.
• Solutions using one or more models: Set Term expansion which uses the included word chunker as a noun phrase extractor and NP2Vec, Topics and trend analysis for analyzing temporal corpora.

The library consists of core modules (topologies), data pipelines, utilities and end-to-end model examples with training and inference scripts. We look at these as a set of building blocks that were needed for implementing NLP use cases based on our pragmatic research experience. Each of the models includes algorithm descriptions and results in the documentation.

Some of the components, with provided pre-trained models, are exposed as REST service APIs through NLP Architect server. NLP Architect server is designed to provide predictions across different models in NLP Architect. It also includes a web front-end exposing the model annotations for visualizations. The server supports extensions via a template for developers to add a new service. For detailed documentation see this page.

NLP Architect server in action

NLP Architect utilizes the following open source deep learning frameworks:

• TensorFlow
• Intel-Optimized TensorFlow with MKL-DNN
• Dynet

Documentation

Framework documentation on NLP models, algorithms, and modules, and instructions on how to contribute can be found at our main documentation site.

Installation

Prerequisites

Make sure pip and setuptools and venv are up to date before installing.

pip3 install -U pip setuptools

We recommend installing NLP Architect in a virtual environment to self-contain the work done using the library.

To create and activate a new virtual environment:

python3 -m venv .nlp_architect_env
source .nlp_architect_env/bin/activate

Installing using pip

To install NLP Architect using pip package manager:

pip install nlp-architect

Installing from source

To get started, clone our repository:

git clone https://github.com/NervanaSystems/nlp-architect.git
cd nlp-architect

Selecting a backend

NLP Architect supports CPU, GPU and Intel Optimized Tensorflow (MKL-DNN) backends. Users can select the desired backend using a dedicated environment variable (default: CPU). (MKL-DNN and GPU backends are supported only on Linux)

export NLP_ARCHITECT_BE=CPU/MKL/GPU

Installation

NLP Architect is installed using pip and it is recommended to install in development mode.

Default:

pip3 install .

Development mode:

pip3 install -e .

Once installed, the nlp_architect command provides additional options to work with the library, issue nlp_architect -h to see all options.

Packages

Package	Description
nlp_architect.api	Model server API interfaces
nlp_architect.common	Common packages
nlp_architect.contrib	Framework extensions
nlp_architect.data	Datasets, data loaders and data classes
nlp_architect.models	NLP, NLU and End-to-End neural models
nlp_architect.pipelines	End-to-end NLP apps
nlp_architect.server	API Server and demos UI
nlp_architect.solutions	Solution applications
nlp_architect.utils	Misc. I/O, metric, pre-processing and text utilities
examples	Example files for each model
tutorials	Misc. Jupyter tutorials

NLP Architect is an active space of research and development; Throughout future releases new models, solutions, topologies and framework additions and changes will be made. We aim to make sure all models run with Python 3.5+. We encourage researchers and developers to contribute their work into the library.

Citation

If you use NLP Architect in your research, please use the following citation:

@misc{izsak_peter_2018_1477518,
  author       = {Izsak, Peter and
                  Bethke, Anna and
                  Korat, Daniel and
                  Yaccobi, Amit and
                  Mamou, Jonathan and
                  Guskin, Shira and
                  Nittur Sridhar, Sharath and
                  Keller, Andy and
                  Pereg, Oren and
                  Eirew, Alon and
                  Tsabari, Sapir and
                  Green, Yael and
                  Kothapalli, Chinnikrishna and
                  Eavani, Harini and
                  Wasserblat, Moshe and
                  Liu, Yinyin and
                  Boudoukh, Guy and
                  Zafrir, Ofir and
                  Tewani, Maneesh},
  title        = {NLP Architect by Intel AI Lab},
  month        = nov,
  year         = 2018,
  doi          = {10.5281/zenodo.1477518},
  url          = {https://doi.org/10.5281/zenodo.1477518}
}

Disclaimer

The NLP Architect is released as reference code for research purposes. It is not an official Intel product, and the level of quality and support may not be as expected from an official product. Additional algorithms and environments are planned to be added to the framework. Feedback and contributions from the open source and NLP research communities are more than welcome.

Contact

Contact the NLP Architect development team through Github issues or email: [e-mail]