Manage Deep Learning Networks with Intel® Optimization for Chainer*

Chainer* is a Python*-based deep learning framework aiming at flexibility and intuition. It provides automatic differentiation APIs based on the define-by-run approach (a.k.a. dynamic computational graphs) as well as object-oriented high-level APIs to build and train neural networks. It supports various network architectures including feed-forward nets, convnets, recurrent nets and recursive nets. It also supports per-batch architectures. Forward computation can include any control flow statements of Python without lacking the ability of backpropagation. It makes code intuitive and easy to debug. Intel® optimization for Chainer, is currently integrated with the latest release of Intel® Math Kernel Library for Deep Neural Networks (Intel® MKL-DNN) 2017 optimized for Intel® Advanced Vector Extensions 2 (Intel® AVX) and Intel® Advanced Vector Extensions 512 (Intel®AVX-512) instructions which are supported in Intel® Xeon® and Intel® Xeon Phi™ processors.

We recommend these Linux* distributions.

• Ubuntu* 14.04/16.04 LTS 64bit
• CentOS* 7 64bit

The following versions of Python can be used:

• 2.7.5+, 3.5.2+, and 3.6.0+

Above recommended environments are tested. We cannot guarantee that Intel® optimization for Chainer works on other environments including Windows* and macOS*, even if Intel optimization for Chainer looks to be running correctly.

Before installing Intel Optimization for Chainer, we recommend to upgrade setuptools if you are using an old one:

$ pip install -U setuptools

The following packages are required to install IntelChainer.

• NumPy 1.9, 1.10, 1.11, 1.12, 1.13
• Six 1.9+
• Swig 3.0.9+

Image dataset support

• pillow 2.3+

HDF5 serialization support

• h5py 2.5+

Testing utilities

• pytest 3.2.5+

Intel® MKL-DNN

• You don’t need to manually install Intel MKL-DNN, when build Intel optimization for Chainer, Intel MKL-DNN will be downloaded and built automatically, thus, boost, glog and gflags are also required.

You can use setup.py to install Intel optimization for Chainer from the tarball:

$ python setup.py install

Use pip to uninstall Chainer:

$ pip uninstall chainer

We are providing the Docker image and Dockerfile for Ubuntu and Centos based on python2 and python3, respectively. For details see: How to build and run Intel optimization for Chainer Docker image.

Training test with mnist dataset:

$ cd examples/mnist
$ python train_mnist.py -g -1

Training test with cifar datasets:

• run the CIFAR-100 dataset:

$ cd examples/cifar
$ python train_cifar.py –g -1 --dataset='cifar100'

• run the CIFAR-10 dataset:

$ cd examples/cifar
$ python train_cifar.py –g -1 --dataset='cifar10'

For Single Node Performance Test Configurations, please refer to following wiki:

https://github.com/intel/chainer/wiki/Intel-Chainer-Single-Node-Performance-Test-Configurations

MIT License (see LICENSE file).

Tokui, S., Oono, K., Hido, S. and Clayton, J., Chainer: a Next-Generation Open Source Framework for Deep Learning, Proceedings of Workshop on Machine Learning Systems(LearningSys) in The Twenty-ninth Annual Conference on Neural Information Processing Systems (NIPS), (2015) URL, BibTex

• Intel® optimization for Chainer github
•  Release notes