Neural Networks

Neural Network Libraries

Keras:

• Keras is a neural network library with high-level API to enable rapid prototyping.
• https://github.com/fchollet/keras
• https://keras.io/

Lasagne:

• Lasagne is similar in spirit to Keras, it provides a higher-level API for building neural networks in Theano.
• https://github.com/Lasagne/Lasagne
• https://lasagne.readthedocs.io/en/latest/

Theano:

• Theano is similar in spirit to TensorFlow, Theano is a library intended for tensor calculations. It is not Google-sponsored.

TensorFlow:

• TensorFlow is a neural network library with low-level C++ implementation, and high-level Python API, enabling it to run on a variety of hardware kernels

MXNet:

• MXNet is an Apache project for building neural networks. Again - mainly intended to do tensor calculations.

Gluon:

• Gluon is an API for MXNet intended to provide a higher-level API to enable rapid prototyping.

Neural Network Datasets

Facial recognition/facial matching:

• LFW - Labeled faces in the wild
• Yale Faces - consists of original (very small) and B (larger)
• MegaFace - UW facial recognition dataset
• FDDB - face detection database

Data Engineering Pipelines

Data engineering scenarios:

• https://github.com/data-engineering-scenarios/kaggle-sql-jupyter-keras - SQL data on credit card fraud hosted in Dockerized containers, loaded into Keras+TensorFlow Jupyter notebook running in Docker container

Software related to building neural network pipelines:

Fuel:

• Fuel is a library for creating machine learning pipelines
• https://github.com/mila-udem/fuel

• https://github.com/mila-udem/blocks - blocks
• https://github.com/dribnet/kerosene - kerosene
• https://github.com/dribnet/chips - blocks/fuel helpers