Sequence models

This repo contains a walk-through RNNs (recurrent neural networks), the family of neural networks for processing sequential data.

The material focuses implementation using Keras Sequential API and is derived from or largely inspired by:

• sequence notes by Adam Green
• Keras Intro by Mike Irvine
• Machine Learning Mastery's Keras Tutorial
• Understanding LSTM Networks from Cohlah's blog

Original content developed and maintained by Dania Meira - meira.dania@gmail.com

Distributed under the MIT license.

Getting started with Keras

Install Tensorflow and Keras in a new conda enrironment:

$ conda create -n keras_cpu -y pip jupyterlab pandas scikit-learn matplotlib
$ conda activate keras_cpu
$ pip install --ignore-installed --upgrade tensorflow
$ pip install keras

Table of Contents

00-keras-intro-sequential-api.ipynb

• Keras Sequential API first look
• Binary classification problem
• Toy Dataset
• Simple multi-layer perceptron with one hidden layer

01-keras-introduction-mnist.ipynb

• Keras Sequential API
• Multi-classification problem
• MNIST dataset

02-recurrent.ipynb

• recurrent neural network - motivations & mechanics
• character level language modeling

03-lstm.ipynb

• LSTMs - motivations & mechanics
• GRUs
• sin wave prediction

Further Resources

• Directory of tutorials and open-source code repositories for working with Keras
• deeplearning.ai Sequence Course
• Keras Blog
• How to configure the number of layers and nodes in a neural network
• Time series prediction with LSTM