Introduction

Keras is a deeplearning library that encapslutes lower level tensor operations from Theano or Tensorflow. To demonstrate Keras, we'll run over the following examples...

Keras Introduction

Keras vs Poisonous Mushrooms

This tutorial demonstrates a simple classification task implemented with Keras. Hopefully, we won't eat the wrong srooms.

Things we'll cover:

• Extract features from categorical data
• Define a simple neural network
• Define Callbacks (Tensorflow logs)
• Train the network
• Take a peak at the underlying network in Tensorboard
• Evaluate the network against a test dataset

Keras Tensorflow Comparison

Tensorflow vs Poisonous Mushrooms

This tutorial implements the same poisonous mushroom classification task in Tensorflow. The above image is of the poisonous Lepiota Castanea mushroom. While Tensorflow invovles a bit more work, it isn't deadly!

In this tutorial, we will cover:

• Deep dive into the dense layers and the model definition
• Decompose Keras' compile method to Tensorflow operations
• Better understand Keras' fit method (with many lines of Tensorflow code)

Yellowstone Visitors (Monthly Average)

Yellowstone Vistiors Part 1

Before we dive into Reccurent Nerual Networks, we'll look at a Yellowstone recreational visitor timeseries.

Yellowstone Visitors (LSTM)

Yellowstone Vistiors Part 2

This tutorial attempts to predict Yellowstone visitors based on timeseries data. We'll define a Long Short Term Memory network in Keras.

This tutorial will:

• Prepare Time Series Data for the LSTM network
  • Convert non-stationary to stationary data
  • Scale data
  • Create labels from time series data
• Define a stateful LSTM network
• Train the LSTM model
• Compare the LSTM predictions against monthly averages