Exercise 1 - AirBnB revisited

Introduction

Context: Flashback to AirBnb Data that we used in Workshop 3 of M1

In this workshop we are going to explore the insideairbnb using deep learning techniques. We used the data in a research project some years ago and you can check out a conference presentation below:

1. Build a neural net for price prediction.

Your job is to build a feed-forward network using Keras to predict the price. Can you beat the performance of other models that we explored in M1 (using the same set of independent variables)? Start with a “simpler” baseline model using 1 2 layers only (one input layer one output layer).

2. Tune the network.

Experiment with different set-ups changing e.g.:

• number of neurons
• number of layers
• number of epochs and batch size
• activation functions (reasonable choices)
• optimizers and losses (again: reasonable choices)

Use the Keras documentation and community sources to identify best practices. However, do not spend hours on grid search!

3. Prevent overfitting

This part is related to 2. Explore the training process (over the epochs). Explore and describe 2 common approaches used to prevent overfitting in deep feed-forward neural nets.