This challenge organized by ENS Ulm and Collège de France was about predicting real estate prices with tabular data and images. One of the main the difficulties of this challenge was that we had two types of data: tabular data and images. Moreover, each real estate good contained between 1 and 6 images, which all had different sizes. Therefore, a lot of efforts into data importation had to be put. I decided to make a concatenation of a multilayer perceptron (MLP) for the tabular data and a convolutional neural network (CNN) for the images using Keras.
My rank on the public leaderboard is 8th out of the 167 participants. You can find the leaderboard here: https://challengedata.ens.fr/participants/challenges/68/ranking/public
-The code that I've written (jupyter notebook) for this challenge.
-A PDF file explaining my approach.
Unfortunately, since the data is not publishable, it can't be uploaded here. However, you can download it on this page, if you register with an account: https://challengedata.ens.fr/participants/challenges/68/
Username: VictorHoffmann