[Pycon] [new paper] "Cenk Bircanoğlu" - House Price Prediction with Distributed Keras

[info a pycon.it]

Title: House Price Prediction with Distributed Keras
Duration: 45 (includes Q&A)
Q&A Session: 0
Language: en
Type: Talk

Abstract: Regression problem is the first problem that every machine learning courses start with. Finding relationship between the variables with the statistical processes are the main idea of regression problems. In this study, we worked on the dataset of Kaggle that is called House Prices: Advanced Regression Techniques. Different network models experimented on Keras in a distributed way with Spark. Elephas is an extension of Keras, which allows you to run distributed deep learning models at scale with Spark. In this talk, first I am going to give information about MultiLayer Neural Networks. Then, as you know, Spark ML library already has the MultiLayer Neural Network implementation in it. Due to the fact that Spark ML can not run on GPU, Keras may be the best alternative to Spark ML. But Keras can not run distributedly. Elephas can make it doable, to run on GPU and also distributed. In this talk, I going to give information about Elephas, and point out the positive and negative parts of it as a comparison between Keras and Spark ML. However, I will also provide code samples on the House Prices dataset and compare them to values such as time complexity, accuracy results, and so on.

Tags: [u'distributed-systems', u'Keras', u'spark', u'regression', u'Deep-Learning']