This repository contains the machine learning component of the recommender engine. The component is written using Apache Spark.

The code in this repository has a simple structure. It does the following things:

• Loads the user reviews from a directory which Flume saves the reviews. The path of the directory is taken from arguments.
• Trains a learning model. (We use ALS algorithm to do that.)
• Saves the learning model to a path, which is specified in the arguments.
• Calculates Root Mean Square Error of the model and writes to the output.

This component cannot be run by itself. In order to run this component, all of the project components must be run using docker-compose. See Recommender Engine - Docker Files

If you want to change the code and run, after you change the code do the following steps:

• Go to the root directory of the repository.
• Run mvn clean package
• Get the recom-engine-docker repository from here.
• Move the jar file that is created from target/recom-engine-ml-1.0-SNAPSHOT.jar to images/spark/master/target under recom-engine-docker's root directory.
• Go to the root diretory of recom-engine-docker and run the command docker-compose up.

There is just one output of this component, the RMSE of the learning model. If you want to see the output, do the following steps:

• First run all of the components with docker-compose.
• Go to the worker's web page from http://localhost:8082.
• In this page, you can see the running drivers, click stdout link of the ml application, then you can watch the outputs of the application.

• Oğuzhan Bölükbaş
• Sercan Ersoy
• Yasin Uygun