Getting Started

dependency

Hadoop is crucially depended by LASER and has to be more than 2.0 version. And other dependencies are managed by maven and will be automatically download.

package

Paste following instruction to your shell.

mvn package

configure

There is a example:

src/main/properties/laser.properties.examble

And you can rename it to XX.properties, where XX is the collection name.

There are four section in configure, e.g. couchbase, metaq, msgpack, laser.

couchbase

Couchbase is the storage for user profile, and LASER retrieves each user's profile via client. If you have a couchbase cluster, cluster should be separated by comma.

metaq

Similar to couchbase, zookeeper should be separated by comma if several address. Besides, users can specify the consumer, or implement their own consumer.

msgpack

LASER and the predict servers communicate each by msgpack and ip should be separated by comma.

laser

The section is the core configure for training framework, output specifies the output directory on HDFS. user_feature_dimension and item_feature_dimension decides the user and item feature dimension. online_retraining_frequency and offline_retraining_frequency specify the retrain frequency, which should be given in quartz's format

run

hadoop jar LASER.jar --configure=configure file

note

The configure file should be on HDFS.