=================================================== README =================================================== * Dependencies: - Java 1.8 - Postgres 9.3 - Matlab =================================================== * Execution steps 1) Import the provided datasets using the \i command from pgadmin console; 2) Configure database credentials in config.properties; 3) Remove comments ("#") from config.properties for the chosen dataset and learning setting. For example, to invoke TUCKER_CLUS using rank=2 for the Tucker decomposition, remove comments on the following lines: ______________________ dataset=PV_ITALY multiPlant=1 plantAtt=idsito startH=2 endH=20 dateField=data rank=2 table=it_dataset_complete setting=202 numTargets=1 variant=hourly ______________________ 4) Invoke the algorithm as in the scripts provided in the scripts file. Please note that the first invocation should include "-init" in order to create the tables responsible for storing the predictions and the errors. For example, the following command runs the algorithm repeatedly using the dates of the split #2 (subfolder dates) as test dates, with a training sliding window of 30 days: java -jar dm-toolkit.jar -alg TUCKER_CLUS -windowSize 30 -run 2 -init > out.log Predictions and errors will be stored in separate tables in the database ("dm_errors","dm_predictions"). Results can be extracted with a query. For example, the following query allows to extract the average prediction errors (RMSE) for each dataset, algorithm and learning setting: SELECT dataset, algorithm, setting, avg(rmse)as rmse, count(*) FROM dm_errors GROUP BY dataset, algorithm, setting;