Utilising the Weka learning library for the DANCER Project
This project is using the Weka machine learning library to find patterns in the everyday operation performed by people in their daily life. The data used, should be imported before hand in an InfluxDB library. The data for each one case, comprise of two time series, the context and the actions.
This InfluxDB series contains the data from sensors along with their timestamps. To make their processing easier, all data are included with a single timestamp. If a reading is non existent for a specific time, the 0 or NaN can be set a the value. The identifier of each sensor, is set as the column header for the respective values.
The actions are stored in a separate time series with the same name as the context one, plus the suffix '.actions.'. The format of this time series is the following: time | sequence_number | action | device | parameters | source
WekaDancer is build on the Eclipse IDE, so you can just download the project as a zip or clone the git repository. Then open Eclipse and import the download as an existing project:
- Select the root directory if you clone the git repository or
- Select the archive file if you downloaded as a zip
Multiple libraries are required so load Weka and this is why these have been included in the repository, under the lib/ directory.
However, in order to run the project the correct java compiler should be set in the Eclipse project configuration. This should be 1.7, although both Sun and OpenJRE should work.
It is also compatible with version 1.8 but further testing is needed and minor problems might appear.
To run the project a set of argument should be given. An example is of the required one is as follows: WekaDancer [-h InfluxDB_host] [-d InfluxDB_database] [-s InfluxDB_series] [-f ARFF_filename] [-a algorithm] [-b begin_timestamp] [-e end_timestamp]