How-Tos

Using a library

Sometimes you are running a large number of experiments with many aspects in common, for example, a standard metric configuration or a standard cross-validation methodology. It may be convenient to encode fixed aspects of the work in a library that can be shared across multiple studies implemented in librec-auto.

To create an external library, create an XML file that looks like this:

<librec-auto-library>
        <alg name="itemknn50">
                <class>itemknn</class>
                <similarity type="item">pcc</similarity>
                <neighborhood-size>50</neighborhood-size>
                <shrinkage>50</shrinkage>
        </alg>
</librec-auto-library>

Note that each internal element (here alg) has an associated name attribute. This is what is used to look up elements in the library. You can have any number of elements in a library file.

To use an external library in your librec-auto configuration file, you first have to import it. Put the following information at the beginning of your configuration file:

<library>local-library.xml</library>

Replacing “local library” with a path to the library file. You can have multiple library files by adding additional library directives. They will be consulted in the order that they appear in the file.

To use a predefined element from a library file, you only need to reference it by name. For example, the following element is sufficient to use the item kNN algorithm defined above.

<alg ref="itemknn50"/>

You can always override any aspect of the imported element by supplying your own element. For example, to set the shrinkage value to 20, we would say the following:

<alg ref="itemknn50">
        <shrinkage>20</shrinkage>
</alg>

All of the other parts of the algorithm specification would be unchanged.

Producing graphical output

There is a post-processing script for producing basic summary plots of your study. This script was primarily designed for studies that make use of cross-validation. Two types of visualization are produced:

• Bar plots with the average metric value for each experiment in the study, for each metric.
• Box plots that show the distribution of metric values across the different folds of each experiment in a study (if using cross-validation), for each metric.

The script can optionally pop up a browser window that contains the graphics.

In order to use this script, you will need to add it to the post-processing portion of the configuration file. Here is an example:

<script lang="python3" src="system">
        <script-name>result_graphics.py</script-name>
        <param name="browser">true</param>
</script>

The plots are stored in the post directory under the names viz-bar- metric.jpg and viz-box- metric.jpg where metric is the name of the LibRec metric that was calculated.

How to check your configuration files for errors