After installation you may call tempo_eval from the command line using the tempo_eval command. Calling it without arguments will generate a Markdown-formatted evaluation report in the current directory for all built-in datasets and annotations.

$ tempo_eval


Most likely that is not what you want. To learn about command line options, you may pass the parameter --help.

$ tempo_eval --help

Output Directory

You can change the output directory using the --dir option:

$ tempo_eval --dir MY_OUTPUT_DIR

Specify the Corpus

To specify the datasets for which you want to create evaluation reports, you can pass one or more corpus names using the --corpus parameter:

$ tempo_eval --corpus [CORPUS0 [CORPUS1 [...]]]

Possible corpus names are for example gtzan or ballroom and correspond to the folder names of the built-in annotations.

Output Format

tempo_eval is designed to create Markdown reports which can easily be hosted on GitHub Pages. While Markdown is very readable even without having been rendered, sometimes it is preferable to have an HTML version. This can be generated simply by specifying html as format:

$ tempo_eval --format html

Alternatively, Markdown can be viewed in Chrome with the help of an extension (e.g. with Markdown Preview Plus).

Report Size

By default, tempo_eval generates a very comprehensive report. Not all report elements are always needed (or wanted). To create shorter versions, pass the --size parameter:

$ tempo_eval --size S

Valid values are S, M, and L (the default). XL is reserved for experimental use.

Custom Datasets

In order to generate reports for your custom annotations and estimates you may specify either one on the command line using the --estimates and --references parameters:

$ tempo_eval --references FOLDER_WITH_REFERENCE_JAMS \
             --estimates FOLDER_WITH_ESTIMATE_JAMS

Both the given reference and estimates folders are walked recursively and any found JAMS are read. To make comparison with built-in reference annotations easy, make sure to always specify an annotation_metadata.version and corpus (see also here). In essence, you want to match the used corpus names of the built-in reference annotations, which are lowercase and contain underscores _ instead of spaces.