IDyOM Project

h1. IDyOM Output

{{>toc}}

When used with the <code>:output-path</code> argument, the <code>idyom</code> function writes the output of the modelling to a file. The format is *space-separated values* which can be read directly into software such as R, matlab and excel for further analysis. The output differs depending on the degree of detail requested with the <code>detail</code> argument. The following describes the rows and columns at each level of detail.

h2. detail = 3 (Note level)

Each line in the file represents a note in a composition and each column represents a property of that composition. The columns are:

*Database identifiers*

* dataset.id

* melody.id

* note.id

* melody.name

*Musical properties of the event*

These correspond to the basic viewpoints listed in the [[List of viewpoints]].

* articulation

* comma

* voice

* ornament

* dyn 

* accidental

* mpitch

* phrase

* bioi

* deltast

* barlength

* pulses 

* tempo 

* mode 

* keysig 

* dur 

* cpitch 

* onset

*Properties of the model* 

There are columns representing the order used in predicting the event for each PPM model used in prediction. The column names are labelled according to the target basic viewpoint predicted by the model, the model type (stm or ltm) and the source viewpoint used for prediction (e.g., <code>cpitch.order.ltm.cpint-cpintfref</code>). 

There are columns representing the order used in predicting the event for each PPM model used in prediction. The column names are labelled according to the target basic viewpoint predicted by the model, the model type (stm or ltm) and the source viewpoint used for prediction (e.g., <code>cpitch.weight.ltm.cpint-cpintfref</code>). The weights for the STM/LTM combination are also recorded for each target viewpoint (e.g., <code>cpitch.weight.ltm</code> and <code>cpitch.weight.stm</code>).

*Output of the model*

There are columns representing the probability, information content and entropy estimated by the model for each target viewpoint predicted. For example,

* cpitch.probability

* cpitch.information.content 

* cpitch.entropy

The full distribution is also recorded in columns named to represent both the target viewpoint and the value of the element in the distribution. For example, 

* cpitch.60 

* cpitch.61 

* cpitch.62 

* cpitch.63 

* cpitch.64

* ...

Finally, the overall probability, information.content and entropy are recorded using the following columns:

* probability

* information.content

* entropy

If there is only one target viewpoint, then these values will be equal to those for the single target viewpoint. If there is more than one target viewpoint then these reflect the joint probability distribution computed from the individual distributions for each target viewpoint.

h2. detail = 2 (Composition level)

Each line in the file represents a composition in the dataset and each column represents a property of that composition. The columns are:

* melody.id

* melody.name 

* mean.information.content

h2. detail = 1 (Dataset level)

Not yet implemented. However, the <code>idyom</code> function returns a single number representing the average information content of the entire dataset.