Feature selection

The next block is the central one:

features:
 - CDPC_FEATURES
 - CPC_FEATURES
 - PITCH_FEATURES
 - TPC_FEATURES
 - INT_FEATURES
 - INTCLASS_FEATURES
 - FUZZYINT_FEATURES
 - PARSON_FEATURES
 - DURCLASS_FEATURES
 - IOICLASS_FEATURES
 - DURCLASS_RUNLENGTH: durclass_mean_seg_len, veryshort_mean_seg_len, short_mean_seg_len, medium_mean_seg_len, long_mean_seg_len, verylong_mean_seg_len
 - GENERAL_EVENT_DENSITY: event_density
 - GENERAL_SELF_SIM_MATRIX_PHRASES_FEATURES
 - INT_CHROMATIC_SEQUENCES_AV_LEN
 - INT_CHROMATIC_SEQUENCES_RATIO
 - VI_IOI
 - VI_DUR
 - PARSON_CONST_DIRECTION_AV_LEN
 - PARSON_UP_DOWN_RATIO
 - PITCH_WAVINESS
 - PC_AV_NUM_UNIQUE_PC_PHRASES

Here, the features to extract from the melodies in the input set are defined. The Block consists of a YAML list of features, which all start with the name of a feature definition file (without the extension .yml.) The name of a FDF can be followed by a colon : and a comma-separated list of feature labels. A trailing ~ sign excludes the feature. If the list is missing, all features defined in the FDF are used. (Although the demo is called “all features”, the list included here actually does not contain all available features. A complete list can be found here).

This list will trigger melfeature to calculate about 600 features, which will not be exlained here in detail. Instead we will have a look at the result file and examine some of feature values as an example.

Next part: Segmentation.