prepare_data_long_term only uses a fraction of the dataset

[leonvictor]

Hi,

the paper mentions that training for long term generation is done on Holden and al. 's whole dataset while the preparing process only uses a fraction.

See

QuaterNet/prepare_data_long_term.py

Line 69 in 9d8485b

locomotion_files = get_files(output_directory + '/motionsynth_data/data/processed/edin_locomotion')

where only the edin_locomotion folder is processed (the dataset is composed of this and some other folders)

Is this intended ? If not, now you know !