For an explanation of the project please view the report
##Data These will be the features we use:
- artist_term
- Echo Nest tags
- danceability (REMOVED, datasheets come back as all 0s)
- between 0 and 1
- duration
- in seconds
- end_of_fade_in
- end time in seconds
- energy (REMOVED, datasheets come back as all 0s)
- between 0 and 1
- key (estimation)
- key signiatures start at 0 (C) and ascend the chomatic scale 1 (D-flat)
- loudness
- measured in dB
- mode (estimation)
- major or minor
- song_hotness (some are NaN)
- algorithmic estimation
- start_of_fade_out
- start time in seconds
- tempo
- in BPM
- time_signature (estimation)
- beats per bar
- track_id
- year
- year of release, 0 if not known
http://labrosa.ee.columbia.edu/millionsong/pages/field-list
To build the csv file for use with MATLAB first download the MillionSongSubset files. Then cd into the directory with getCsvFromData.py
and run the command python getCsvFromData.py <path to MillionSongSubset/data/>
This data folder contains a file merged_song_database.db which is a combination of subset_artist_similarity.db,
subset_artist_term.db, and subset_track_metadata.db from the original subset data.
To get the song_data.csv I ran this query against the database
select
songs.*,
group_concat(artist_term.term, ';')
from songs
inner join artist_term
on artist_term.artist_id = songs.artist_id
group by track_id;which just appends a list of the artist tems separated by a ';' to each track. That should give us a decent start.
You can download any sqlite db browser like [sqlite browser]: http://sqlitebrowser.org/ to run your own SQL queries against it.