Create simplified label files¶

Instead of having to have a labels.csv and a metadata.csv which are matched via another numeric index, we use the clip_id for the labels.csv files so we can match directly with the Mel spectrogram .npz files without the need of the additional *metadata.csv file.

(These files are used in Tutorial Part 1)

In [1]:

import pandas as pd
from os.path import join

In [2]:

# DEFINE PATHS and FILE PATTERNS

METADATA_PATH = 'metadata'

# INPUT
# here, %s will be replaced by 'instrumental', 'genres' or 'moods'
LABEL_FILE_PATTERN = join(METADATA_PATH, 'ismir2018_tut_part_1_%s_labels_subset_post.csv') 
META_FILE_PATTERN = join(METADATA_PATH, 'ismir2018_tut_part_1_%s_metadata_subset.csv') 

# OUTPUT
OUT_LABEL_FILE_PATTERN = join(METADATA_PATH, 'ismir2018_tut_part_1_%s_labels_subset_w_clipid.csv') 

In [3]:

tasks = ['instrumental','genres','moods']

In [4]:

for task in tasks:
    label_file = LABEL_FILE_PATTERN % task
    meta_file = META_FILE_PATTERN % task
    label_file_out = OUT_LABEL_FILE_PATTERN % task

    labels = pd.read_csv(label_file, index_col=0)
    metadata = pd.read_csv(meta_file, index_col=0)

    print("Task:", task)
    print("Labels shape:", labels.shape)
    print("Metadata shape:", metadata.shape)
    print("Assigning clip id from metadata to labels")

    # replace the numeric index in labels by clip_id from metadata column 'clip_id'
    clip_ids_sorted_by_index_of_labels = metadata.loc[labels.index]['clip_id']
    labels.index = clip_ids_sorted_by_index_of_labels
    #print(labels.head())

    # write 
    labels.to_csv(label_file_out)
    print("Created " + label_file_out)

Task: instrumental
Labels shape: (1680, 1)
Metadata shape: (1703, 10)
Assigning clip id from metadata to labels
Created metadata/ismir2018_tut_part_1_instrumental_labels_subset_w_clipid.csv
Task: genres
Labels shape: (1998, 8)
Metadata shape: (1998, 10)
Assigning clip id from metadata to labels
Created metadata/ismir2018_tut_part_1_genres_labels_subset_w_clipid.csv
Task: moods
Labels shape: (719, 4)
Metadata shape: (719, 10)
Assigning clip id from metadata to labels
Created metadata/ismir2018_tut_part_1_moods_labels_subset_w_clipid.csv

In [5]:

# just to check
labels.head()

Out[5]:

	loud	quiet	soft	strange
clip_id
30064	0	0	0	1
5862	0	0	0	0
38362	1	0	0	0
44901	0	0	0	1
16246	0	0	1	1

In [6]:

# just to check
metadata.head()

Out[6]:

	clip_id	mp3_path	track_number	title	artist	album	url	segmentStart	segmentEnd	original_url
13708	30064	D:/Research/Data/MIR/MagnaTagATune/mp3_full/c/...	7	-A Lake-	LVX Nova	LVX Nova	http://www.magnatune.com/artists/albums/lvxnov...	59	88	http://he3.magnatune.com/all/07--A%20Lake--LVX...
2697	5862	D:/Research/Data/MIR/MagnaTagATune/mp3_full/f/...	2	-BWV54 - II Recitative-	American Bach Soloists	J.S. Bach Solo Cantatas	http://www.magnatune.com/artists/albums/abs-so...	30	59	http://he3.magnatune.com/all/02--BWV54%20-%20I...
17495	38362	D:/Research/Data/MIR/MagnaTagATune/mp3_full/0/...	9	-Die Today-	Rocket City Riot	Last Of The Pleasure Seekers	http://www.magnatune.com/artists/albums/rocket...	88	117	http://he3.magnatune.com/all/09--Die%20Today--...
20431	44901	D:/Research/Data/MIR/MagnaTagATune/mp3_full/f/...	11	-In Gaway-	The Headroom Project	Jetuton Andawai	http://www.magnatune.com/artists/albums/headro...	30	59	http://he3.magnatune.com/all/11--In%20Gaway--T...
7423	16246	D:/Research/Data/MIR/MagnaTagATune/mp3_full/f/...	4	-Industrial Blues-	The Headroom Project	Jetuton Andawai	http://www.magnatune.com/artists/albums/headro...	30	59	http://he3.magnatune.com/all/04--Industrial%20...