issues
search
bzamecnik
/
ml
Machine learning projects, often on audio datasets
MIT License
92
stars
39
forks
source link
Evaluate the errors in pitch class vector prediction
#6
Open
bzamecnik
opened
8 years ago
bzamecnik
commented
8 years ago
overall metrics for the full dataset splits
accuracy (on binary classes, full chords)
hamming distance (on binary classes, individual tones)
AUC (on probabilities, more informative)
metrics on separate songs
how they differ?
confusion matrix (matrices)
12x 2x2 - for each pitch class separately
1x 4096x4096 - for whole pitch class vectors
plot a line of error per time frame
which places are more erroneous?
How to take into consideration class imbalance?
imbalance in tones vs. silence (1 vs. 0)
imbalance across pitch classes