def score_model(model, x, b, t):
# First, transform the data
if model is not None:
try:
xt = model.dot(x)
except:
return 0.0
else:
xt = x
# Then, run the segmenter
kmin, kmax = segmenter.get_num_segs(b[-1])
boundary_beats = segmenter.get_segments(xt, kmin=kmin, kmax=kmax)
if len(boundary_beats) < 2 or len(t) < 2:
return 0.0
t = np.unique(t)
boundary_times = mir_eval.util.adjust_events(b[boundary_beats],
t_min=0.0,
t_max=t[-1])[0]
truth_intervals = mir_eval.util.boundaries_to_intervals(t)
pred_intervals = mir_eval.util.boundaries_to_intervals(boundary_times)
score = mir_eval.segment.detection(truth_intervals,
pred_intervals,
trim=True)[-1]
return score
def score_model(model, x, b, t):
# First, transform the data
if model is not None:
try:
xt = model.dot(x)
except:
return 0.0
else:
xt = x
# Then, run the segmenter
kmin, kmax = segmenter.get_num_segs(b[-1])
boundary_beats = segmenter.get_segments(xt, kmin=kmin, kmax=kmax)
if len(boundary_beats) < 2 or len(t) < 2:
return 0.0
t = np.unique(t)
boundary_times = mir_eval.util.adjust_events(b[boundary_beats], t_min=0.0, t_max=t[-1])[0]
truth_intervals = mir_eval.util.boundaries_to_intervals(t)[0]
pred_intervals = mir_eval.util.boundaries_to_intervals(boundary_times)[0]
score = mir_eval.segment.detection(truth_intervals, pred_intervals, trim=True)[-1]
return score
def score_model(model, x, b, t):
# First, transform the data
if model is not None:
xt = model.dot(x)
else:
xt = x
# Then, run the segmenter
kmin, kmax = segmenter.get_num_segs(b[-1])
boundary_beats = segmenter.get_segments(xt, kmin=kmin, kmax=kmax)
if len(boundary_beats) < 2 or len(t) < 2:
return 0.0
boundary_times = mir_eval.util.adjust_events(b[boundary_beats],
t_min=0.0,
t_max=t[-1])[0]
# Convert boundaries to intervals
truth_intervals = mir_eval.util.boundaries_to_intervals(t)[0]
pred_intervals = mir_eval.util.boundaries_to_intervals(boundary_times)[0]
score = mir_eval.segment.boundary_detection(truth_intervals,
pred_intervals)[-1]
return score
make_metadata(annotation)
for tag, bounds in zip(labels, levels):
add_level(annotation, tag, make_intervals(bounds, beats))
return J
if __name__ == '__main__':
parameters = segmenter.process_arguments()
# Load the features
print '- ', os.path.basename(parameters['input_song'])
X, beats = segmenter.features(parameters['input_song'])
# Load the boundary transformation
W_bound = segmenter.load_transform(parameters['transform_boundary'])
print '\tapplying boundary transformation...'
X_bound = W_bound.dot(X)
# Find the segment boundaries
print '\tpredicting segments...'
kmin, kmax = segmenter.get_num_segs(beats[-1])
levels, labels = get_all_levels(X_bound, kmin, kmax)
J = make_jams(levels, labels, beats)
with open(parameters['output_file'], 'w') as f:
json.dump(J, f, indent=2)
