def main(pool):
import museval
results, task_list, eval_out_bundle_path, model_name = main_estimate(pool)
check_task_list(results, task_list, True)
print(results)
method = museval.MethodStore()
method.add_evalstore(results, model_name)
method.save(eval_out_bundle_path)
pool.imap_unordered(func=functools.partial(separate_and_evaluate,
model=args.model,
niter=args.niter,
alpha=args.alpha,
softmask=args.softmask,
output_dir=args.outdir,
eval_dir=args.evaldir),
iterable=mus.tracks,
chunksize=1))
pool.close()
pool.join()
for scores in scores_list:
results.add_track(scores)
else:
results = museval.EvalStore()
for track in tqdm.tqdm(mus.tracks):
scores = separate_and_evaluate(track=track,
model=args.model,
niter=args.niter,
alpha=args.alpha,
softmask=args.softmask,
output_dir=args.outdir,
eval_dir=args.evaldir)
results.add_track(scores)
print(results)
method = museval.MethodStore()
method.add_evalstore(results, 'x-umx')
method.save('x-umx.pandas')
print(target)
print('################')
for tag in tags:
if tag in exclude_tags:
continue
museval_path = os.path.join('exp', tag, 'bss_eval.pkl')
if not os.path.exists(museval_path):
continue
museval_data = pd.read_pickle(museval_path)
# ------------------------------------------------------------------------------------------------------------------
# median over frames, median over tracks
method_median_median = museval.MethodStore(frames_agg='median', tracks_agg='median')
method_median_median.df = museval_data
agg_median_median = method_median_median.agg_frames_tracks_scores()
agg_median_median = pd.DataFrame(agg_median_median, index=None)
median_median = {'tag': [tag], 'target': [target]}
# add museval metrics to method results dicts
for row in agg_median_median.itertuples():
if target == 'all':
metric = row.Index[2]
median_median[metric] = [row.score]
elif row.Index[1] == target:
metric = row.Index[2]
median_median[metric] = [row.score]
import seaborn as sns
import pandas as pd
import matplotlib.pyplot as plt
import museval
comparisons = museval.MethodStore()
comparisons.add_sisec18()
agg_df = comparisons.agg_frames_scores()
sns.set()
sns.set_context("notebook")
metrics = ['SDR']
selected_targets = ['vocals', 'drums', 'bass', 'other']
oracles = ['IBM1', 'IBM2', 'IRM1', 'IRM2', 'MWF', 'IMSK']
# Convert to Pandas Dataframes
agg_df['oracle'] = agg_df.method.isin(oracles)
agg_df = agg_df[agg_df.target.isin(selected_targets)].dropna()
# Get sorting keys (sorted by median of SDR:vocals)
df_sort_by = agg_df[(agg_df.metric == "SDR") & (agg_df.target == "vocals")]
methods_by_sdr = df_sort_by.score.groupby(
df_sort_by.method).median().sort_values().index.tolist()
# df = df[df.target == "vocals"]
g = sns.FacetGrid(agg_df,
row="target",
col="metric",
row_order=selected_targets,