Merge branch 'master' of https://gitlab.lis-lab.fr/paul.best/f0_estimation

plot_scores_bars.py

@@ -20,7 +20,8 @@ for i, metric in enumerate(metrics):
     for specie in species:
         df = pd.read_csv(f'scores/{specie}_scores{"_minusvocs" if drop_noisy_vocs else ""}{"_minusbins" if drop_noisy_bins else ""}.csv', index_col=0)
         df['Specificity'] = 1 - df['False alarm']
-        ok.loc[specie, df.index] = df[metric]
+        ok.loc[specie.replace('_',' '), df.index] = df[metric]
+    
     ok = ok[algos]
     # bar plot
     m_ax = ax[int(i//2), i%2]

run_all.py

@@ -60,7 +60,7 @@ for specie in species if args.specie =='all' else args.specie.split(' '):
             out = pd.read_csv(f'{fn[:-4]}_preds.csv')
             # check if everything has already been computed, and if yes skip the file
             for algo in algos:
-                if out[algo].isna().all():
+                if algo in out.columns and out[algo].isna().all():
                     out.drop(algo, axis=1, inplace=True)
             if pd.Series(algos).isin(out.columns).all():
                 continue

train_crepe.py

@@ -18,15 +18,15 @@ model.load_state_dict(torch.load('/home/paul.best/.local/lib/python3.9/site-pack
 model = model.train().to('cuda')
 
 FS, n_in, f0 = 16000, 1024, 10
-dt = n_in // 8 # winsize / 8
 norm = lambda s : (s-np.mean(s))/max(1e-8, np.std(s))
 
 if not os.path.isfile(f'crepe_ft/train_set_{suffix}.pkl'):
     df = []
     for specie in ([args.only] if args.only else set(species)-{args.omit} if args.omit else species):
-        wavpath, fs, nfft, downsample = species[specie].values()
+        wavpath, fs, nfft, downsample, step = species[specie].values()
+        dt = int(n_in * step) # winsize / 8
         files = glob.glob(wavpath)
-        for fn in tqdm.tqdm(pd.Series(files).sample(min(len(files), 1000)), desc=specie):
+        for fn in tqdm.tqdm(pd.Series(files).sample(min(len(files), 1000)), desc='Peparing dataset for '+specie):
             if os.path.isfile(f'noisy_pngs/{fn[:-4]}.png'):
                 continue
             annot = pd.read_csv(fn[:-4]+'.csv').drop_duplicates(subset='Time')