remove 'num_keyword' parameter

examples/hi_xiaowen/s0/run.sh

@@ -101,7 +101,6 @@ if [ ${stage} -le 3 ] && [ ${stop_stage} -ge 3 ]; then
     --batch_size 256 \
     --checkpoint $score_checkpoint \
     --score_file $result_dir/score_longwav.txt  \
-    --num_keywords $num_keywords  \
     --num_workers 8
 
   for keyword in 0 1; do

kws/bin/compute_det_longwav.py

@@ -93,3 +93,4 @@ if __name__ == '__main__':
                                                        false_alarm_per_hour,
                                                        false_reject_rate))
             threshold += args.step
+            

kws/bin/score_longwav.py

@@ -58,9 +58,6 @@ def get_args():
     parser.add_argument('--score_file',
                         required=True,
                         help='output score file')                     
-    parser.add_argument('--num_keywords',
-                        required=True,
-                        help='the number of keywords')                        
     parser.add_argument('--jit_model',
                         action='store_true',
                         default=False,
@@ -106,22 +103,22 @@ def main():
         device = torch.device('cuda' if use_cuda else 'cpu')
     model = model.to(device)
     model.eval()
-
     score_abs_path = os.path.abspath(args.score_file)
-    num_keywords = int(args.num_keywords)
     with torch.no_grad(), open(score_abs_path, 'w', encoding='utf8') as fout:
         for batch_idx, batch in enumerate(test_data_loader):
             keys, feats, target, lengths = batch
             feats = feats.to(device)
             lengths = lengths.to(device)
             logits = model(feats)
+            num_keywords = logits.shape[2]
             logits = logits.cpu()
             for i in range(len(keys)):
                 key = keys[i]
                 score = logits[i][:lengths[i]]
                 for keyword_i in range(num_keywords):
                     keyword_scores = score[:, keyword_i]
-                    score_frames = ' '.join(['{:.3g}'.format(x) for x in keyword_scores.tolist()])
+                    score_frames = ' '.join(['{:.6f}'.format(x) 
+                                            for x in keyword_scores.tolist()])
                     fout.write('{} {}\n'.format(key, score_frames))
             if batch_idx % 10 == 0:
                 print('Progress batch {}'.format(batch_idx))

learnkws/learn_mask.py

@@ -0,0 +1,30 @@
+'''
+Date: 2022-03-04 18:10:52
+LastEditors: Cyan
+LastEditTime: 2022-03-07 10:21:34
+'''
+
+import torch
+
+
+def padding_mask(lengths: torch.Tensor) -> torch.Tensor:
+    """
+    Examples:
+        >>> lengths = torch.tensor([2, 2, 3], dtype=torch.int32)
+        >>> mask = padding_mask(lengths)
+        >>> print(mask)
+        tensor([[False, False,  True],
+                [False, False,  True],
+                [False, False, False]])
+    """
+    batch_size = lengths.size(0)
+    max_len = int(lengths.max().item())
+    seq = torch.arange(max_len, dtype=torch.int64, device=lengths.device)
+    seq = seq.expand(batch_size, max_len)
+    return seq >= lengths.unsqueeze(1)
+
+if __name__ == '__main__':
+    lengths = torch.tensor([2, 2, 3], dtype=torch.int32)
+    print(lengths.numel())
+    mask = padding_mask(lengths)
+    print(mask, mask.size())

learnkws/test_learn.py

@@ -0,0 +1,13 @@
+'''
+Date: 2022-03-04 18:10:52
+LastEditors: Cyan
+LastEditTime: 2022-03-07 10:21:34
+'''
+
+if __name__ == '__main__':
+    a = [1,2,3,4,5,6,7]
+    for i in range(len(a)):
+        print('i = ', i)
+        if a[i] >= 3:
+            i += 2
+        # print('a[i] = ' , a[i])