Fix some minor bugs and add instructions

README.md

@@ -15,6 +15,7 @@ This is a PyTorch implementation of the parser described in ["Rethinking Self-At
 
 * Python 3.6 or higher.
 * The Python package requirements can be installed through the `requirements.sh` file.
+* Run `make` in ./EVALB. 
 
 ## Pre-trained models
 

best_parser_training_script.sh

@@ -1,5 +1,5 @@
 #!/usr/bin/env bash
-CUDA_VISIBLE_DEVICES=6 python src_joint/main.py train \
+python src_joint/main.py train \
  --model-path-base models/joint_xlnet_clean_large_3_layers_no_resdrop_lambda \
   --epochs 100 \
  --use-xlnet \
@@ -20,4 +20,4 @@ CUDA_VISIBLE_DEVICES=6 python src_joint/main.py train \
  --dep-dev-ptb-path data/ptb_dev_3.3.0.sd.clean \
  --lal-d-kv 128 \
  --lal-d-proj 128 \
- --no-lal-resdrop
+ --no-lal-resdrop

src_joint/KM_parser.py

@@ -11,7 +11,7 @@
 if use_cuda:
     torch_t = torch.cuda
     def from_numpy(ndarray):
-        return torch.from_numpy(ndarray).pin_memory().cuda(async=True)
+        return torch.from_numpy(ndarray).pin_memory().cuda(non_blocking=True)
 else:
     print("Not using CUDA!")
     torch_t = torch
@@ -331,7 +331,7 @@ def pad_and_rearrange(self, q_s, k_s, v_s, batch_idxs):
         q_padded = q_s.new_zeros((n_head, mb_size, len_padded, d_k))
         k_padded = k_s.new_zeros((n_head, mb_size, len_padded, d_k))
         v_padded = v_s.new_zeros((n_head, mb_size, len_padded, d_v))
-        invalid_mask = q_s.new_ones((mb_size, len_padded), dtype=torch.uint8)
+        invalid_mask = q_s.new_ones((mb_size, len_padded), dtype=torch.bool)
 
         for i, (start, end) in enumerate(zip(batch_idxs.boundaries_np[:-1], batch_idxs.boundaries_np[1:])):
             q_padded[:,i,:end-start,:] = q_s[:,start:end,:]
@@ -983,7 +983,7 @@ def pad_and_rearrange(self, q_s, k_s, v_s, batch_idxs):
             q_padded = q_s.repeat(mb_size, 1, 1) # (d_l * mb_size) x 1 x d_k
         k_padded = k_s.new_zeros((n_head, mb_size, len_padded, d_k))
         v_padded = v_s.new_zeros((n_head, mb_size, len_padded, d_v))
-        invalid_mask = q_s.new_ones((mb_size, len_padded), dtype=torch.uint8)
+        invalid_mask = q_s.new_ones((mb_size, len_padded), dtype=torch.bool)
 
         for i, (start, end) in enumerate(zip(batch_idxs.boundaries_np[:-1], batch_idxs.boundaries_np[1:])):
             if self.q_as_matrix:
@@ -1638,7 +1638,7 @@ def parse_batch(self, sentences, golds=None):
             features = all_encoder_layers[-1]
 
             if self.encoder is not None:
-                features_packed = features.masked_select(all_word_end_mask.to(torch.uint8).unsqueeze(-1)).reshape(-1,
+                features_packed = features.masked_select(all_word_end_mask.to(torch.bool).unsqueeze(-1)).reshape(-1,
                                                                                                                   features.shape[
                                                                                                                       -1])
 
@@ -1736,7 +1736,7 @@ def parse_batch(self, sentences, golds=None):
             # features = all_encoder_layers[-1]
             features = transformer_outputs[0]
 
-            features_packed = features.masked_select(all_word_end_mask.to(torch.uint8).unsqueeze(-1)).reshape(-1,
+            features_packed = features.masked_select(all_word_end_mask.to(torch.bool).unsqueeze(-1)).reshape(-1,
                                                                                                               features.shape[
                                                                                                                   -1])
 
@@ -1779,8 +1779,8 @@ def parse_batch(self, sentences, golds=None):
         else:
             assert self.bert is not None
             features = self.project_bert(features)
-            fencepost_annotations_start = features.masked_select(all_word_start_mask.to(torch.uint8).unsqueeze(-1)).reshape(-1, features.shape[-1])
-            fencepost_annotations_end = features.masked_select(all_word_end_mask.to(torch.uint8).unsqueeze(-1)).reshape(-1, features.shape[-1])
+            fencepost_annotations_start = features.masked_select(all_word_start_mask.to(torch.bool).unsqueeze(-1)).reshape(-1, features.shape[-1])
+            fencepost_annotations_end = features.masked_select(all_word_end_mask.to(torch.bool).unsqueeze(-1)).reshape(-1, features.shape[-1])
 
         fp_startpoints = batch_idxs.boundaries_np[:-1]
         fp_endpoints = batch_idxs.boundaries_np[1:] - 1
@@ -1976,4 +1976,4 @@ def make_tree():
         tree_list = make_tree()
         assert len(tree_list) == 1
         tree = tree_list[0]
-        return tree, score
+        return tree, score

test.sh

@@ -1,7 +1,8 @@
 #!/usr/bin/env bash
 python src_joint/main.py test \
 --dataset ptb \
+--eval-batch-size 8 \
 --consttest-ptb-path data/23.auto.clean \
---deptest-ptb-path data/ptb_test_3.3.0.sd \
+--deptest-ptb-path data/ptb_test_3.3.0.sd.clean \
 --embedding-path data/glove.gz \
 --model-path-base best_parser.pt