refactor

atulkum · atulkum · commit 5803311daf64 · 2019-01-19T20:15:06.000-08:00
diff --git a/neural_ner/config.py b/neural_ner/config.py
@@ -1,4 +1,5 @@
 import os
+import torch
 
 class Config(object):
     pass
@@ -43,3 +44,5 @@ class Config(object):
 config.is_cuda = True
 
 config.is_l2_loss = False
+
+config.is_cuda = config.is_cuda and torch.cuda.is_available()
diff --git a/neural_ner/data_utils/batcher.py b/neural_ner/data_utils/batcher.py
@@ -2,9 +2,9 @@
 
 import os
 import numpy as np
-from .utils import load_sentences, prepare_dataset
-from .vocab import Vocab
-from .sentence_utils import pad_items, pad_chars
+from data_utils.utils import load_sentences, prepare_dataset
+from data_utils.vocab import Vocab
+from data_utils.sentence_utils import pad_items, pad_chars
 import torch
 
 class DatasetConll2003(object):
diff --git a/neural_ner/data_utils/utils.py b/neural_ner/data_utils/utils.py
@@ -1,6 +1,6 @@
 import codecs
-from .sentence_utils import prepare_sentence
-from .tag_scheme_utils import update_tag_scheme
+from data_utils.sentence_utils import prepare_sentence
+from data_utils.tag_scheme_utils import update_tag_scheme
 
 def load_sentences(path, tag_scheme):
     sentences = []
diff --git a/neural_ner/data_utils/vocab.py b/neural_ner/data_utils/vocab.py
@@ -3,10 +3,9 @@
 import os
 import codecs
 import re
-import numpy as np
 
-from .sentence_utils import get_char_word_seq, Constants
-from .utils import load_sentences
+from data_utils.sentence_utils import get_char_word_seq, Constants
+from data_utils.utils import load_sentences
 from collections import Counter
 
 def create_freq_map(item_list):
diff --git a/neural_ner/model.py b/neural_ner/model.py
@@ -21,7 +21,8 @@
     torch.cuda.manual_seed_all(123)
 
 def get_model(vocab, config, model_file_path, is_eval=False):
-    model = NER_SOFTMAX_CHAR_CRF(vocab, config)
+    #model = NER_SOFTMAX_CHAR_CRF(vocab, config)
+    model = NER_SOFTMAX_CHAR(vocab, config)
 
     if is_eval:
         model = model.eval()
@@ -120,7 +121,7 @@ def forward(self, batch):
     def neg_log_likelihood(self, logits, y, s_lens):
         log_smx = F.log_softmax(logits, dim=2)
         loss = F.nll_loss(log_smx.transpose(1, 2), y, ignore_index=Constants.TAG_PAD_ID, reduction='none')
-        loss = loss.squeeze(1).sum(dim=1) / s_lens.float()
+        loss = loss.sum(dim=1) / s_lens.float()
         loss = loss.mean()
         return loss
 
