tmnt 0.7.50__py3-none-any.whl → 0.7.51__py3-none-any.whl

tmnt/modeling.py

@@ -788,7 +788,7 @@ class SelfEmbeddingCrossEntropyLoss(_Loss):
         - **loss**: loss tensor with shape (batch_size,).
     """
 
-    def __init__(self, teacher_right=True, metric_loss_temp=1.0, batch_axis=0, **kwargs):
+    def __init__(self, teacher_right=True, metric_loss_temp=0.5, batch_axis=0, **kwargs):
         super(SelfEmbeddingCrossEntropyLoss, self).__init__(batch_axis, **kwargs)
         self.cross_entropy_loss = nn.CrossEntropyLoss()
         self.metric_loss_temp = metric_loss_temp
@@ -799,11 +799,14 @@ class SelfEmbeddingCrossEntropyLoss(_Loss):
         the function computes the kl divergence between the negative distances
         and the smoothed label matrix.
         """
+        batch_size = l1.size()[0]
         x1_norm = torch.nn.functional.normalize(x1, p=2, dim=1)
         x2_norm = torch.nn.functional.normalize(x2, p=2, dim=1)
         cross_side_distances = torch.mm(x1_norm, x2_norm.transpose(0,1)) / self.metric_loss_temp
         single_side_distances = torch.mm(x2_norm, x2_norm.transpose(0,1)) / self.metric_loss_temp if self.teacher_right \
             else torch.mm(x1_norm, x1_norm.transpose(0,1)) / self.metric_loss_temp
+        # need to normalize these
+        single_side_distances = single_side_distances / single_side_distances.sum(axis=1,keepdim=True).expand(batch_size, batch_size)
         # multiply by the batch size to obtain the sum loss (kl_loss averages instead of sum)
         return self.cross_entropy_loss(cross_side_distances, single_side_distances.to(single_side_distances.device))
 

{tmnt-0.7.50.dist-info → tmnt-0.7.51.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: tmnt
-Version: 0.7.50
+Version: 0.7.51
 Summary: Topic modeling neural toolkit
 Home-page: https://github.com/mitre/tmnt.git
 Author: The MITRE Corporation

{tmnt-0.7.50.dist-info → tmnt-0.7.51.dist-info}/RECORD

@@ -5,7 +5,7 @@ tmnt/distribution.py,sha256=Pmyc5gwDd_-jP7vLVb0vdNQaSSvF1EuiTZEWg3KfmI8,10866
 tmnt/estimator.py,sha256=i37NVmUseDuEWfk4cwZcShsRrbINLbtrqRzDAPmJUwU,77249
 tmnt/eval_npmi.py,sha256=ODRDMsBgDM__iCNEX399ck7bAhl7ydvgDqmpfR7Y-q4,5048
 tmnt/inference.py,sha256=Sw7GO7QiWVEtbPJKBjFB7AiKRmUOZbFZn3tCrsStzWw,17845
-tmnt/modeling.py,sha256=9aVXSDfiNCM6fzRkvdrvmT1Vv69k_XJ_t12b9I-T_qA,34847
+tmnt/modeling.py,sha256=wKDuUsw2bvsrvJ7LkcnSXAPh8cvUSd8y3Q7eGAf_JeU,35049
 tmnt/preprocess/__init__.py,sha256=gwMejkQrnqKS05i0JVsUru2hDUR5jE1hKC10dL934GU,170
 tmnt/preprocess/tokenizer.py,sha256=-ZgowfbHrM040vbNTktZM_hdl6HDTqxSJ4mDAxq3dUs,14050
 tmnt/preprocess/vectorizer.py,sha256=RkdivqP76qAJDianV09lONad9NbfBVWLZgIbU_P1-zo,15796
@@ -17,9 +17,9 @@ tmnt/utils/ngram_helpers.py,sha256=VrIzou2oQHCLBLSWODDeikN3PYat1NqqvEeYQj_GhbA,1
 tmnt/utils/pubmed_utils.py,sha256=3sHwoun7vxb0GV-arhpXLMUbAZne0huAh9xQNy6H40E,1274
 tmnt/utils/random.py,sha256=qY75WG3peWoMh9pUyCPBEo6q8IvkF6VRjeb5CqJOBF8,327
 tmnt/utils/recalibrate.py,sha256=TmpB8An8bslICZ13UTJfIvr8VoqiSedtpHxec4n8CHk,1439
-tmnt-0.7.50.dist-info/LICENSE,sha256=qFZJrfJ7Zi4IXDiyiGVrHWic_l1h2tc36tI8Z7rK9bs,11356
-tmnt-0.7.50.dist-info/METADATA,sha256=kVFUVgKEDDFi7uUxamXxAcsKUD4QjPvsUJGtLiCrGAM,1443
-tmnt-0.7.50.dist-info/NOTICE,sha256=p0kYIVAkReTFaGb4C-qPa7h5ztze6hGzOpjCMMbOipU,425
-tmnt-0.7.50.dist-info/WHEEL,sha256=GJ7t_kWBFywbagK5eo9IoUwLW6oyOeTKmQ-9iHFVNxQ,92
-tmnt-0.7.50.dist-info/top_level.txt,sha256=RpYgUl187sXnqmiwKjZZdcDlHz2AALs6bGdUcukyd_E,5
-tmnt-0.7.50.dist-info/RECORD,,
+tmnt-0.7.51.dist-info/LICENSE,sha256=qFZJrfJ7Zi4IXDiyiGVrHWic_l1h2tc36tI8Z7rK9bs,11356
+tmnt-0.7.51.dist-info/METADATA,sha256=2KT4XrKVkPIKkkgeO_iEzHrG_ERSwX2Td1KTdyiKMn8,1443
+tmnt-0.7.51.dist-info/NOTICE,sha256=p0kYIVAkReTFaGb4C-qPa7h5ztze6hGzOpjCMMbOipU,425
+tmnt-0.7.51.dist-info/WHEEL,sha256=GJ7t_kWBFywbagK5eo9IoUwLW6oyOeTKmQ-9iHFVNxQ,92
+tmnt-0.7.51.dist-info/top_level.txt,sha256=RpYgUl187sXnqmiwKjZZdcDlHz2AALs6bGdUcukyd_E,5
+tmnt-0.7.51.dist-info/RECORD,,