Initial commit

README.md

@@ -51,18 +51,19 @@ Use pip to install dependencies in your desired python environment
 Download one of the pre-trained models and run the following command:
 
 ```bash
-allennlp predict [path-to-data.jsonl] [path-to-model.tar.gz] \
---predictor predictor_aclarc \
+allennlp predict [path-to-model.tar.gz] [path-to-data.jsonl] \
+--predictor [predictor-type] \
 --include-package scicite \
---output-file [out-path.jsonl]
+--overrides "{'model':{'data_format':''}}"
 ```
 
 Where 
 * `[path-to-data.jsonl]` contains the data in the same format as the training data.
 * `[path-to-model.tar.gz]` is the path to the pretrained model
+* `[predictor-type]` is one of `predictor_scicite` (for the SciCite dataset format) or `predictor_aclarc` (for the ACL-ARC dataset format).
 * `--output-file [out-path.jsonl]` is an optional argument showing the path to the output. If you don't pass this, the output will be printed in the stdout.
 
-You need to convert your data to be according to the training data.
+If you are using your own data, you need to first convert your data to be according to the SciCite data format.
 
 #### Pretrained models
 
@@ -106,7 +107,7 @@ Where the model output and logs will be stored in `[path-to-serialization-dir/]`
 
 ## Citing
 
-If you found our dataset, or code useful, please cite our NAACL 2019 paper:
+If you found our dataset, or code useful, please cite [Structural Scaffolds for Citation Intent Classification in Scientific Publications](https://arxiv.org).
 
 ```
 @InProceedings{Cohan2019Structural,

experiment_configs/aclarc-experiment.json

@@ -1,7 +1,12 @@
+local stringToBool(s) =
+  if s == "true" then true
+  else if s == "false" then false
+  else error "invalid boolean: " + std.manifestJson(s);
+
 local DIM =
-  if std.extVar("elmo")) then
+  if stringToBool(std.extVar("elmo")) then
     1124
-  else:
+  else
     100;
 {
   "random_seed": std.extVar("SEED"),

experiment_configs/scicite-experiment.json

@@ -1,7 +1,12 @@
+local stringToBool(s) =
+  if s == "true" then true
+  else if s == "false" then false
+  else error "invalid boolean: " + std.manifestJson(s);
+
 local DIM =
-  if std.extVar("elmo")) then
+  if stringToBool(std.extVar("elmo")) then
     1224
-  else:
+  else
     200;
 {
   "random_seed": std.extVar("SEED"),

scicite/models/scaffold_bilstm_attention_classifier.py

@@ -185,8 +185,9 @@ def decode(self, output_dict: Dict[str, torch.Tensor]) -> Dict[str, torch.Tensor
         argmax_indices = np.argmax(predictions, axis=-1)
         labels = [self.vocab.get_token_from_index(x, namespace="labels")
                  for x in argmax_indices]
-        output_dict['class_probs'] = class_probabilities
+        output_dict['probabilities'] = class_probabilities
         output_dict['positive_labels'] = labels
+        output_dict['prediction'] = labels
         citation_text = []
         for batch_text in output_dict['citation_text']:
             citation_text.append([self.vocab.get_token_from_index(token_id.item()) for token_id in batch_text])

scicite/predictors/predictor_acl_arc.py

@@ -30,24 +30,15 @@ def predict_json(self, inputs: JsonDict) -> JsonDict:
             intent=citation.intent,
             citing_paper_id=citation.citing_paper_id,
             cited_paper_id=citation.cited_paper_id,
-            citation_excerpt_index=citation.citation_excerpt_index,
-            citation_id=citation.citation_id
+            citation_excerpt_index=citation.citation_excerpt_index
         )
         outputs = self._model.forward_on_instance(instance)
-        predictions = {}
 
-        label_to_index = {v: k for k, v in outputs['all_labels'].items()}
-        for i, prob in enumerate(outputs['class_probs']):
-            predictions[outputs['all_labels'][i]] = prob
-
-        label = max(predictions.items(), key=operator.itemgetter(1))[0]
         return_dict['citation_id'] = citation.citation_id
         return_dict['citingPaperId'] = outputs['citing_paper_id']
         return_dict['citedPaperId'] = outputs['cited_paper_id']
-        return_dict['probabilities'] = predictions
-        return_dict['prediction'] = label
-        return_dict['original_label'] = citation.intent
-        # return_dict['attention_dist'] = outputs['attn_dist']
+        return_dict['probabilities'] = outputs['probabilities']
+        return_dict['prediction'] = outputs['prediction']
         return return_dict
 
     @overrides
@@ -56,7 +47,7 @@ def dump_line(self, outputs: JsonDict) -> str:  # pylint: disable=no-self-use
         If you don't want your outputs in JSON-lines format
         you can override this function to output them differently.
         """
-        keys = ['citation_id', 'prediction', 'probabilities', 'original_label', 'citation_text', 'attention_dist', 'original_citation_text']
+        keys = ['citation_id', 'prediction', 'probabilities', 'citation_text']
         for k in outputs.copy():
             if k not in keys:
                 outputs.pop(k)

scicite/predictors/predictor_scicite.py

@@ -40,23 +40,15 @@ def predict_json(self, inputs: JsonDict) -> JsonDict:
                 citation_excerpt_index=citation.citation_excerpt_index
             )
             outputs = self._model.forward_on_instance(instance)
-            predictions = {}
 
-            label_to_index = {v: k for k, v in outputs['all_labels'].items()}
-            for i, prob in enumerate(outputs['class_probs']):
-                predictions[outputs['all_labels'][i]] = prob
-
-            label = max(predictions.items(), key=operator.itemgetter(1))[0]
-            return_dict['citation_id'] = citation.citation_id
-            return_dict['citingPaperId'] = outputs['citing_paper_id']
-            return_dict['citedPaperId'] = outputs['cited_paper_id']
+            return_dict['citingPaperId'] = outputs.get('citing_paper_id')
+            return_dict['citedPaperId'] = outputs.get('cited_paper_id')
             return_dict['citation_id'] = citation.citation_id
-            return_dict['probabilities'] = predictions
-            return_dict['prediction'] = label
+            return_dict['probabilities'] = outputs.get('probabilities')
+            return_dict['prediction'] = outputs['prediction']
             return_dict['original_label'] = citation.intent
-            return_dict['citation_text'] = outputs['citation_text']
-            return_dict['original_citation_text'] = citation.text
-            return_dict['attention_dist'] = outputs['attn_dist']
+            return_dict['citation_text'] = outputs.get('citation_text')
+            return_dict['attention_dist'] = outputs.get('attn_dist')
         return return_dict
 
     @overrides
@@ -65,7 +57,7 @@ def dump_line(self, outputs: JsonDict) -> str:  # pylint: disable=no-self-use
         If you don't want your outputs in JSON-lines format
         you can override this function to output them differently.
         """
-        keys = ['citedPaperId', 'citingPaperId', 'excerptCitationIntents']
+        keys = ['citedPaperId', 'citingPaperId', 'excerptCitationIntents', 'prediction']
         for k in outputs.copy():
             if k not in keys:
                 outputs.pop(k)

tests/models/model_test_aclarc.py

@@ -17,7 +17,7 @@
 
 sys.path.append(str(Path('.').absolute()))
 
-from scicite.allennlp_modules.train_multitask_two_tasks import train_model_from_file
+from scicite.training.train_multitask_two_tasks import train_model_from_file
 from scicite.constants import root_path
 from scicite.models.scaffold_bilstm_attention_classifier import ScaffoldBilstmAttentionClassifier
 from scicite.dataset_readers.citation_data_reader_scicite import SciciteDatasetReader

tests/models/model_test_scicite.py

@@ -17,7 +17,7 @@
 
 sys.path.append(str(Path('.').absolute()))
 
-from scicite.allennlp_modules.train_multitask_two_tasks import train_model_from_file
+from scicite.training.train_multitask_two_tasks import train_model_from_file
 from scicite.constants import root_path
 from scicite.models.scaffold_bilstm_attention_classifier import ScaffoldBilstmAttentionClassifier
 from scicite.dataset_readers.citation_data_reader_scicite import SciciteDatasetReader

tests/predictors/aclarc_predictor_test.py

@@ -14,7 +14,7 @@
 import sys
 sys.path.append(str(Path('.').absolute()))
 
-from scicite.allennlp_modules.train_multitask_two_tasks import train_model_from_file
+from scicite.training.train_multitask_two_tasks import train_model_from_file
 from scicite.constants import root_path
 
 sys.path.append(root_path)