Merge pull request #143 from NLPatVCU/development

Development

.gitignore

@@ -1,3 +1,10 @@
+# medaCy
+*.ann
+*.txt
+
+# macOS
+.DS_Store
+
 # Byte-compiled / optimized / DLL files
 __pycache__/
 *.py[cod]

docs/source/medacy.ner.model.rst

@@ -4,4 +4,5 @@ medacy.ner.model package
 .. toctree::
 
     medacy.ner.model.model
+    medacy.ner.model.spacy_model
     medacy.ner.model.stratified_k_fold

docs/source/medacy.ner.model.spacy_model.rst

@@ -0,0 +1,7 @@
+medacy.ner.model.spacy_model module
+=========================
+
+.. automodule:: medacy.ner.model.spacy_model
+    :members:
+    :undoc-members:
+    :show-inheritance:

medacy/__init__.py

@@ -1,2 +1,2 @@
-__version__ = '0.1.0'
+__version__ = '0.1.1'
 __authors__ = "Andriy Mulyar, Corey Sutphin, Bobby Best, Steele Farnsworth, Bridget McInnes"

medacy/__main__.py

@@ -0,0 +1,113 @@
+import argparse
+import logging
+from datetime import datetime
+import time
+import importlib
+
+from medacy.data import Dataset
+from medacy.ner import Model
+from medacy.ner import SpacyModel
+
+def setup(args):
+    dataset = Dataset(args.dataset)
+
+    pipeline = None
+
+    if args.pipeline == 'spacy':
+        model = SpacyModel
+        return dataset, model
+
+    else:      
+        labels = list(dataset.get_labels())
+
+        pipeline_arg = args.pipeline
+
+        #Parse the argument as a class name in module medacy.ner.pipelines
+        module = importlib.import_module("medacy.ner.pipelines")
+        pipeline_class = getattr(module, pipeline_arg)
+
+        if args.word_embeddings is not None:
+            pipeline = pipeline_class(entities=labels, word_embeddings=args.word_embeddings)
+        else:
+            pipeline = pipeline_class(entities=labels)
+
+
+    model = Model(pipeline)
+
+    return dataset, model
+
+def train(args, dataset, model):
+    if args.filename is None:
+        response = input('No filename given. Continue without saving the model at the end? (y/n) ')
+        if response.lower() == 'y':
+            model.fit(dataset)
+        else:
+            print('Cancelling. Add filename with -f or --filename.')
+    else:
+        model.fit(dataset)
+        model.dump(args.filename)
+
+def predict(args, dataset, model):
+    model.load(args.model_path)
+    model.predict(dataset, prediction_directory=args.predictions, groundtruth_directory=args.groundtruth)
+
+def cross_validate(args, dataset, model):
+    model.cross_validate(num_folds=args.k_folds, training_dataset=dataset, prediction_directory=args.predictions,groundtruth_directory=args.groundtruth)
+
+def main():
+    # Argparse setup
+    parser = argparse.ArgumentParser(prog='medacy', description='Train and evaluate medaCy pipelines.')
+    parser.add_argument('-p', '--print_logs', action='store_true', help='Use to print logs to console.')
+    parser.add_argument('-pl', '--pipeline', default='ClinicalPipeline', help='Pipeline to use for training. Write the exact name of the class.')
+    parser.add_argument('-d', '--dataset', required=True, help='Directory of dataset to use for training.')
+    parser.add_argument('-w', '--word_embeddings', help='Path to word embeddings.')
+    subparsers = parser.add_subparsers()
+
+    # Train arguments
+    parser_train = subparsers.add_parser('train', help='Train a new model.')
+    parser_train.add_argument('-f', '--filename', help='Filename to use for saved model.')
+    parser_train.set_defaults(func=train)
+
+    # Predict arguments
+    parser_predict = subparsers.add_parser('predict', help='Run predictions on the dataset using a trained model.')
+    parser_predict.add_argument('-m', '--model_path', required=True, help='Trained model to load.')
+    parser_predict.add_argument('-gt', '--groundtruth', action='store_true', help='Use to store groundtruth files.')
+    parser_predict.add_argument('-pd', '--predictions', action='store_true', help='Use to store prediction files.')
+    parser_predict.set_defaults(func=predict)
+
+    # Cross Validation arguments
+    parser_validate = subparsers.add_parser('validate', help='Cross validate a model on a given dataset.')
+    parser_validate.add_argument('-k', '--k_folds', default=5, type=int, help='Number of folds to use for cross-validation.')
+    parser_validate.add_argument('-gt', '--groundtruth', action='store_true', help='Use to store groundtruth files.')
+    parser_validate.add_argument('-pd', '--predictions', action='store_true', help='Use to store prediction files.')
+    parser_validate.set_defaults(func=cross_validate)
+
+    # Parse initial args
+    args = parser.parse_args()
+
+    # Logging
+    logging.basicConfig(filename='medacy.log', format='%(asctime)-15s: %(message)s', level=logging.INFO)
+    if args.print_logs:
+        logging.getLogger().addHandler(logging.StreamHandler())
+    start_time = time.time()
+    current_time = datetime.fromtimestamp(start_time).strftime('%Y_%m_%d_%H.%M.%S')
+    logging.info('\nSTART TIME: ' + current_time)
+
+    # Run proper function
+    dataset, model = setup(args)
+    args.func(args, dataset, model)
+
+    # Calculate/print end time
+    end_time = time.time()
+    current_time = datetime.fromtimestamp(end_time).strftime('%Y_%m_%d_%H.%M.%S')
+    logging.info('END TIME: ' + current_time)
+
+    # Calculate/print time elapsed
+    seconds_elapsed = end_time - start_time
+    minutes_elapsed, seconds_elapsed = divmod(seconds_elapsed, 60)
+    hours_elapsed, minutes_elapsed = divmod(minutes_elapsed, 60)
+
+    logging.info('H:M:S ELAPSED: %d:%d:%d' % (hours_elapsed, minutes_elapsed, seconds_elapsed))
+
+if __name__ == '__main__':
+    main()

medacy/data/dataset.py

@@ -182,6 +182,63 @@ def get_data_files(self):
     def __iter__(self):
         return self.get_data_files().__iter__()
 
+    def get_labels(self):
+        """
+        Get all of the entities/labels used in the dataset.
+
+        :return: A set of strings. Each string is a label used.
+        """
+        labels = set()
+        data_files = self.all_data_files
+
+        for datafile in data_files:
+            ann_path = datafile.get_annotation_path()
+            annotations = Annotations(ann_path)
+            labels.update(annotations.get_labels())
+
+        return labels
+
+    def get_training_data(self, data_format='spacy'):
+        """
+        Get training data in a specified format.
+
+        :param data_format: The specified format as a string.
+
+        :return: The requested data in the requested format.
+        """
+        # Only spaCy format is currently supported.
+        if data_format != 'spacy':
+            raise TypeError("Format %s not supported" % format)
+
+        training_data = []
+
+        # Add each entry in dataset with annotation to train_data
+        for data_file in self.all_data_files:
+            txt_path = data_file.get_text_path()
+            ann_path = data_file.get_annotation_path()
+            annotations = Annotations(ann_path, source_text_path=txt_path)
+            training_data.append(annotations.get_entity_annotations(format='spacy'))
+
+        return training_data
+
+    def get_subdataset(self, indices):
+        """
+        Get a subdataset of data files based on indices.
+
+        :param indices: List of ints that represent the indexes of the data files to split off.
+
+        :return: Dataset object with only the specified data files.
+        """
+        subdataset = Dataset(self.data_directory)
+        data_files = subdataset.get_data_files()
+        sub_data_files = []
+
+        for i in range(len(data_files)):
+            if i in indices:
+                sub_data_files.append(data_files[i])
+
+        subdataset.all_data_files = sub_data_files
+        return subdataset
 
     def metamap(self, metamap, n_jobs=multiprocessing.cpu_count() - 1, retry_possible_corruptions=True):
         """

medacy/ner/__init__.py

@@ -1 +1,2 @@
 from .model.model import Model
+from .model.spacy_model import SpacyModel

medacy/ner/model/__init__.py

@@ -1,2 +1,2 @@
 from .model import Model
-from .stratified_k_fold import SequenceStratifiedKFold
+from .stratified_k_fold import SequenceStratifiedKFold

medacy/ner/model/_model.py

@@ -58,8 +58,14 @@ def construct_annotations_from_tuples(doc, predictions):
     annotations = {'entities': {}, 'relations': []}
     T_num = 1
     for prediction in predictions:
-        (entity, start, end) = prediction
-        labeled_text = doc.text[start:end]
+        if len(prediction) == 3:
+            (entity, start, end) = prediction
+            labeled_text = doc.text[start:end]
+        elif len(prediction) == 4:
+            (entity, start, end, labeled_text) = prediction
+        else:
+            raise ValueError("Incorrect prediction length.")
+
         annotations['entities']['T%i' % T_num] = (entity, start, end, labeled_text)
         T_num += 1
-    return Annotations(annotations)
+    return Annotations(annotations)