Add initial app files

app.py

@@ -0,0 +1,75 @@
+import gradio as gr
+from spacy import displacy 
+from transformers import (AutoModelForTokenClassification, 
+                          AutoTokenizer, 
+                          pipeline,
+                          )
+
+model_checkpoint = "jsylee/scibert_scivocab_uncased-finetuned-ner"
+
+model = AutoModelForTokenClassification.from_pretrained(model_checkpoint, 
+                                                        num_labels=5,
+                                                        id2label={0: 'O', 1: 'DRUG', 2: 'DRUG', 3: 'ADVERSE EFFECT', 4: 'ADVERSE EFFECT'} # for grouping BIO tags back together
+                                                        )
+tokenizer = AutoTokenizer.from_pretrained(model_checkpoint)
+
+model_pipeline = pipeline(task="ner", model=model, tokenizer=tokenizer, device=0, grouped_entities=True)
+
+def extract_entities(sentence):
+    """ Extract drug and reaction entities, and show using displaCy's NER visualizer.
+    
+    source: https://github.com/jsylee/personal-projects/blob/master/Hugging%20Face%20ADR%20Fine-Tuning/SciBERT%20ADR%20Fine-Tuning.ipynb
+    """
+    tokens = model_pipeline(sentence)
+    entities = []
+    
+    for token in tokens:
+        label = token["entity_group"]
+
+        if label != "0":
+            # label 0 corresponds to "Outside" any entity we care about
+            token["label"] = label
+            entities.append(token)
+    
+    params = [{"text": sentence,
+               "ents": entities,
+               "title": None}]
+    
+    return displacy.render(params, style="ent", manual=True, options={
+        "colors": {
+                   "DRUG": "#f08080",
+                   "ADVERSE EFFECT": "#9bddff",
+               },
+    })
+
+# the following examples of adverse effects are taken from Wikipedia: 
+# https://en.wikipedia.org/wiki/Adverse_effect#Medications
+
+examples = [
+    "Abortion, miscarriage or uterine hemorrhage associated with misoprostol (Cytotec), a labor-inducing drug.",
+    "Addiction to many sedatives and analgesics, such as diazepam, morphine, etc.",
+    "Birth defects associated with thalidomide",
+    "Bleeding of the intestine associated with aspirin therapy",
+    "Cardiovascular disease associated with COX-2 inhibitors (i.e. Vioxx)",
+    "Deafness and kidney failure associated with gentamicin (an antibiotic)",
+    "Death, following sedation, in children using propofol (Diprivan)",
+    "Depression or hepatic injury caused by interferon",
+    "Diabetes caused by atypical antipsychotic medications (neuroleptic psychiatric drugs)"
+]
+
+footer = """
+<hr>
+This app automatically extracts drug names and adverse effects from the input text. An adverse effect occurs when a drug harms a patient in any way.
+
+The extraction is done by a <a href=https://huggingface.co/jsylee/scibert_scivocab_uncased-finetuned-ner>SciBERT model</a> fine-tuned on the <a href=https://huggingface.co/datasets/ade_corpus_v2>`ade_corpus_v2`</a> dataset. Fine-tuning code <a href=https://github.com/jsylee/personal-projects/blob/master/Hugging%20Face%20ADR%20Fine-Tuning/SciBERT%20ADR%20Fine-Tuning.ipynb>here</a>.
+
+This was made during the November 2021 Hugging Face Community Event.
+
+By <a href=http://www.columbia.edu/~jsl2239/>Justin S. Lee</a>
+"""
+
+iface = gr.Interface(fn=extract_entities, inputs=gr.inputs.Textbox(lines=5, placeholder="Abortion, miscarriage or uterine hemorrhage associated with misoprostol..."),
+                     outputs="html", examples=examples,
+                     title="NER for Drug Names and Adverse Effects",
+                     article=footer)
+iface.launch()

requirements.txt

@@ -0,0 +1,2 @@
+spacy == 3.2.0
+transformers == 4.11.3