added initial NER app

app.py

@@ -0,0 +1,52 @@
+import streamlit as st
+from annotated_text import annotated_text
+import transformers
+
+ENTITY_TO_COLOR = {
+    'PER': '#8ef',
+    'LOC': '#faa',
+    'ORG': '#afa',
+    'MISC': '#fea',
+}
+
+@st.cache(allow_output_mutation=True, show_spinner=False)
+def get_pipe():
+    model_name = "dslim/bert-base-NER"
+    model = transformers.AutoModelForTokenClassification.from_pretrained(model_name)
+    tokenizer = transformers.AutoTokenizer.from_pretrained(model_name)
+    pipe = transformers.pipeline("token-classification", model=model, tokenizer=tokenizer, aggregation_strategy="simple")
+    return pipe
+
+def parse_text(text, prediction):
+    start = 0
+    parsed_text = []
+    for p in prediction:
+        parsed_text.append(text[start:p["start"]])
+        parsed_text.append((p["word"], p["entity_group"], ENTITY_TO_COLOR[p["entity_group"]]))
+        start = p["end"]
+    parsed_text.append(text[start:])
+    return parsed_text
+
+st.set_page_config(page_title="Named Entity Recognition")
+st.title("Named Entity Recognition")
+st.write("Type text into the text box and then press 'Predict' to get the named entities.")
+
+default_text = "My name is John Smith. I work at Microsoft. I live in Paris. My favorite painting is the Mona Lisa."
+
+text = st.text_area('Enter text here:', value=default_text)
+submit = st.button('Predict')
+
+with st.spinner("Loading model..."):
+    pipe = get_pipe()
+
+if (submit and len(text.strip()) > 0) or len(text.strip()) > 0:
+
+    prediction = pipe(text)
+
+    parsed_text = parse_text(text, prediction)
+
+    st.header("Prediction:")
+    annotated_text(*parsed_text)
+
+    st.header('Raw values:')
+    st.json(prediction)

requirements.txt

@@ -0,0 +1,3 @@
+transformers
+torch
+st-annotated-text