intent classifier app

app.py

@@ -0,0 +1,42 @@
+import streamlit as st
+import streamlit.components.v1 as components
+from infer_intent import IntentClassifier
+import matplotlib.pyplot as plt
+
+st.set_page_config(layout="wide")
+st.title("Intent classifier")
+
+@st.cache_resource
+def get_intent_classifier():
+    cls = IntentClassifier()
+    return cls
+
+cls = get_intent_classifier()
+query = st.text_input("Enter a query", value="What is the weather today")
+pred_result, proba_result = cls.find_intent(query)
+
+st.markdown(f"prediction = :green[{pred_result}]")
+keys = list(proba_result.keys())
+values = list(proba_result.values())
+
+# Creating the bar plot
+fig, ax = plt.subplots()
+ax.barh(keys, values)
+
+# Adding labels and title
+ax.set_xlabel('Intent')
+ax.set_ylabel('Values')
+ax.set_title('Intents probability score')
+
+col1, col2 = st.columns([2,4])
+
+with col1:
+    st.pyplot(fig)
+
+with col2:
+    exp = st.expander("Explore training data")
+    with exp:
+        html_file = "reports/web_search_intents.html"
+        with open(html_file, 'r', encoding='utf-8') as f:
+            plotly_html = f.read()
+            components.html(plotly_html, height=900, width=900)

infer_intent.py

@@ -0,0 +1,64 @@
+from transformers import AutoModelForSequenceClassification, AutoTokenizer
+import torch
+
+
+class IntentClassifier:
+    def __init__(self):
+        self.id2label = {0: 'information_intent',
+                    1: 'yelp_intent',
+                    2: 'navigation_intent',
+                    3: 'travel_intent',
+                    4: 'purchase_intent',
+                    5: 'weather_intent',
+                    6: 'translation_intent',
+                    7: 'unknown'}
+        self.label2id = {label:id for id,label in self.id2label.items()}
+
+        self.tokenizer = AutoTokenizer.from_pretrained("chidamnat2002/intent_classifier")
+        self.intent_model = AutoModelForSequenceClassification.from_pretrained('chidamnat2002/intent_classifier',
+                                                                        num_labels=8,
+                                                                        torch_dtype=torch.bfloat16,
+                                                                        id2label=self.id2label,
+                                                                        label2id=self.label2id)
+
+    def find_intent(self, sequence, verbose=False):
+        inputs = self.tokenizer(sequence,
+                return_tensors="pt",  # ONNX requires inputs in NumPy format
+                padding="max_length",  # Pad to max length
+                truncation=True,       # Truncate if the text is too long
+                max_length=64)
+
+        self.intent_model.eval()
+        with torch.no_grad():
+            outputs = self.intent_model(**inputs)
+            logits = outputs.logits
+            prediction = torch.argmax(logits, dim=1).item()
+            probabilities = torch.softmax(logits, dim=1)
+            rounded_probabilities = torch.round(probabilities, decimals=3)
+            
+            pred_result = self.id2label[prediction]
+            proba_result = dict(zip(self.label2id.keys(), rounded_probabilities.tolist()[0]))
+            if verbose:
+                print(sequence + " -> " + pred_result)
+                print(proba_result,  "\n")
+            return pred_result, proba_result
+
+
+def main():
+    text_list = [
+        'floor repair cost', 
+        'pet store near me', 
+        'who is the us president', 
+        'italian food',
+        'sandwiches for lunch',
+        "cheese burger cost",
+        "What is the weather today",
+        "what is the capital of usa",
+        "cruise trip to carribean",
+    ]
+    cls = IntentClassifier()
+    for sequence in text_list:
+        cls.find_intent(sequence)
+
+if __name__ == '__main__':
+    main()

requirements.txt

@@ -0,0 +1,4 @@
+transformers==4.45.1
+torch==2.4.1
+streamlit==1.38.0
+matplotlib==3.9.2