# src/streamlit_app.py import os import joblib import streamlit as st # ===================== FIX: Download NLTK corpora for TextBlob ===================== @st.cache_resource def ensure_nltk_corpora(): import nltk nltk_dir = "/tmp/nltk_data" os.makedirs(nltk_dir, exist_ok=True) # Tell NLTK to use this writable folder in HF Spaces os.environ["NLTK_DATA"] = nltk_dir if nltk_dir not in nltk.data.path: nltk.data.path.append(nltk_dir) # TextBlob/NLTK resources needed for tokenization + lemmatization nltk.download("punkt", download_dir=nltk_dir, quiet=True) nltk.download("punkt_tab", download_dir=nltk_dir, quiet=True) # important on many HF setups nltk.download("wordnet", download_dir=nltk_dir, quiet=True) nltk.download("omw-1.4", download_dir=nltk_dir, quiet=True) ensure_nltk_corpora() # ================================================================================ # ---- Custom analyzer MUST exist to load your saved pipeline ---- from textblob import TextBlob from sklearn.feature_extraction.text import ENGLISH_STOP_WORDS stops = ENGLISH_STOP_WORDS def ekkok(text): words = TextBlob(str(text)).words return [ w.lemmatize(pos="v").lower() for w in words if w.lower() not in stops and w.isalpha() ] # --------------------------------------------------------------- st.set_page_config(page_title="Stress Detection (TF-IDF)", page_icon="🧠", layout="centered") @st.cache_resource def load_pipeline(): # If your pickle is in repo root, change to: "tfidf_lr_pipeline.pkl" return joblib.load("src/tfidf_lr_pipeline.pkl") pipeline = load_pipeline() st.title("🧠 Stress Detection from Text") st.write("This app uses a **TF-IDF + Logistic Regression** pipeline to predict **Stress (1)** or **No Stress (0)**.") default_example = "I feel stressed and overwhelmed with my daily tasks." text = st.text_area( "Enter text (e.g., a social media post):", height=180, placeholder="Type or paste text here..." ) col1, col2 = st.columns(2) with col1: predict_btn = st.button("Predict", use_container_width=True) with col2: example_btn = st.button("Use example", use_container_width=True) if example_btn: text = default_example st.session_state["last_text"] = text if "last_text" in st.session_state and not text.strip(): text = st.session_state["last_text"] st.markdown("---") if predict_btn: if not text.strip(): st.warning("Please enter some text.") else: pred = int(pipeline.predict([text])[0]) st.subheader("Result") if pred == 1: st.error("Prediction: **Stress (1)**") else: st.success("Prediction: **No Stress (0)**") if hasattr(pipeline, "predict_proba"): proba = pipeline.predict_proba([text])[0] # [P(0), P(1)] st.write(f"Confidence (No Stress): **{proba[0]*100:.2f}%**") st.write(f"Confidence (Stress): **{proba[1]*100:.2f}%**") st.progress(float(proba[1])) st.caption("Saved model: tfidf_lr_pipeline.pkl (TF-IDF + Logistic Regression)")