NLPDisasterTweets / src /streamlit_app.py
EnYa32's picture
Update src/streamlit_app.py
ca4f1f1 verified
Raw History Blame Contribute Delete
3.98 kB
import re
from pathlib import Path
import joblib
import numpy as np
import streamlit as st
# -------------------------
# Page config
# -------------------------
st.set_page_config(
page_title='Disaster Tweet Classifier (TF-IDF + LogReg)',
page_icon='🚨',
layout='centered'
)
st.title('🚨 Disaster Tweet Classifier')
st.write('Classify tweets as **Disaster (1)** or **Non-Disaster (0)** using a TF-IDF + Logistic Regression model.')
# -------------------------
# Paths (HF friendly)
# -------------------------
BASE_DIR = Path(__file__).resolve().parent
MODEL_PATH = BASE_DIR / 'final_clf.pkl'
THRESH_PATH = BASE_DIR / 'threshold.pkl'
# -------------------------
# Helpers (same logic as notebook)
# -------------------------
def build_text(text: str, keyword: str, location: str) -> str:
text = '' if text is None else str(text)
keyword = '' if keyword is None else str(keyword)
location = '' if location is None else str(location)
# same markers as training
return text + ' [KEY] ' + keyword + ' [LOC] ' + location
def clean_text(s: str) -> str:
s = '' if s is None else str(s)
s = s.lower()
s = re.sub(r'http\S+|www\.\S+', ' URL ', s)
s = re.sub(r'@\w+', ' USER ', s)
s = re.sub(r'\n|\r', ' ', s)
s = re.sub(r'\s+', ' ', s)
return s.strip()
@st.cache_resource
def load_artifacts():
if not MODEL_PATH.exists():
raise FileNotFoundError(
f'Model not found: {MODEL_PATH.name}. Put it in the same folder as app.py.'
)
if not THRESH_PATH.exists():
raise FileNotFoundError(
f'Threshold not found: {THRESH_PATH.name}. Put it in the same folder as app.py.'
)
model = joblib.load(MODEL_PATH)
threshold = joblib.load(THRESH_PATH)
# make sure threshold is float
threshold = float(threshold)
return model, threshold
# -------------------------
# Load model
# -------------------------
try:
model, threshold = load_artifacts()
except Exception as e:
st.error(str(e))
st.stop()
# -------------------------
# UI
# -------------------------
with st.form('predict_form'):
text = st.text_area('Tweet text', height=140, placeholder='Paste the tweet text here...')
c1, c2 = st.columns(2)
with c1:
keyword = st.text_input('Keyword (optional)', placeholder='e.g., earthquake')
with c2:
location = st.text_input('Location (optional)', placeholder='e.g., California')
submitted = st.form_submit_button('Predict')
st.caption(f'Using threshold = **{threshold:.2f}** (tuned on validation).')
# Example buttons
with st.expander('Examples (click to copy)'):
st.code('Just happened a terrible car crash', language='text')
st.code('Forest fire near La Ronge Sask. Canada', language='text')
st.code('What a beautiful sunset, the sky is ablaze!', language='text')
# -------------------------
# Prediction
# -------------------------
if submitted:
if not text or str(text).strip() == '':
st.warning('Please enter a tweet text.')
st.stop()
full_text = build_text(text, keyword, location)
full_text = clean_text(full_text)
try:
proba = float(model.predict_proba([full_text])[:, 1][0])
except Exception as e:
st.error(f'Prediction failed: {e}')
st.stop()
pred = int(proba >= threshold)
if pred == 1:
st.success('Prediction: **Disaster (1)**')
else:
st.info('Prediction: **Non-Disaster (0)**')
st.write(f'**Disaster probability:** {proba:.3f}')
# Simple explanation
st.markdown('---')
st.subheader('What was sent to the model?')
st.code(full_text, language='text')
st.markdown('---')
st.subheader('Notes')
st.write(
'- The model uses TF-IDF features (word and bigram features).\n'
'- The threshold was tuned on a validation split to improve F1-score.\n'
'- Keyword and location are appended to the text using special markers.'
)