# app.py
import os
import re
import joblib
import numpy as np
import gradio as gr
# -------------------------
# Helper: safe-loading
# -------------------------
def try_load(path_options):
for p in path_options:
if p is None:
continue
if os.path.exists(p):
try:
model = joblib.load(p)
print(f"Loaded: {p}")
return model, p
except Exception as e:
print(f"Failed to load {p}: {e}")
return None, None
ROOT = os.path.dirname(__file__) if "__file__" in globals() else os.getcwd()
MODEL_DIR = os.path.join(ROOT, "models")
# try multiple plausible names/locations
tfidf_candidates = [
os.path.join(MODEL_DIR, "tfidf_vectorizer.pkl"),
os.path.join(MODEL_DIR, "tfidf.pkl"),
os.path.join(ROOT, "tfidf_vectorizer.pkl"),
os.path.join(ROOT, "tfidf.joblib"),
os.path.join(MODEL_DIR, "tfidf_vectorizer.joblib"),
os.path.join(MODEL_DIR, "tfidf.joblib"),
]
logreg_candidates = [
os.path.join(MODEL_DIR, "logreg_model.pkl"),
os.path.join(MODEL_DIR, "logreg.pkl"),
os.path.join(ROOT, "logreg_model.pkl"),
os.path.join(ROOT, "logreg.pkl"),
os.path.join(MODEL_DIR, "logreg.joblib"),
]
lgbm_candidates = [
os.path.join(MODEL_DIR, "lgbm_model.pkl"),
os.path.join(MODEL_DIR, "lgbm.pkl"),
os.path.join(ROOT, "lgbm_model.pkl"),
os.path.join(ROOT, "lgbm.pkl"),
]
tfidf, tfidf_path = try_load(tfidf_candidates)
logreg, logreg_path = try_load(logreg_candidates)
lgbm, lgbm_path = try_load(lgbm_candidates)
DEFAULT_LABELS = ['negative', 'neutral', 'positive']
# -------------------------
# Text preprocessing
# -------------------------
def clean_text(t):
if t is None:
return ""
s = str(t).lower()
s = re.sub(r"\s+", " ", s)
s = re.sub(r"[^a-z0-9\s']", " ", s)
return s.strip()
# -------------------------
# Prediction logic
# -------------------------
import warnings
warnings.filterwarnings("ignore")
def get_model_classes(model):
if hasattr(model, "classes_"):
return list(model.classes_)
if hasattr(model, "classes"):
return list(model.classes)
return DEFAULT_LABELS
def predict_one(text, model_choice="Logistic Regression"):
text_clean = clean_text(text)
if not text_clean:
return {
"label": "neutral",
"confidence": 0.0,
"html": "No text provided",
"error": None
}
if tfidf is None:
return {"label": None, "confidence": 0.0, "html": "", "error": "Vectorizer (tfidf) not found. Upload tfidf_vectorizer.pkl to models/."}
X = tfidf.transform([text_clean])
try:
if model_choice == "Logistic Regression" and logreg is not None:
probs = logreg.predict_proba(X)[0]
classes = get_model_classes(logreg)
elif model_choice == "LightGBM" and lgbm is not None:
try:
probs = lgbm.predict_proba(X)[0]
except Exception:
probs = lgbm.predict_proba(X.toarray())[0]
classes = get_model_classes(lgbm)
else:
if logreg is not None:
probs = logreg.predict_proba(X)[0]; classes = get_model_classes(logreg)
elif lgbm is not None:
try: probs = lgbm.predict_proba(X)[0]
except: probs = lgbm.predict_proba(X.toarray())[0]
classes = get_model_classes(lgbm)
else:
return {"label": None, "confidence": 0.0, "html": "", "error": "No model found. Upload logreg_model.pkl or lgbm_model.pkl to models/."}
except Exception as e:
return {"label": None, "confidence": 0.0, "html": "", "error": f"Prediction error: {e}"}
idx = int(np.argmax(probs))
label = classes[idx]
confidence = float(probs[idx])
colors = {
'positive': '#16a34a',
'neutral': '#f59e0b',
'negative': '#ef4444'
}
bars_html = ""
for c, p in zip(classes, probs):
col = colors.get(str(c).lower(), "#3b82f6")
pct = float(p) * 100.0
bars_html += f"""
"""
header_html = f"""
Prediction:
{label.upper()} ({confidence:.2f})
{bars_html}
"""
return {"label": label, "confidence": float(confidence), "html": header_html, "error": None}
# -------------------------
# Dark theme CSS
# -------------------------
css = """
body { background: linear-gradient(135deg,#0f172a 0%,#1e293b 100%); color:#e5e7eb; }
.app-card {
border-radius: 12px;
padding: 18px;
box-shadow: 0 10px 25px rgba(0, 0, 0, 0.6);
background: rgba(31,41,55,0.9);
color: #f9fafb;
}
.title {
font-weight: 800;
font-size: 22px;
margin-bottom: 6px;
color: #f9fafb;
}
.subtitle {
color: #9ca3af;
margin-bottom: 12px;
}
.gr-button {
border-radius: 10px;
padding: 10px 16px;
font-weight:700;
background: #3b82f6 !important;
color: white !important;
}
"""
examples = [
["Looking forward to our demo next week! Confirm time please.", "Logistic Regression"],
["Not interested at this time, thanks.", "LightGBM"],
["Can you share pricing and features?", "Logistic Regression"],
]
with gr.Blocks(css=css, theme=gr.themes.Base()) as demo:
with gr.Row():
with gr.Column(scale=2):
gr.HTML("๐ SvaraAI โ Reply Classifier
"
"
Classify replies as positive / neutral / negative.
"
"
โน๏ธ About
"
"
This demo uses a TF-IDF vectorizer and a saved classifier (Logistic Regression / LightGBM). "
"Upload your saved pickles to models/ as described in README.md.
"
f"- TF-IDF: `{tfidf_path or 'NOT FOUND'}` \n"
f"- LogReg: `{logreg_path or 'NOT FOUND'}` \n"
f"- LGBM: `{lgbm_path or 'NOT FOUND'}` \n")
download_note = gr.Markdown("If a model is missing upload it to models/ or rename files appropriately.")
def run_and_format(text, model_choice):
res = predict_one(text, model_choice)
if res.get("error"):
return f"**Error:** {res['error']}", "", gr.update(value=f"{res['error']}
")
label = res["label"]
conf = res["confidence"]
html = res["html"]
md = f"**Prediction:** **{label.upper()}** โ confidence **{conf:.2f}**"
return md, str(round(conf, 3)), gr.update(value=html)
btn.click(run_and_format, inputs=[inp, model_choice], outputs=[output_label, error_box, output_html])
clear.click(lambda: ("**Prediction:** _waiting for input_", "", gr.update(value="Probabilities will appear here")), [], [output_label, error_box, output_html])
gr.HTML("๐ Built for the SvaraAI assignment โข Upload your model pickles into models/
")
if __name__ == "__main__":
demo.launch(server_name="0.0.0.0", server_port=int(os.environ.get("PORT", 7860)))