Update app.py

app.py

@@ -1,28 +1,66 @@
 import gradio as gr
-from transformers import pipeline
+import joblib
+import pandas as pd
+import numpy as np
+import json
+import re
+from nltk.stem import SnowballStemmer
+from sklearn.feature_extraction.text import TfidfVectorizer
 
-# Load your Tigrinya model (replace with your actual model)
-# For now, using a multilingual model as placeholder
-classifier = pipeline(
-    "text-classification",
-    model="Davlan/bert-base-multilingual-cased-finetuned-amharic",  # Replace with your model
-    tokenizer="Davlan/bert-base-multilingual-cased-finetuned-amharic"
-)
+# Load model and metadata
+model = joblib.load('model.joblib')
+with open('feature_names.json', 'r') as f:
+    feature_names = json.load(f)['feature_names']
 
-def detect_harmful(text):
-    if not text.strip():
-        return "Please enter Tigrinya text"
+# Tigrinya text preprocessing
+def preprocess_tigrinya(text):
+    # Basic cleaning
+    text = re.sub(r'[^\w\s]', '', text)
+    text = text.lower()
+    
+    # Simple stemmer (use NLTK's Arabic stemmer as closest match)
+    stemmer = SnowballStemmer("arabic")
+    words = text.split()
+    stemmed = [stemmer.stem(word) for word in words]
+    return " ".join(stemmed)
+
+# Feature extraction
+def extract_features(text):
+    processed = preprocess_tigrinya(text)
     
-    results = classifier(text)
-    return {
-        "Harmful" if res["label"] == "LABEL_1" else "Safe": res["score"] 
-        for res in results
+    # Create feature vector (customize based on your original features)
+    features = {
+        "word_count": len(processed.split()),
+        "unique_words": len(set(processed.split())),
+        "char_count": len(processed),
+        "contains_hate_keyword": int(any(kw in processed for kw in ["ዘሕደረ", "ጸረ"]))
     }
+    
+    # Create DataFrame with all expected features
+    df = pd.DataFrame(columns=feature_names)
+    df = df.append(features, ignore_index=True).fillna(0)
+    return df
+
+# Prediction function
+def predict(text):
+    if not text.strip():
+        return "ጽሑፍ ኣእትዉ! (Please enter text)"
+    
+    features_df = extract_features(text)
+    proba = model.predict_proba(features_df)[0]
+    return {"ጉዳት ዘለዎ (Harmful)": float(proba[1]), 
+            "ሰላማዊ (Safe)": float(proba[0])}
 
+# Gradio interface
 gr.Interface(
-    fn=detect_harmful,
-    inputs=gr.Textbox(label="Enter Tigrinya Text", placeholder="ጽሑፍ ኣብዚ..."),
-    outputs=gr.Label(label="Detection Results"),
+    fn=predict,
+    inputs=gr.Textbox(label="ትግርኛ ጽሑፍ ኣእትዉ (Enter Tigrinya Text)", 
+                     placeholder="እዚ ጽሑፍ ጉዳት ዘለዎ ይመስል..."),
+    outputs=gr.Label(label="ውጽኢት (Prediction)"),
     title="ጉዳት ዘለዎ ጽሑፍ ኣሳታሚ ትግርኛ (Tigrinya Harmful Content Detector)",
-    description="ብትግርኛ ዝተጻሕፈ ጉዳት ዘለዎ ጽሑፍ ይለለጥ።"
+    description="ብትግርኛ ዝተጻሕፈ ጉዳት ዘለዎ ጽሑፍ ይለለጥ።",
+    examples=[
+        ["እዚ ጽሑፍ ጥሩ እዩ"],  # Safe example
+        ["ኣፍታዊ ጥልመት ኣለካ!"]  # Harmful example
+    ]
 ).launch()