predict_document.py

import joblib
import pandas as pd
from PIL import Image
import pytesseract
from scipy.sparse import hstack, csr_matrix

print("🔄 Loading model and preprocessors...")

bundle = joblib.load("/Users/vidyasagarkaruturi/Downloads/machine learning/src/document_classifier_xgb.pkl")

model = bundle["model"]
word_vectorizer = bundle["word_vectorizer"]
char_vectorizer = bundle["char_vectorizer"]
scaler = bundle["scaler"]

print("✅ Loaded successfully")

# ===============================
# LABEL MAPPING (Important!)
# ===============================
label_map = {
    0: "advertisement",
    1: "budget",
    2: "email",
    3: "file folder",
    4: "form",
    5: "handwritten",
    6: "invoice",
    7: "letter",
    8: "memo",
    9: "news article",
    10: "presentation",
    11: "questionnaire",
    12: "resume",
    13: "scientific publication",
    14: "scientific report",
    15: "specification"
}

# ===============================
# FEATURE EXTRACTION
# ===============================

def extract_features(text):
    return {
        "char_count": len(text),
        "digit_count": sum(c.isdigit() for c in text),
        "uppercase_count": sum(c.isupper() for c in text),
        "currency_count": text.count("$") + text.count("€"),
        "line_count": text.count("\n"),
    }

# ===============================
# PREDICTION FUNCTION
# ===============================

def predict_document(image_path):

    print(f"\n📄 Running OCR on: {image_path}")

    img = Image.open(image_path)
    text = pytesseract.image_to_string(img)
    print("text starts -------------------------------------------")
    print(text)
    print("text ends -------------------------------------------")

    # TEXT FEATURES
    word_features = word_vectorizer.transform([text])
    print("word_features starts -------------------------------------------")
    print(word_features)
    print("word_features ends -------------------------------------------")
    char_features = char_vectorizer.transform([text])
    print("char_features starts -------------------------------------------")
    print(char_features)
    print("char_features ends -------------------------------------------")
    text_features = hstack([word_features, char_features])
    print("text_features starts -------------------------------------------")
    print(text_features)
    print("text_features ends -------------------------------------------")

    # NUMERIC FEATURES
    numeric_dict = extract_features(text)
    print("numeric_dict starts -------------------------------------------")
    print(numeric_dict)
    print("numeric_dict ends -------------------------------------------")
    numeric_df = pd.DataFrame([numeric_dict])
    print("numeric_df starts -------------------------------------------")
    print(numeric_df)
    print("numeric_df ends -------------------------------------------")
    numeric_scaled = scaler.transform(numeric_df)
    print("numeric_scaled starts -------------------------------------------")
    print(numeric_scaled)
    print("numeric_scaled ends -------------------------------------------")
    numeric_sparse = csr_matrix(numeric_scaled)
    print("numeric_sparse starts -------------------------------------------")
    print(numeric_sparse)
    print("numeric_sparse ends -------------------------------------------")

    # COMBINE
    final_features = hstack([text_features, numeric_sparse])
    print("final_features starts -------------------------------------------")
    print(final_features)
    print("final_features ends -------------------------------------------")

    # PREDICT
    prediction = model.predict(final_features)[0]
    probability = model.predict_proba(final_features).max()

    print("\n🎯 Prediction:")
    print("Document Type:", label_map[prediction])
    print("Confidence:", round(probability * 100, 2), "%")


# ===============================
# RUN
# ===============================

if __name__ == "__main__":

    # image_path = "/Users/vidyasagarkaruturi/Desktop/Screenshot 2025-10-31 at 9.25.49 PM.png"
    image_path="/Users/vidyasagarkaruturi/Desktop/Screenshot 2025-10-13 at 9.20.04 AM.png" 
    predict_document(image_path)