app.py

import streamlit as st
import pandas as pd
import torch
import os
import re
import base64

try:
    from transformers import AutoTokenizer, AutoModelForSequenceClassification
    from huggingface_hub import login
    TRANSFORMERS_AVAILABLE = True
except ImportError as e:
    st.error(f"Failed to import transformers: {str(e)}. Please install it with `pip install transformers`.")
    TRANSFORMERS_AVAILABLE = False

# Set page configuration
st.set_page_config(page_title="WizNerd Insp", page_icon="🚀", layout="centered")

# Custom CSS
st.markdown("""
    <style>
    @import url('https://fonts.googleapis.com/css2?family=Tw+Cen+MT&display=swap');
    html, body, [class*="css"]  {
        font-family: 'Tw Cen MT', sans-serif !important;
    }
    .stTable table {
        font-family: 'Tw Cen MT', sans-serif !important;
    }
    </style>
""", unsafe_allow_html=True)

# Load Hugging Face token
HF_TOKEN = os.getenv("HF_TOKEN")

# Model name
MODEL_NAME = "amiguel/class_insp_program"

# =============================================================================
# FIXED: Label mapping must match EXACTLY what the model was trained with
# The model was trained with 13 classes (Flare Tip and Flare TIP were merged)
# =============================================================================
LABEL_TO_CLASS = {
    0: "Campaign",
    1: "Corrosion Monitoring",
    2: "Flare Tip",  # This now covers both "Flare Tip" and "Flare TIP"
    3: "FU Items",
    4: "Intelligent Pigging",
    5: "Lifting",
    6: "Non Structural Tank",
    7: "Piping",
    8: "Pressure Safety Device",
    9: "Pressure Vessel (VIE)",
    10: "Pressure Vessel (VII)",
    11: "Structure",
    12: "Flame Arrestor"
}

NUM_LABELS = len(LABEL_TO_CLASS)  # Should be 13

# Required columns - UPDATED
REQUIRED_COLS = ["MaintItem text", "Functional Loc.", "Description"]

# Title
st.title("🚀 Scope Inspektor 🚀")

# Avatars
USER_AVATAR = "https://raw.githubusercontent.com/achilela/vila_fofoka_analysis/9904d9a0d445ab0488cf7395cb863cce7621d897/USER_AVATAR.png"
BOT_AVATAR = "https://raw.githubusercontent.com/achilela/vila_fofoka_analysis/991f4c6e4e1dc7a8e24876ca5aae5228bcdb4dba/Ataliba_Avatar.jpg"

# Sidebar
with st.sidebar:
    st.header("Upload Documents 📂")
    uploaded_file = st.file_uploader(
        "Choose an XLSX or CSV file",
        type=["xlsx", "csv"],
        label_visibility="collapsed"
    )
    
    # Show model info
    st.markdown("---")
    st.markdown(f"**Model:** `{MODEL_NAME}`")
    st.markdown(f"**Classes:** {NUM_LABELS}")

# Initialize session state
if "messages" not in st.session_state:
    st.session_state.messages = []
if "file_processed" not in st.session_state:
    st.session_state.file_processed = False
if "file_data" not in st.session_state:
    st.session_state.file_data = None
if "last_uploaded_file" not in st.session_state:
    st.session_state.last_uploaded_file = None


# File processing function with cache
@st.cache_data
def process_file(uploaded_file, _cache_key):
    if uploaded_file is None:
        return None
    
    try:
        # Read file
        if uploaded_file.type == "application/vnd.openxmlformats-officedocument.spreadsheetml.sheet":
            df = pd.read_excel(uploaded_file)
        else:
            df = pd.read_csv(uploaded_file)
        
        # Check if all required columns are present
        missing_cols = [col for col in REQUIRED_COLS if col not in df.columns]
        if missing_cols:
            st.error(f"Missing required columns: {', '.join(missing_cols)}. Please upload a file with 'MaintItem text', 'Functional Loc.', and 'Description'.")
            return None
        
        # Pre-process and concatenate required columns for classification
        df = df.dropna(subset=REQUIRED_COLS, how='all')  # Keep rows with at least some data
        df["input_text"] = df[REQUIRED_COLS].apply(
            lambda row: " ".join([re.sub(r'\s+', ' ', str(val).lower().strip()) for val in row if pd.notna(val)]), axis=1
        )
        
        return {"type": "table", "content": df, "original_df": df.copy()}
        
    except Exception as e:
        st.error(f"📄 Error processing file: {str(e)}")
        return None


# Model loading function - FIXED
@st.cache_resource
def load_model(hf_token):
    if not TRANSFORMERS_AVAILABLE:
        return None
    try:
        if not hf_token:
            st.error("🔐 Please set the HF_TOKEN environment variable.")
            return None
        
        login(token=hf_token)
        
        # Load tokenizer
        tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME, token=hf_token)
        
        # =================================================================
        # FIXED: Load model WITHOUT specifying num_labels
        # Let it auto-detect from config.json, or use ignore_mismatched_sizes
        # =================================================================
        try:
            # First try: Load without specifying num_labels (uses config.json)
            model = AutoModelForSequenceClassification.from_pretrained(
                MODEL_NAME, 
                token=hf_token
            )
        except Exception as e1:
            # Fallback: Try with explicit num_labels and ignore size mismatch
            st.warning(f"Auto-load failed, trying with explicit config: {str(e1)}")
            model = AutoModelForSequenceClassification.from_pretrained(
                MODEL_NAME,
                num_labels=NUM_LABELS,
                token=hf_token,
                ignore_mismatched_sizes=True  # This allows loading even if sizes differ
            )
        
        device = "cuda" if torch.cuda.is_available() else "cpu"
        model.to(device)
        model.eval()
        
        # Log successful load
        st.sidebar.success(f"✅ Model loaded on {device}")
        
        return model, tokenizer
        
    except Exception as e:
        st.error(f"🤖 Model loading failed: {str(e)}")
        import traceback
        st.error(f"Full traceback:\n```\n{traceback.format_exc()}\n```")
        return None


# Classification function - IMPROVED with confidence scores
def classify_instruction(prompt, context, model, tokenizer, return_confidence=False):
    model.eval()
    device = model.device
    
    if isinstance(context, pd.DataFrame):
        predictions = []
        confidences = []
        
        # Process in batches for efficiency
        batch_size = 32
        texts = context["input_text"].tolist()
        
        for i in range(0, len(texts), batch_size):
            batch_texts = texts[i:i + batch_size]
            
            # Prepare inputs
            inputs = tokenizer(
                batch_texts, 
                return_tensors="pt", 
                padding=True, 
                truncation=True, 
                max_length=128
            )
            inputs = {k: v.to(device) for k, v in inputs.items()}
            
            with torch.no_grad():
                outputs = model(**inputs)
                probs = torch.softmax(outputs.logits, dim=-1)
                batch_preds = outputs.logits.argmax(dim=-1).cpu().numpy()
                batch_confs = probs.max(dim=-1).values.cpu().numpy()
            
            for pred, conf in zip(batch_preds, batch_confs):
                # Handle case where prediction ID exceeds our mapping
                if pred in LABEL_TO_CLASS:
                    predictions.append(LABEL_TO_CLASS[pred])
                else:
                    predictions.append(f"Unknown ({pred})")
                confidences.append(float(conf))
        
        if return_confidence:
            return predictions, confidences
        return predictions
    else:
        # Single text classification
        text = str(context) if context else prompt
        inputs = tokenizer(text, return_tensors="pt", padding=True, truncation=True, max_length=128)
        inputs = {k: v.to(device) for k, v in inputs.items()}
        
        with torch.no_grad():
            outputs = model(**inputs)
            probs = torch.softmax(outputs.logits, dim=-1)
            prediction = outputs.logits.argmax().item()
            confidence = probs[0, prediction].item()
        
        pred_label = LABEL_TO_CLASS.get(prediction, f"Unknown ({prediction})")
        
        if return_confidence:
            return pred_label, confidence
        return pred_label


# Excel download function - inserts Item Class before MaintItem text
def get_excel_download_link(df, filename="predicted_classes.xlsx"):
    from io import BytesIO
    
    # Create output dataframe with Item Class before MaintItem text
    output_df = df.copy()
    
    # Get column list and find position of MaintItem text
    cols = list(output_df.columns)
    if "Item Class" in cols:
        cols.remove("Item Class")
    if "Confidence" in cols:
        cols.remove("Confidence")
    
    # Find MaintItem text position
    if "MaintItem text" in cols:
        maint_idx = cols.index("MaintItem text")
        # Insert Item Class and Confidence before MaintItem text
        cols.insert(maint_idx, "Confidence")
        cols.insert(maint_idx, "Item Class")
    else:
        # Fallback: put at beginning
        cols.insert(0, "Confidence")
        cols.insert(0, "Item Class")
    
    # Remove input_text column if present (internal use only)
    if "input_text" in cols:
        cols.remove("input_text")
    
    # Reorder columns
    output_df = output_df[[c for c in cols if c in output_df.columns]]
    
    # Save to Excel
    buffer = BytesIO()
    output_df.to_excel(buffer, index=False, engine='openpyxl')
    buffer.seek(0)
    
    b64 = base64.b64encode(buffer.getvalue()).decode()
    href = f'<a href="data:application/vnd.openxmlformats-officedocument.spreadsheetml.sheet;base64,{b64}" download="{filename}">📥 Download Excel</a>'
    return href


# CSV download function
def get_csv_download_link(df, filename="predicted_classes.csv"):
    # Remove input_text column
    output_df = df.drop(columns=["input_text"], errors="ignore")
    
    # Reorder to put Item Class before MaintItem text
    cols = list(output_df.columns)
    if "Item Class" in cols and "MaintItem text" in cols:
        cols.remove("Item Class")
        if "Confidence" in cols:
            cols.remove("Confidence")
        maint_idx = cols.index("MaintItem text")
        cols.insert(maint_idx, "Confidence")
        cols.insert(maint_idx, "Item Class")
        output_df = output_df[[c for c in cols if c in output_df.columns]]
    
    csv = output_df.to_csv(index=False)
    b64 = base64.b64encode(csv.encode()).decode()
    href = f'<a href="data:file/csv;base64,{b64}" download="{filename}">📥 Download CSV</a>'
    return href


# Load model
if "model" not in st.session_state:
    model_data = load_model(HF_TOKEN)
    if model_data is None and TRANSFORMERS_AVAILABLE:
        st.error("Failed to load model. Check HF_TOKEN.")
        st.stop()
    elif TRANSFORMERS_AVAILABLE:
        st.session_state.model, st.session_state.tokenizer = model_data

model = st.session_state.get("model")
tokenizer = st.session_state.get("tokenizer")

# Check for new file upload and clear cache
if uploaded_file and uploaded_file != st.session_state.last_uploaded_file:
    st.cache_data.clear()
    st.session_state.file_processed = False
    st.session_state.file_data = None
    st.session_state.last_uploaded_file = uploaded_file

# Process uploaded file once
if uploaded_file and not st.session_state.file_processed:
    cache_key = f"{uploaded_file.name}_{uploaded_file.size}"
    file_data = process_file(uploaded_file, cache_key)
    if file_data:
        st.session_state.file_data = file_data
        st.session_state.file_processed = True
        st.write(f"✅ File uploaded with {len(file_data['content'])} rows. Please provide an instruction to classify.")

# Display chat messages
for message in st.session_state.messages:
    avatar = USER_AVATAR if message["role"] == "user" else BOT_AVATAR
    with st.chat_message(message["role"], avatar=avatar):
        st.markdown(message["content"])

# Chat input handling
if prompt := st.chat_input("Ask your inspection question..."):
    if not TRANSFORMERS_AVAILABLE:
        st.error("Transformers library not available.")
        st.stop()

    # Add user message
    with st.chat_message("user", avatar=USER_AVATAR):
        st.markdown(prompt)
    st.session_state.messages.append({"role": "user", "content": prompt})

    # Handle response
    if model and tokenizer:
        try:
            with st.chat_message("assistant", avatar=BOT_AVATAR):
                if st.session_state.file_data:
                    file_data = st.session_state.file_data
                    if file_data["type"] == "table":
                        with st.spinner("Classifying..."):
                            predictions, confidences = classify_instruction(
                                prompt, file_data["content"], model, tokenizer, return_confidence=True
                            )
                        
                        # Add predictions to dataframe
                        result_df = file_data["content"].copy()
                        result_df["Item Class"] = predictions
                        result_df["Confidence"] = [f"{c:.2%}" for c in confidences]
                        
                        # Display preview (first 10 rows)
                        st.write("**Predicted Item Classes (preview):**")
                        display_cols = ["Item Class", "Confidence"] + REQUIRED_COLS
                        st.dataframe(result_df[display_cols].head(10), use_container_width=True)
                        
                        # Stats
                        st.write(f"**Total rows classified:** {len(predictions)}")
                        st.write("**Class distribution:**")
                        st.write(result_df["Item Class"].value_counts())
                        
                        # Average confidence
                        avg_conf = sum(confidences) / len(confidences)
                        st.write(f"**Average confidence:** {avg_conf:.2%}")
                        
                        # Download links
                        st.markdown("---")
                        col1, col2 = st.columns(2)
                        with col1:
                            st.markdown(get_excel_download_link(result_df), unsafe_allow_html=True)
                        with col2:
                            st.markdown(get_csv_download_link(result_df), unsafe_allow_html=True)
                        
                        response = f"✅ Classification completed for {len(predictions)} rows."
                    else:
                        predicted_class, confidence = classify_instruction(
                            prompt, file_data["content"], model, tokenizer, return_confidence=True
                        )
                        response = f"The Item Class is: **{predicted_class}** (confidence: {confidence:.2%})"
                else:
                    predicted_class, confidence = classify_instruction(
                        prompt, "", model, tokenizer, return_confidence=True
                    )
                    response = f"The Item Class is: **{predicted_class}** (confidence: {confidence:.2%})"
                
                st.markdown(response)
                st.session_state.messages.append({"role": "assistant", "content": response})
        except Exception as e:
            st.error(f"⚡ Classification error: {str(e)}")
            import traceback
            st.error(f"```\n{traceback.format_exc()}\n```")
    else:
        st.error("🤖 Model not loaded!")