app.py

from fastapi import FastAPI, UploadFile, Form, HTTPException, Depends, status, File
from fastapi.middleware.cors import CORSMiddleware
from fastapi.responses import JSONResponse, StreamingResponse
import pandas as pd
import numpy as np
from sklearn.naive_bayes import CategoricalNB
from sklearn.preprocessing import LabelEncoder
from sklearn.model_selection import train_test_split
from sklearn.metrics import confusion_matrix
import json
import io
from typing import Dict, List, Optional, Any
from pydantic import BaseModel, Field
import matplotlib.pyplot as plt
import seaborn as sns
from fastapi.encoders import jsonable_encoder

app = FastAPI(
    title="Categorical Naive Bayes API",
    description="API for uploading CSVs, training a Categorical Naive Bayes model, and making predictions.",
    version="1.0.0"
)

app.add_middleware(
    CORSMiddleware,
    allow_origins=["*"],
    allow_credentials=True,
    allow_methods=["*"],
    allow_headers=["*"],
)

class TrainOptions(BaseModel):
    target_column: str = Field(..., description="The name of the target column.")
    feature_columns: List[str] = Field(..., description="List of feature column names.")

class PredictionFeatures(BaseModel):
    features: Dict[str, str] = Field(..., description="Dictionary of feature values for prediction.")

class UploadResponse(BaseModel):
    message: str
    columns: List[str]
    column_types: Dict[str, str]
    unique_values: Dict[str, List[Any]]
    row_count: int

class TrainResponse(BaseModel):
    message: str
    accuracy: float
    target_classes: List[str]

class PredictResponse(BaseModel):
    prediction: str
    probabilities: Dict[str, float]

class ModelState:
    def __init__(self):
        self.model: Optional[CategoricalNB] = None
        self.feature_encoders: Dict[str, LabelEncoder] = {}
        self.target_encoder: Optional[LabelEncoder] = None
        self.X_test: Optional[pd.DataFrame] = None
        self.y_test: Optional[np.ndarray] = None

model_state = ModelState()

def get_model_state():
    return model_state

@app.get("/api/health", tags=["Health"], summary="Health Check", response_model=Dict[str, str])
async def health_check():
    """Check API health."""
    return {"status": "healthy"}

@app.post("/api/upload", tags=["Data"], summary="Upload CSV File", response_model=UploadResponse, status_code=status.HTTP_200_OK)
async def upload_csv(
    file: UploadFile = File(..., description="CSV file to upload")
) -> UploadResponse:
    """Upload a CSV file and get metadata about its columns."""
    if not file.filename or not file.filename.lower().endswith('.csv'):
        raise HTTPException(
            status_code=status.HTTP_400_BAD_REQUEST, 
            detail="Only CSV files are allowed"
        )
    
    try:
        contents = await file.read()
        # Check if the file content is valid
        if len(contents) == 0:
            raise HTTPException(
                status_code=status.HTTP_400_BAD_REQUEST,
                detail="Uploaded file is empty"
            )
        
        # Try to parse the CSV
        try:
            df = pd.read_csv(io.StringIO(contents.decode('utf-8')))
        except UnicodeDecodeError:
            # Try another encoding if UTF-8 fails
            try:
                df = pd.read_csv(io.StringIO(contents.decode('latin-1')))
            except Exception:
                raise HTTPException(
                    status_code=status.HTTP_400_BAD_REQUEST,
                    detail="Unable to decode CSV file. Please ensure it's properly formatted."
                )
        except Exception as e:
            raise HTTPException(
                status_code=status.HTTP_400_BAD_REQUEST,
                detail=f"Error parsing CSV: {str(e)}"
            )
        
        if df.empty:
            raise HTTPException(
                status_code=status.HTTP_400_BAD_REQUEST,
                detail="CSV file contains no data"
            )
        
        # Process the data
        columns = df.columns.tolist()
        column_types = {col: str(df[col].dtype) for col in columns}
        
        # Limit the number of unique values to prevent excessive response sizes
        unique_values = {}
        for col in columns:
            unique_vals = df[col].unique().tolist()
            # Limit to 100 values max to prevent excessive response size
            if len(unique_vals) > 100:
                unique_values[col] = unique_vals[:100] + ["... (truncated)"]
            else:
                unique_values[col] = unique_vals
                
        # Convert NumPy objects to Python native types
        for col, values in unique_values.items():
            unique_values[col] = [v.item() if isinstance(v, np.generic) else v for v in values]
            
        return UploadResponse(
            message="File uploaded and processed successfully",
            columns=columns,
            column_types=column_types,
            unique_values=unique_values,
            row_count=len(df)
        )
    except HTTPException:
        # Re-raise HTTP exceptions
        raise
    except Exception as e:
        raise HTTPException(
            status_code=status.HTTP_500_INTERNAL_SERVER_ERROR, 
            detail=f"An unexpected error occurred: {str(e)}"
        )

@app.post("/api/train", tags=["Model"], summary="Train Model", response_model=TrainResponse, status_code=status.HTTP_200_OK)
async def train_model(
    file: UploadFile = File(..., description="CSV file to train on"),
    options: TrainOptions = Depends(),
    state: ModelState = Depends(get_model_state)
) -> TrainResponse:
    """Train a Categorical Naive Bayes model on the uploaded CSV.
    
    Parameters:
    - file: CSV file with the training data
    - options: Training options specifying target column and feature columns
    """
    try:
        contents = await file.read()
        
        try:
            df = pd.read_csv(io.StringIO(contents.decode('utf-8')))
        except UnicodeDecodeError:
            df = pd.read_csv(io.StringIO(contents.decode('latin-1')))
            
        # Validate columns exist in the DataFrame
        missing_columns = [col for col in [options.target_column] + options.feature_columns 
                          if col not in df.columns]
        if missing_columns:
            raise HTTPException(
                status_code=status.HTTP_400_BAD_REQUEST,
                detail=f"Columns not found in CSV: {', '.join(missing_columns)}"
            )
        
        # Initialize data structures
        X = pd.DataFrame()
        state.feature_encoders = {}
        
        # Encode features
        for column in options.feature_columns:
            if df[column].isna().any():
                raise HTTPException(
                    status_code=status.HTTP_400_BAD_REQUEST,
                    detail=f"Column '{column}' contains missing values. Please preprocess your data."
                )
            encoder = LabelEncoder()
            X[column] = encoder.fit_transform(df[column])
            state.feature_encoders[column] = encoder
            
        # Encode target
        if df[options.target_column].isna().any():
            raise HTTPException(
                status_code=status.HTTP_400_BAD_REQUEST,
                detail=f"Target column '{options.target_column}' contains missing values."
            )
            
        state.target_encoder = LabelEncoder()
        y = state.target_encoder.fit_transform(df[options.target_column])
        
        # Train/test split
        X_train, X_test, y_train, y_test = train_test_split(
            X, y, test_size=0.2, random_state=42, stratify=y if len(np.unique(y)) > 1 else None
        )
        
        # Train the model
        state.model = CategoricalNB()
        state.model.fit(X_train, y_train)
        accuracy = float(state.model.score(X_test, y_test))
        state.X_test = X_test
        state.y_test = y_test
        
        return TrainResponse(
            message="Model trained successfully",
            accuracy=accuracy,
            target_classes=list(state.target_encoder.classes_)
        )
    except HTTPException:
        raise
    except Exception as e:
        raise HTTPException(
            status_code=status.HTTP_500_INTERNAL_SERVER_ERROR,
            detail=f"Model training failed: {str(e)}"
        )

@app.post("/api/predict", tags=["Model"], summary="Predict", response_model=PredictResponse, status_code=status.HTTP_200_OK)
async def predict(
    features: PredictionFeatures,
    state: ModelState = Depends(get_model_state)
) -> PredictResponse:
    """Predict the target class for given features using the trained model.
    
    Parameters:
    - features: Dictionary of feature values for prediction
    """
    if state.model is None:
        raise HTTPException(
            status_code=status.HTTP_400_BAD_REQUEST, 
            detail="Model not trained yet. Please train a model before making predictions."
        )
    
    try:
        # Validate that all required features are provided
        required_features = set(state.feature_encoders.keys())
        provided_features = set(features.features.keys())
        
        missing_features = required_features - provided_features
        if missing_features:
            raise HTTPException(
                status_code=status.HTTP_400_BAD_REQUEST,
                detail=f"Missing required features: {', '.join(missing_features)}"
            )
            
        # Validate extra features
        extra_features = provided_features - required_features
        if extra_features:
            raise HTTPException(
                status_code=status.HTTP_400_BAD_REQUEST,
                detail=f"Unexpected features provided: {', '.join(extra_features)}"
            )
        
        # Encode features
        encoded_features = {}
        for column, value in features.features.items():
            try:
                encoded_features[column] = state.feature_encoders[column].transform([value])[0]
            except ValueError:
                # Handle unknown category values
                raise HTTPException(
                    status_code=status.HTTP_400_BAD_REQUEST,
                    detail=f"Unknown value '{value}' for feature '{column}'. Allowed values: {', '.join(map(str, state.feature_encoders[column].classes_))}"
                )
                
        # Make prediction
        X = pd.DataFrame([encoded_features])
        prediction = state.model.predict(X)
        prediction_proba = state.model.predict_proba(X)
        predicted_class = state.target_encoder.inverse_transform(prediction)[0]
        
        # Generate probabilities
        class_probabilities = {
            state.target_encoder.inverse_transform([i])[0]: float(prob)
            for i, prob in enumerate(prediction_proba[0])
        }
        
        return PredictResponse(
            prediction=predicted_class,
            probabilities=class_probabilities
        )
    except HTTPException:
        raise
    except Exception as e:
        raise HTTPException(
            status_code=status.HTTP_500_INTERNAL_SERVER_ERROR, 
            detail=f"Prediction failed: {str(e)}"
        )

@app.get(
    "/api/plot/confusion-matrix", 
    tags=["Model"], 
    summary="Confusion Matrix Plot",
    response_class=StreamingResponse,
    responses={
        200: {
            "content": {"image/png": {}},
            "description": "PNG image of confusion matrix"
        },
        400: {
            "description": "Model not trained or no test data available"
        }
    }
)
async def plot_confusion_matrix(state: ModelState = Depends(get_model_state)):
    """Return a PNG image of the confusion matrix for the last test set."""
    if state.model is None or state.X_test is None or state.y_test is None:
        raise HTTPException(
            status_code=status.HTTP_400_BAD_REQUEST, 
            detail="Model not trained or no test data available."
        )
    
    try:
        y_pred = state.model.predict(state.X_test)
        cm = confusion_matrix(state.y_test, y_pred)
        
        # Create plot
        fig, ax = plt.subplots(figsize=(7, 6))
        cax = ax.matshow(cm, cmap=plt.cm.Blues)
        plt.title('Confusion Matrix')
        plt.xlabel('Predicted')
        plt.ylabel('Actual')
        plt.colorbar(cax)
        
        # Add labels
        classes = state.target_encoder.classes_ if state.target_encoder else []
        ax.set_xticks(np.arange(len(classes)))
        ax.set_yticks(np.arange(len(classes)))
        ax.set_xticklabels(classes, rotation=45, ha="left")
        ax.set_yticklabels(classes)
        
        # Add numbers to the plot
        for (i, j), z in np.ndenumerate(cm):
            ax.text(j, i, str(z), ha='center', va='center', 
                   color='white' if cm[i, j] > cm.max() / 2 else 'black')
        
        plt.tight_layout()
        
        # Save to buffer
        buf = io.BytesIO()
        plt.savefig(buf, format='png', dpi=150)
        plt.close(fig)
        buf.seek(0)
        
        # Create response with cache control headers
        response = StreamingResponse(buf, media_type="image/png")
        response.headers["Cache-Control"] = "max-age=3600"  # Cache for 1 hour
        return response
    except Exception as e:
        raise HTTPException(
            status_code=status.HTTP_500_INTERNAL_SERVER_ERROR, 
            detail=f"Failed to generate confusion matrix: {str(e)}"
        )

@app.get(
    "/api/plot/feature-log-prob", 
    tags=["Model"], 
    summary="Feature Log Probability Heatmap",
    response_class=StreamingResponse,
    responses={
        200: {
            "content": {"image/png": {}},
            "description": "PNG heatmap of feature log probabilities"
        },
        400: {
            "description": "Model not trained"
        }
    }
)
async def plot_feature_log_prob(state: ModelState = Depends(get_model_state)):
    """Return a PNG heatmap of feature log probabilities for each class."""
    if state.model is None or state.target_encoder is None:
        raise HTTPException(
            status_code=status.HTTP_400_BAD_REQUEST, 
            detail="Model not trained."
        )
    
    try:
        feature_names = list(state.feature_encoders.keys())
        class_names = list(state.target_encoder.classes_)
        
        # Calculate plot size based on data
        fig_height = max(4, 2 * len(feature_names))
        fig, axes = plt.subplots(len(feature_names), 1, figsize=(10, fig_height))
        if len(feature_names) == 1:
            axes = [axes]
            
        for idx, feature in enumerate(feature_names):
            encoder = state.feature_encoders[feature]
            categories = encoder.classes_
            data = []
            
            for class_idx, class_name in enumerate(class_names):
                # For each class, get the log prob for each value of this feature
                log_probs = state.model.feature_log_prob_[class_idx, idx, :]
                data.append(log_probs)
                
            data = np.array(data)
            ax = axes[idx]
            
            # Create heatmap
            sns.heatmap(
                data, 
                annot=True, 
                fmt=".2f", 
                cmap="Blues", 
                xticklabels=categories, 
                yticklabels=class_names, 
                ax=ax
            )
            
            ax.set_title(f'Log Probabilities for Feature: {feature}')
            ax.set_xlabel('Feature Value')
            ax.set_ylabel('Class')
            
        plt.tight_layout()
        
        # Save to buffer
        buf = io.BytesIO()
        plt.savefig(buf, format='png', dpi=150)
        plt.close(fig)
        buf.seek(0)
        
        # Create response with cache control headers
        response = StreamingResponse(buf, media_type="image/png")
        response.headers["Cache-Control"] = "max-age=3600"  # Cache for 1 hour
        return response
    except Exception as e:
        raise HTTPException(
            status_code=status.HTTP_500_INTERNAL_SERVER_ERROR, 
            detail=f"Failed to generate feature log probability plot: {str(e)}"
        )

if __name__ == "__main__":
    import uvicorn
    import os
    
    # Get port from environment variable or default to 7860 (for HF Spaces)
    port = int(os.environ.get("PORT", 7860))
    
    # Configure logging for better visibility
    log_config = uvicorn.config.LOGGING_CONFIG
    log_config["formatters"]["access"]["fmt"] = "%(asctime)s - %(levelname)s - %(message)s"
    log_config["formatters"]["default"]["fmt"] = "%(asctime)s - %(levelname)s - %(message)s"
    
    uvicorn.run(
        "app:app", 
        host="0.0.0.0", 
        port=port, 
        log_level="info",
        reload=True,
        log_config=log_config
    )