pages/KNN.py

import streamlit as st

st.set_page_config(page_title="KNN Algorithm", page_icon="📍", layout="wide")


# Page Title
st.markdown("<h1>K-Nearest Neighbors (KNN)</h1>", unsafe_allow_html=True)

# Introduction
st.markdown("### 🧠 What is KNN?")
st.markdown("""
K-Nearest Neighbors (KNN) is a simple, powerful **supervised learning algorithm** used for both **classification** and **regression** tasks.

- It doesn't build a model during training.
- Instead, it stores the data and makes predictions based on the **K nearest data points** during inference.
""")

# How It Works
st.markdown("### ⚙️ How KNN Works")
with st.expander("Training Phase"):
    st.markdown("""
    - KNN does not perform any actual training.
    - It memorizes the entire dataset and waits for a test point to arrive.
    """)

with st.expander("Prediction - Classification"):
    st.markdown("""
    1. Select a value for **K**.
    2. Calculate distances from the new point to all training points.
    3. Pick the **K closest** data points.
    4. Use **majority voting** to determine the class.
    """)

with st.expander("Prediction - Regression"):
    st.markdown("""
    1. Choose **K**.
    2. Find the **K nearest** neighbors.
    3. Predict the output by:
        - **Averaging** the target values (Standard)
        - **Weighted average**, giving more weight to closer points
    """)

# Distance Metrics
st.markdown("### 📏 Distance Metrics")
st.markdown("""
- **Euclidean**: Most common (straight-line distance)
- **Manhattan**: Grid-like path distance
- **Minkowski**: Generalized form

These metrics help determine how “close” two points are.
""")

# Overfitting vs Underfitting
st.markdown("### ⚖️ Underfitting, Overfitting & Best Fit")
st.markdown("""
- **Low K (e.g., 1)**: Model becomes too sensitive → Overfitting  
- **High K (e.g., 20+)**: Model becomes too generalized → Underfitting  
- **Best Fit**: Use cross-validation to find a balanced K.
""")

# Cross-Validation
st.markdown("### 🔁 Cross-Validation for K")
st.markdown("""
To choose the best `K`, we split the dataset and check performance on unseen data:

- **Training Error**: Model accuracy on training data.
- **CV Error**: Model accuracy on validation data.
- Goal: Find K that minimizes both errors.
""")

# Hyperparameters
st.markdown("### 🛠️ KNN Hyperparameters")
st.markdown("""
- `k`: Number of neighbors (main parameter)
- `weights`: `'uniform'` vs `'distance'`
- `metric`: Distance metric (`euclidean`, `manhattan`, etc.)
- `n_jobs`: Parallel processing for faster computation
""")

# Feature Scaling
st.markdown("### 📊 Feature Scaling")
st.markdown("""
Distance-based algorithms require scaling:
- **Normalization**: Rescales features to [0, 1]
- **Standardization**: Transforms data to mean 0, std dev 1

👉 Apply scaling **before** using KNN!
""")

# Weighted KNN
st.markdown("### 🧮 Weighted KNN")
st.markdown("""
In **Weighted KNN**, closer neighbors contribute more to the prediction.

Why? Because closer points are usually more relevant.
""")

# Decision Regions
st.markdown("### 🗺️ Decision Boundaries in KNN")
st.markdown("""
KNN draws boundaries between different classes:
- `K=1`: Very jagged (overfits)
- `K` increases: Boundaries smooth out (less overfitting)

Visualization helps you understand this behavior better.
""")

# Hyperparameter Search
st.markdown("### 🔍 Hyperparameter Tuning Methods")
st.markdown("""
- **Grid Search**: Try every possible combination (exhaustive)
- **Random Search**: Randomly tries values
- **Bayesian Optimization**: Learns from results to choose next best combination
""")

# Final Note
st.markdown("### ✅ Summary")
st.markdown("""
KNN is intuitive, requires no training, but is sensitive to:
- Feature scaling
- Choice of K
- Noise in data

Use **cross-validation**, **scaling**, and **hyperparameter tuning** to get the best results.
""")