[Kuenzang Dorji] - Fab Futures - Data Science

Home About

Week 4: Machine Learning(28 November 2025)

Lession about

This section covers machine learning fundamentals focusing on neural networks, which use layered neurons with activation functions (sigmoid, tanh, ReLU) to process weighted inputs. Training involves backpropagation with gradient descent variants (SGD, ADAM, L-BFGS) and techniques to prevent overfitting (dropout, regularization, early stopping). Key concepts include loss functions (MSE, cross-entropy), network architectures (deep vs. wide), and practical considerations like vanishing gradients and model optimization (pruning, quantization). It also references major frameworks (PyTorch, TensorFlow, Jax), model repositories (Hugging Face), and specialized applications like large language models and edge computing deployments.

Assignments: We are asked to Fit a machine learning model to our datasets

import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
from sklearn.preprocessing import PolynomialFeatures
from sklearn.linear_model import LinearRegression

# 1. Load dataset
df = pd.read_csv("datasets/viii_2023.csv")

# Clean column names
df.columns = df.columns.str.strip().str.replace('\n', '')

# Remove completely empty rows
df = df.dropna(how='all').reset_index(drop=True)

# 2. Convert numeric columns
df_numeric = df.copy()
for col in df_numeric.columns:
    if col.lower() != 'name':
        df_numeric[col] = pd.to_numeric(
            df_numeric[col].astype(str).str.replace(r'[^0-9.]', '', regex=True),
            errors='coerce'
        )

# Get numeric columns
numeric_cols = df_numeric.select_dtypes(include=[np.number]).columns.tolist()
print("Numeric subject columns detected:", numeric_cols)

if len(numeric_cols) == 0:
    raise ValueError("❌ No numeric columns could be converted! Check your CSV content.")

# 3. ML Polynomial fit & plot
plt.figure(figsize=(14, 8))
x = np.arange(len(df_numeric))

for subject in numeric_cols:
    y = df_numeric[subject].values

    # Skip subject if all values are NaN
    if np.all(np.isnan(y)):
        print(f"⚠ Skipping {subject}: all values are NaN")
        continue

    # Remove NaNs for fitting
    x_clean = x[~np.isnan(y)]
    y_clean = y[~np.isnan(y)]

    # Polynomial regression (degree 2)
    poly = PolynomialFeatures(degree=2)
    X_poly = poly.fit_transform(x_clean.reshape(-1, 1))
    model = LinearRegression().fit(X_poly, y_clean)
    y_pred = model.predict(X_poly)

    # Plot actual & fitted
    plt.plot(x_clean, y_clean, marker='o', label=f"{subject} — Actual")
    plt.plot(x_clean, y_pred, linestyle='--', label=f"{subject} — Model Fit")

# Graph settings
plt.title("Students' Marks with ML Model Fit")
plt.xlabel("Student Index")
plt.ylabel("Marks")
plt.grid(True)
plt.legend()
plt.tight_layout()
plt.show()

Numeric subject columns detected: ['Unnamed: 0', 'Dzongkha', 'English', 'Geography', 'History', 'ICT', 'Maths', 'Science', 'Unnamed: 8']
⚠ Skipping Unnamed: 0: all values are NaN
⚠ Skipping Unnamed: 8: all values are NaN

Explanation

This code performs polynomial regression on student marks data to model academic performance trends. It cleans the dataset by converting columns to numeric values and removing empty rows, then fits second-degree polynomial models to each subject using scikit-learn's PolynomialFeatures and LinearRegression. The visualization plots both actual marks (dots) and ML-predicted trend lines (dashed) for all subjects, showing how polynomial curves capture performance patterns across the student cohort while handling missing data appropriately.