Hey,
I'm a web developer - building UIs with React, TypeScript, Tailwind, and modern web tools. A few months ago, I decided to step into Machine Learning.
No prior ML background. Just curiosity and basic Python knowledge.I started Andrew Ng’s Machine Learning Specialization and created my own study plan. Here’s what the first 30 days looked like — the concepts that clicked, the mistakes I made, and what surprised me most.
Week 1: Building the Math Foundation
I started with prerequisites because they’re essential:
- Linear Algebra (vectors, matrices, eigenvalues)
- Calculus (derivatives, partial derivatives, gradients)
- Probability & Statistics
Concept: Why Gradients Matter
Gradient descent is the engine behind almost all modern ML.
Imagine you’re trying to reach the bottom of a valley in the dark. The gradient tells you the direction and steepness of the slope. You take a small step downhill.
(w = w - learning_rate * gradient)
Repeat until you reach the minimum.
This simple idea powers neural networks, linear regression, and more.
Weeks 2–3: Supervised Learning (The Real Fun Begins)
I dove into regression and classification:
- Linear & multiple linear regression
- Logistic regression for classification
- Cost functions, gradient descent, regularization, feature scaling, softmax
I built small projects in Colab:
- House price predictor (linear regression)
- Spam email classifier (logistic regression)
Concept: Regularization
Regularization is like putting guardrails on your model. Without it, the model can memorize noise in the training data (overfitting). With L2 regularization, we penalize large weights, helping the model generalize better.
Weeks 4–5: Neural Networks & Tree Ensembles
This is where things got exciting:
- Built my first neural networks in TensorFlow (forward propagation + backpropagation)
- Learned decision trees, Random Forests, and XGBoost
Concept: Backpropagation
Forward propagation makes a prediction.
Backpropagation figures out why the prediction was wrong and updates every weight accordingly using the chain rule. It’s like tracing a bug through a chain of React components — but for thousands of parameters.
What Surprised Me Most
- The math finally made sense when I saw it used in real models.
- Building small projects > passive watching.
- ML and web development are a powerful combination — I can already imagine turning these models into real web features.
What’s Next?
I’m continuing the plan:
- More neural network practice
- Full mini-projects
- Course 3 (Unsupervised learning)
I’ll be sharing regular updates here — code, lessons, and notebooks. If you’re also learning ML as a web developer, drop a comment. I’d love to hear your journey too!
