Day 1: Python Foundations for Scientific Computing 🤖

Welcome to Day 1 of 100 Days of AI!

Today, we’ll revisit Python — not as beginners, but as scientific programmers. Before diving into Machine Learning or Deep Learning, it’s crucial to understand how numerical computation works under the hood.

🔹 Why Scientific Computing in Python?

AI/ML involves handling large datasets and performing millions of mathematical operations (matrix multiplications, vector dot products, optimisation ).

👉 Standard Python (for loops, lists) is too slow for this.
👉 Libraries like NumPy make Python nearly as fast as C because:

• They store data in contiguous memory blocks (unlike Python lists).

• Operations are vectorized (batch operations at once instead of looping).

• They use optimized C/Fortran backends under the hood (BLAS, LAPACK).

🔹 Step 1: NumPy Internals

how Python lists vs NumPy arrays differ.

OUTPUT:

Memory Layout:

🔹 Step 2: Vectorization vs Loops

Example Problem: Add two arrays of size 1 million.

Using Python Loop:

OUTPUT:

Using NumPy Vectorization:

OUTPUT :

NumPy will be ~100x faster. That’s the power of vectorization!

🔹 Step 3: Performance Profiling

Sometimes your code is slow, and you don’t know why. That’s where profiling tools help

OUTPUT:

#TIMEIT automatically runs multiple times to give you average execution time.

Using CProfile :

👉 This shows a function-level breakdown of where time is spent.

🔹 Step 4: Broadcasting (Bonus Concept)

NumPy doesn’t just do element-wise operations — it can automatically expand arrays when shapes don’t match.

👉 NumPy “broadcasted” scalar 5 across the vector [1,2,3] This avoids writing loops.

✅ Takeaways from Day 1

• NumPy arrays are faster & memory-efficient than Python lists.

• Vectorization replaces loops with bulk operations → huge performance gains.

• Profiling tools (Time_it, Cprofile) help detect bottlenecks.

• These optimizations are why AI frameworks (TensorFlow, PyTorch, JAX) are built on top of NumPy-like operations.