Foundations — SciPy — overview of submodules

Below we build every symbol and word the parent note leans on, starting from absolutely nothing. Read top to bottom: each idea is used only after it is drawn.

0. The array — the raw material

Before any algorithm, there is the thing the algorithm works on.

Why the topic needs it: every SciPy function takes arrays in and gives arrays out. When the parent writes A = np.array([[3,1],[1,2]]), that is a 2×2 array — 2 rows, 2 columns. b = np.array([9,8]) is a 1-D array of length 2. If you cannot picture these grids, nothing downstream makes sense.

1. The function $f(x)$ — a machine that turns a number into a number

The parent writes things like lambda x: x**2 and $f(x)=(x-3{)}^2$. What is $f$?

• $f(x)=x^2$ fed $3$ returns $9$. Fed $-2$ returns $4$.
• $f(x)=(x-3{)}^2$ fed $3$ returns $0$ — this will matter.

The picture: plot every pair $(x,f(x))$ as a point and join them → you get a curve. The curve is the function drawn.

Why the topic needs it: quad integrates a function, minimize finds the low point of a function. The function is the object every algorithm chews on.

2. Area under a curve — the symbol ${\int }_a^{b}f(x)dx$

The parent's first derivation is integration. Every piece of that scary symbol has a plain meaning.

Why a sum of rectangles? A rectangle's area is height × width. We can't measure a curved region directly, so we chop it into many thin rectangles, add them, and shrink the width toward zero. That shrinking is why we write $dx$ instead of a fixed width $h$.

Why the topic needs it: scipy.integrate.quad computes this area adaptively — but to know what "the answer" should be, you must know the area idea.

3. Slope and the derivative $\frac{df}{dx}$ — how steep is the curve?

The parent's second derivation minimizes a function using $\frac{df}{dx}=0$. What is that fraction?

Reading $\frac{df}{dx}$: "$d$" again means a tiny change. $df$ = tiny change in output; $dx$ = tiny change in input. Their ratio is rise-over-run in the limit of a tiny run — exactly the slope.

Why the topic needs it: scipy.optimize.minimize walks downhill — it follows the slope down until the slope flattens. See Optimization — gradient descent.

4. The system $Ax=b$ — many equations at once

The parent solves $3x+y=9,x+2y=8$ and writes it as $Ax=b$. Decode each letter.

For the parent's example: $A=\left[\begin{array}{cc}3& 1\\ 1& 2\end{array}\right],x=\left[\begin{array}{c}x\\ y\end{array}\right],b=\left[\begin{array}{c}9\\ 8\end{array}\right]$

How the shorthand unpacks: multiplying a matrix row by the unknown column means "multiply pairwise and add". Row 1 gives $3x+1y$; set equal to $b$'s first entry $9$ → $3x+y=9$. Row 2 gives $x+2y=8$. The two equations are back. See Linear algebra — solving Ax=b.

Why the topic needs it: scipy.linalg.solve finds this $x$ for you (via LU decomposition), and it is the flagship example of the whole parent note.

5. The "mostly-zero" matrix — why $\text{scipy.sparse}$ exists

The picture / the numbers: a $10^6\times 10^6$ grid has $10^{12}$ cells. Each 64-bit float takes 8 bytes, so a dense store needs $8\times 10^{12}$ bytes = 8 terabytes — the parent's exact warning. If only a few entries per row are nonzero, Sparse matrices shrinks this to gigabytes.

6. Frequencies and the FFT — one more submodule word

You don't need the full mathematics here — only to recognise the word when you meet scipy.fft in the submodule map.

7. Probability distribution — for $\text{scipy.stats}$

How these foundations feed the topic

Read it upward: everything rests on the array; functions and matrices grow from it; area and slope grow from functions; and each SciPy submodule is a machine sitting on top of exactly one of these ideas.

Equipment checklist

Cover the right side and test yourself. If any answer surprises you, re-read that section before tackling the parent note.