NumPy

One type, one block of memory — and why everything else follows from that.

zeros, ones, arange, linspace, eye, random — pick a shape, then a filling rule.

Fixed-width boxes — and every surprise that follows from them.

One number per axis that explains reshape, transpose and every free operation.

One range per axis — and why an integer removes an axis while a slice keeps it.

Slicing doesn’t copy — and that will edit your data behind your back exactly once.

Selecting by condition — and why the answer comes back flat.

Indexing with a list of positions — and why two lists zip instead of crossing.

An if-statement over a whole array, without losing its shape.

The memory never moves — only the rule for walking it.

newaxis, squeeze, moveaxis — the tools for making shapes line up.

concatenate extends an axis; stack creates one. Everything else is shorthand.

How NumPy makes mismatched shapes work — without copying anything.

The loop doesn’t disappear — it moves somewhere much faster.

The axis you name is the axis that disappears.

The reductions that don’t reduce — cumsum, diff, and their friends.

argsort is the one that matters — it returns the order, so you can apply it anywhere.

Distinct values, frequency tables, and the vectorised version of `in`.

Why one gap poisons a whole sum, and why `== np.nan` never finds anything.

Seed it, or you can’t tell a real change from noise.

mean vs median, spread, percentiles — and when the average lies to you.

@ is not *, and that one character is most of machine learning.

meshgrid, mgrid, ogrid — how you compute a function of position without a loop.

Kill the loops, stop copying, reuse buffers — in that order.

Binary keeps everything; text loses your dtypes. Know which you want.