Why Every Full Stack Developer Should Learn NumPy in 2026

import numpy as np
import time

prices = np.random.default_rng(0).uniform(100, 5000, 1_000_000)

start = time.perf_counter()
with_gst_loop = [p * 1.18 for p in prices]
print("loop:", round(time.perf_counter() - start, 3), "s")

start = time.perf_counter()
with_gst = prices * 1.18
print("numpy:", round(time.perf_counter() - start, 3), "s")

import numpy as np

scores = np.array([[70, 80, 90],
                   [60, 75, 85],
                   [88, 92, 79]])
col_means = scores.mean(axis=0)     # shape (3,)
centred = scores - col_means         # broadcast across rows
print(col_means)
print(centred)

# FastAPI endpoint returning order statistics
import numpy as np
from fastapi import FastAPI

app = FastAPI()
ORDER_VALUES = np.array([450, 1200, 830, 2999, 640, 1500, 720])

@app.get("/stats/orders")
def order_stats():
    v = ORDER_VALUES
    return {
        "count": int(v.size),
        "mean": float(v.mean()),
        "p50": float(np.percentile(v, 50)),
        "p90": float(np.percentile(v, 90)),
    }

import numpy as np

def moving_average(values, window=7):
    values = np.asarray(values, dtype=float)
    kernel = np.ones(window) / window
    return np.convolve(values, kernel, mode="valid")

daily = [12, 15, 14, 20, 22, 19, 25, 30, 28, 26]
print(moving_average(daily, 3).round(2))

import numpy as np

def cosine_similarity(query, items):
    query = query / np.linalg.norm(query)
    items = items / np.linalg.norm(items, axis=1, keepdims=True)
    return items @ query

items = np.array([[0.1, 0.9, 0.0],
                  [0.8, 0.1, 0.1],
                  [0.2, 0.7, 0.1]])
query = np.array([0.15, 0.85, 0.05])
print(cosine_similarity(query, items).round(3))

scores = cosine_similarity(query, items)
k = 2
top_k = np.argsort(scores)[::-1][:k]
for rank, idx in enumerate(top_k, start=1):
    print(rank, "item", idx, "score", round(float(scores[idx]), 3))

import numpy as np

raw = np.array([120.0, np.nan, 340.0, -5.0, 99999.0])
clean = np.where(np.isnan(raw), 0, raw)       # replace missing
clean = np.clip(clean, 0, 10000)               # cap extremes
print(clean)

import numpy as np

ages = np.array([17, 22, 35, 41, 58, 63])
bins = [0, 18, 30, 45, 60, 120]
labels = np.array(["<18", "18-29", "30-44", "45-59", "60+"])
groups = labels[np.digitize(ages, bins) - 1]
print(list(zip(ages.tolist(), groups.tolist())))

NumPy quick checks before shipping:
[ ] Array dtype is what I expect (float64, int64)
[ ] Shapes printed once while developing
[ ] No Python loops over large numeric arrays
[ ] NumPy values converted before JSON response
[ ] Random generators use a fixed seed in tests