NumPy & Pandas — Deep Dive

Introduction

NumPy gives you fast, typed n-dimensional arrays. Pandas builds DataFrames on top — labeled rows/columns with SQL-like operations.

Understanding the topic

Core concepts to understand:

• ndarray = fixed dtype, contiguous memory.
• Broadcasting = implicit shape alignment.
• Pandas: read_csv, groupby, merge, pivot.
• Index alignment is automatic in Pandas ops.

Syntax reference

Visual flow / code:

import numpy as np
import pandas as pd

# NumPy
x = np.arange(12).reshape(3, 4)
x.mean(axis=0)         # column means
x[x > 5]               # boolean mask

# Pandas
df = pd.DataFrame({
    "country": ["US","DE","US","DE"],
    "sales": [100, 80, 120, 90],
})
df.groupby("country").sum()
df.pivot_table(values="sales", index="country", aggfunc="mean")
df.merge(other_df, on="country", how="left")

Execution workflow

Real-world use

Pandas powers most analytics, dashboards, and ML feature engineering in Python. For data > RAM, use Polars or DuckDB.

Best practices

• Use vectorized ops, not row loops.
• Set proper dtypes (category, int32) to save memory.
• Chain ops with .pipe() for readability.

Common mistakes

• SettingWithCopyWarning — use .loc or .copy().

Hands-on exercise

Interview preparation — practice these questions:

• Broadcasting — explain.
• groupby internals?
• Pandas vs SQL?

Summary

In summary: NumPy = math; Pandas = tables. Vectorize and dtype-tune for speed.