About Projects Courses Mind Map Notes Roadmap Blog Services

Loading…

Newsletter

New tutorials, notes & project breakdowns

Practical ML, GenAI and full-stack lessons, straight to your inbox. No spam, unsubscribe anytime.

Munna Kumar, also known as Code With Munna (CodeWithMunna / CodeWithMunnaX). Data science, ML and GenAI engineering, built for production and taught for free on YouTube.

Explore

About
Projects
Skills
Blog
Tech Setup

Learn

Free Courses
Python Notes
NumPy Notes
Pandas Notes
SQL Notes
YouTube Channel

Work With Me

Services
How I Work
Book a Free Call
Contact

© 2026 Munna Kumar · Made in India

Loading notes

NotesNumPy

🔢 Free notes

NumPy Notes

NumPy from scratch to advanced — arrays, indexing, broadcasting, statistics, random data, linear algebra and image processing, with a live Python playground on every topic. Written by Munna Kumar (Code With Munna) — 77 topics in order, each with examples, key points, and 22 with a video lesson.

Start with What Is NumPy & Why It Exists

Getting Started

1What Is NumPy & Why It ExistsWhat NumPy is, what it is used for, and the exact problems with Python lists that made NumPy necessary for numeric work.
2ndarray vs Python ListHow a NumPy ndarray differs from a Python list — homogeneous types, contiguous memory, vectorized maths, fixed size — with a full comparison table.
3Installing & Importing NumPyHow to install NumPy with pip, import it with the standard alias np, check the installed version, and fix the most common setup errors.
4Your First ArrayCreate your first NumPy array from a Python list, print it, inspect its type, index and slice it, change values, and convert it back to a list.
5Performance: Loops vs VectorizationMeasure the speed difference between Python loops and NumPy vectorized operations with time and timeit, and understand why NumPy wins.

Creating Arrays

6np.array() from ListsCreate 1D, 2D and 3D arrays from Python lists with np.array(), understand how NumPy picks a dtype (type promotion), and avoid the ragged-list error.
7zeros, ones, full & emptyCreate arrays of a given shape filled with 0, 1, any value, or uninitialised memory — plus the *_like versions that copy another array's shape and dtype.
8np.arange()Generate evenly spaced values with a fixed step using np.arange — how it compares with Python's range, negative and float steps, and the float-step pitfall.
9linspace & logspaceCreate a fixed number of evenly spaced values with np.linspace (num, endpoint, retstep) and log-spaced values with np.logspace.
10eye, identity & diagCreate identity matrices with np.eye and np.identity, build diagonal matrices with np.diag, and extract a matrix's diagonal.
11Data Types (dtype)Understand NumPy dtypes (int32, int64, float64, bool, str …), set them with dtype=, convert with astype, and avoid overflow and precision traps.
12Creation Functions Cheat-SheetOne-page reference of every array-creation function — what it does, its key parameters, default dtype and a real output — plus how to choose the right one.

Array Anatomy

13Array AttributesRead an array's metadata — shape, ndim, size, dtype, itemsize and nbytes — and understand how they relate to each other.
141D vs 2D vs 3D ArraysUnderstand dimensions with cricket data: one player's scores (1D), the whole team (2D), and two seasons (3D) — how to build, read and index each.
15Understanding AxesWhat axis 0, axis 1 and axis 2 mean, how sum(axis=…) collapses a dimension, and how to predict the result shape — with ASCII diagrams.

Indexing & Slicing

161D IndexingRead single elements of a 1D NumPy array with positive and negative indexes, and understand what an index really is.
171D SlicingTake a range of elements with start:stop:step — first N, last N, every k-th element and reversing an array.
182D Indexing: a[row, col]Read single cells and whole rows of a 2D array with arr[row, col], and see why arr[row][col] is not the same thing.
192D Slicing: Rows, Columns & Sub-matricesSlice each axis of a 2D array separately to get columns, groups of rows, sub-tables and stepped grids.
203D Indexing & SlicingNavigate a 3D array (seasons × players × matches) with arr[i, j, k], and slice any combination of axes.
21Boolean Indexing & MasksFilter arrays by a condition — build True/False masks, combine them with & | ~ and use them to select or count values.
22Fancy (Integer-Array) IndexingPick any positions, in any order, with a list or array of indexes — including rows, columns and (row, col) pairs.
23np.where()Use np.where to find the positions where a condition is true (1-argument form) or to build a new array with vectorized if/else (3-argument form).
24Modifying Arrays via IndexingChange values in place by assigning to an index, slice, mask or index list — and why changing a slice changes the original.

Array Operations

25Vectorized ArithmeticDo maths on whole arrays without loops: array with scalar, array with array, the element-wise rule and the same-shape requirement.
26Universal Functions (ufuncs)Use NumPy's fast element-wise maths functions — sqrt, exp, log, abs, round/floor/ceil, power, mod, maximum, sin/cos — on whole arrays.
27Comparison & Logical OperationsCompare arrays element by element (==, !=, >, <), combine results with logical_and/or/not, and summarise with any, all and array_equal.
28Errors & Special Values (nan, inf)What NumPy does on division by zero, what nan and inf are, how to detect them with isnan/isinf, control warnings with seterr/errstate, and compute a safe strike rate with np.where.

Statistics & Aggregation

29sum, mean, median, min & maxCollapse a whole array of scores into one number with sum, mean, median, min and max — and learn the function form vs the method form.
30The axis ParameterUse axis=0 and axis=1 to get per-match vs per-player totals from a 2D scores table, plus keepdims and negative axes.
31std & var: Measuring ConsistencyMeasure how spread out scores are with standard deviation and variance — population vs sample, the ddof parameter, and finding the most consistent player.
32argmin & argmaxFind WHERE the smallest or largest value is — which match was the best, which player scored most — with argmin, argmax and unravel_index.
33Percentiles & QuantilesDescribe where a score stands within the data with np.percentile and np.quantile — quartiles, the IQR, and the method parameter.
34cumsum & cumprodBuild running totals and running products — a batter's running score across the season, the team's scoreboard over by over, and compounding growth.
35NaN-safe FunctionsHandle missing data (np.nan) correctly — why one NaN poisons sum and mean, and how nansum, nanmean, nanmax and friends ignore it.
36Describe Your DataPut every statistic together into a mini describe() function — count, mean, std, min, quartiles, max — and print a per-player summary table using only NumPy.

Reshaping & Views

37reshape()Rearrange the same data into a new shape — the total-elements rule, the -1 placeholder, splitting a season into halves, and C vs F order.
38ravel vs flattenTurn any multi-dimensional array into 1D with ravel() or flatten() — and understand why one returns a view and the other always a copy.
39Transpose, .T & swapaxesFlip rows and columns with .T and np.transpose, reorder axes of 3D data, and swap two axes with swapaxes — plus why you would transpose data at all.
40newaxis, expand_dims & squeezeAdd a length-1 dimension with np.newaxis / None or np.expand_dims, and remove length-1 dimensions with np.squeeze — the key to making shapes broadcast.
41Copy vs ViewKnow exactly when two arrays share data — assignment, views (slices, reshape, .T) and copies (.copy(), fancy indexing) — and check it with np.shares_memory and .base.

Broadcasting

42What is Broadcasting?Why NumPy lets arrays of different shapes work together, what 'virtual stretching' means, and the simplest case: scalar broadcasting.
43The Broadcasting RulesThe exact step-by-step algorithm NumPy uses to decide whether two shapes are compatible, how to compute the result shape, and a full compatibility table.
441D with 2D: Per-ColumnRule 2 from class: a 1D array whose length matches the number of columns broadcasts across every row — one value per match (column).
45Per-Row with reshape(-1, 1)Rule 3 — 'the tricky one': apply one value per ROW by turning a 1D array into a column vector with reshape(-1, 1) or np.newaxis.
46Broadcasting Errors & FixesHow to read 'operands could not be broadcast together', the in-place error, the silent (n, n) surprise, and a checklist of fixes.

Combining & Splitting

47np.concatenateJoin two or more arrays end to end along an EXISTING axis — more rows with axis=0, more columns with axis=1 — and the shape rules that decide when it works.
48vstack, hstack & column_stackConvenience wrappers around concatenate: vstack (rows on top of each other), hstack (side by side) and column_stack (1D arrays become columns).
49np.stack: Joining on a New Axisnp.stack combines same-shape arrays along a brand-new axis (dimensions go up by one) — how it differs from concatenate, and what the axis argument controls.
50Splitting ArraysThe reverse of joining: split an array into pieces with np.split, np.array_split (uneven pieces allowed), np.hsplit and np.vsplit.
51append, insert & deleteAdd values at the end (np.append), at a position (np.insert) or remove them (np.delete) — why they always return new arrays and why calling them in a loop is slow.

Sorting & Searching

52Sorting ArraysSort values with np.sort (returns a copy) or ndarray.sort (in place), sort in descending order, and sort 2D arrays along rows, columns or the whole array.
53argsort & Rankingargsort returns the INDICES that would sort an array — the key to ranking players, sorting one array by another, and finding positions with argmax/argmin.
54np.unique: Distinct Values & CountsFind the distinct values of an array (always sorted), count how often each appears, and locate first occurrences with return_counts, return_index and return_inverse.
55Searching: searchsorted, nonzero, argwhereFind WHERE values are: insertion points in sorted data with searchsorted, positions of True/non-zero elements with nonzero, np.where and argwhere.
56Set OperationsCompare two arrays like mathematical sets: intersect1d (common), union1d (all), setdiff1d (only in first), setxor1d, and isin for membership masks.
57Top-N PatternsGet the top or bottom N values and their owners with argsort[::-1][:n], per-row top-N on 2D arrays, and the faster argpartition for large data.

Random Numbers

58Randomness & SeedsWhat 'pseudo-random' really means, how np.random.seed() makes random results reproducible, and why every data scientist sets a seed.
59rand, randn, randint & randomThe four workhorse functions of np.random — uniform floats, standard-normal floats, random integers — and exactly when to use each.
60choice, shuffle & permutationPicking random items from an existing array (with or without replacement, with weights) and randomly reordering arrays in place or as a copy.
61Distributions: normal, uniform, binomialGenerate realistic data from the three most useful probability distributions — normal (bell curve), uniform (flat) and binomial (count of successes).
62The Generator API (default_rng)The modern, recommended way to make random numbers: rng = np.random.default_rng(seed) with rng.integers, rng.random, rng.normal and rng.choice.
63Generating Synthetic IPL DataCombine random functions into a realistic fake dataset — runs, balls, strike rates and match records — for testing, prototyping and simulation.

Linear Algebra

64Element-wise * vs Matrix @The single most important distinction in NumPy linear algebra: * multiplies position by position, while np.dot, @ and np.matmul perform matrix multiplication.
65How Matrix Multiplication WorksStep-by-step mechanics of matrix multiplication: the (m, n) @ (n, p) → (m, p) shape rule and how every cell is a row-times-column dot product.
66Weighted Scoring with @A practical use of matrix multiplication: compute fantasy-cricket points for every player in one line by multiplying a stats matrix with a weights vector.
67Dot Product of VectorsThe dot product of two vectors — what it computes, how to write it (np.dot, @, sum of products), and what it tells you about similarity.
68Identity, Inverse & DeterminantThe identity matrix (np.eye), the inverse of a matrix (np.linalg.inv) and the determinant (np.linalg.det) — what each means and how they connect.
69Solving Linear EquationsTurn a set of simultaneous equations into matrix form Ax = b and solve it in one call with np.linalg.solve.
70Vector Norm (np.linalg.norm)Measure the length of a vector and the distance between two vectors with np.linalg.norm — L2, L1 and max norms, row-wise norms and normalisation.

Real-World NumPy

71Images as ArraysHow a digital image is just a NumPy array — H×W for grayscale, H×W×3 for colour — with uint8 pixel values, channels, and weighted grayscale conversion.
72Crop, Flip & Rotate ImagesGeometric image edits with pure NumPy: cropping with slicing, flipping with [::-1] or np.flip, and rotating with np.rot90.
73Brightness, Invert & ThresholdPixel-value edits with vectorised maths: brightness and contrast with np.clip, avoiding uint8 overflow, inverting (negatives), normalising and thresholding to black-and-white.
74Saving & Loading (.npy / .npz)Store arrays in NumPy's own binary formats with np.save, np.savez and np.savez_compressed, and read them back with np.load — exact dtype and shape preserved.
75Text Files: savetxt, loadtxt, genfromtxtWrite arrays to CSV/text with np.savetxt and read them with np.loadtxt (clean numeric data) or np.genfromtxt (missing values, headers, mixed columns).
76Vectorization & PerformanceWrite fast NumPy code: replace Python loops with ufuncs, np.where and broadcasting, avoid np.append in loops, preallocate, and use in-place operations.
77NumPy Cheat SheetThe whole NumPy course on one page — creation, attributes, indexing, statistics, reshaping, combining, broadcasting, random, linear algebra, file I/O and performance — in quick-reference tables.

Looking for another subject?

Python Notes Pandas Notes SQL Notes