P2-3.1 Scalar, Vector, and Matrix¶

Section ID: P2-3.1 Version: v2026.09.08

AI data can be represented as a single number, a list of values, or a table with rows and columns. These correspond to a scalar, a vector, and a matrix.

Numbers, Lists, and Tables¶

Numeric Data Representations¶

AI models do not calculate directly on real-world sentences, images, sounds, or tables. Most of the time, they turn them into numbers and handle those numbers in the form of arrays.

1. Sentences are turned through tokens into numeric IDs and vectors.
2. Images are turned into arrays that hold pixel values.
3. Table data is read as a matrix or dataframe with a row-and-column structure.

The shape of a numeric array can be distinguished with the following questions.

1. Is there only one value?
2. Are the values gathered in one line?
3. Are the values arranged into rows and columns?
4. Are several pieces of data bundled together at once?

These questions matter in real code as well. One common reason AI code fails is not misunderstanding the meaning of a value, but misunderstanding its shape.

Scalar: One Number¶

A scalar is one number. For example, all of the following can be treated as scalars.

In AI contexts, scalars appear in many places.

• one temperature value
• one probability
• one loss
• one accuracy
• one learning rate

For example, if the average loss of a model is 0.25, the value itself is a scalar.

A scalar looks small, but it matters. When evaluating a model, we often summarize many calculation results into one number. But one number also compresses a great deal of information. The statement that the loss is 0.25 does not tell us which data the model handled well and which data it got wrong.

Position-Wise Addition and Scalar Multiplication¶

Scalar, vector, and matrix do not differ only in shape. Mathematically, what changes is which calculations are possible.

Because a scalar is one number, we can immediately think of ordinary arithmetic.

Vectors can be added by matching values at the same positions.

If a scalar is multiplied by a vector, the same number is multiplied into each value of the vector.

Matrices can also be added position by position when they have the same shape.

These are less like advanced linear algebra and more like the rule calculate by matching the same positions. This is where shape becomes important. Vectors with different lengths or matrices with different numbers of rows and columns cannot immediately be calculated position by position.

1. If the shape is the same, position-wise calculation is possible.
2. If the shape is different, we first have to decide which values should be combined.

Vector: Ordered Values¶

A vector can be read as an ordered list of values.

Here, \(\mathbf{x}\) is a vector with three values.

For example, suppose one person is represented by three values: age: 30, height: 172, and monthly visits: 5.

Written as a vector, it looks like this.

This vector represents one person’s features as a list of numbers. In machine learning, it is common to gather several features this way and use them as one input.

Order matters in a vector. It must already be fixed whether the first value means age, height, or number of visits.

1
2

[30, 172, 5]
-> [age, height, monthly visits]

If the order changes, the same numbers mean something different.

1
2

[172, 30, 5]
-> [height, age, monthly visits]

So a vector is not merely a bunch of numbers. It is a bundle of numbers where each position carries meaning.

Matrix: Rows and Columns¶

A matrix is a structure in which numbers are arranged into rows and columns.

This matrix has 2 rows and 3 columns. We usually call it a \(2 \times 3\) matrix. A row is one horizontal line, a column is one vertical line, and shape is the count of rows and columns.

In AI data, one row can be read as one data sample, and one column can be read as one feature.

For example, if we represent the data of 3 people through age, height, and number of visits, we get the following.

In this case there are 3 rows and 3 columns. That means the 3 rows are 3 people, and the 3 columns are age, height, and monthly visits.

This perspective is very important when reading a machine learning dataset. In many beginner materials, \(X\) means the full input data, and \(y\) means the target value or label corresponding to each row.

Shape and Calculation Conditions¶

Shape is the form of data. It is the information that tells us whether we are looking at one value, a list of values, or an array with rows and columns.

Create a single number, three values, and a two-row, three-column array in NumPy, then print their shape.

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11

import numpy as np

# scalar, vector, and matrix are three arrays for comparing stored data shape.
scalar = np.array(3)
vector = np.array([1, 2, 3])
matrix = np.array([[1, 2, 3], [4, 5, 6]])

# shape gives the length of each axis.
print(scalar.shape)
print(vector.shape)
print(matrix.shape)

This code checks the shape of the three values. The result is read like scalar (), vector (3,), and matrix (2, 3).

Depending on the library or settings, the appearance of a scalar can differ slightly, but the core remains the same. A scalar is one number, a vector is a list of numbers, and a matrix is a numeric array with rows and columns.

Shape matters because whether a calculation is possible changes with shape. For example, vectors with different lengths are difficult to add position by position.

This expression does not align which values should be added together. In code as well, if the shape does not match, an error can occur or the calculation can proceed in a way you did not intend.

Meaning of Rows and Columns¶

A matrix is a numeric array, but the meaning of those numbers is decided by the problem definition. Even for the same \(3 \times 3\) matrix, rows may mean people and columns may mean features in table data; rows and columns may represent pixel positions in part of an image; rows may mean tokens and columns may mean embedding dimensions in sentence data; and rows may mean samples and columns may mean features in batch data.

So when reading a matrix, do not look only at the shape of the numbers. Check what a row represents, what a column represents, what unit each value has, and whether the order of rows and columns is fixed.

When you encounter phrases such as input matrix, feature matrix, or embedding matrix in AI documents, first confirm the meaning of the rows and columns.

Data Shapes in AI Calculation¶

The same array shapes serve different roles in AI calculation.

• embedding: text or items are represented as vectors
• feature: one data sample is represented as a vector of several values
• batch: several samples are grouped into a matrix or a higher-dimensional array
• loss: several calculation results may be summarized into one scalar
• model parameter: weights can take the form of vectors or matrices
• matrix multiplication: it becomes the basic tool for calculating many inputs and weights at once

Student Count and Feature Count¶

Suppose we have the following data.

Assume that in this matrix, each row means a student and the columns mean [study time, quiz score].

Then this matrix means that the first student has study time 2 and quiz score 8, the second student has study time 4 and quiz score 6, and the third student has study time 5 and quiz score 9.

Here, the data for one student is a vector.

If we gather the data of all three students, it becomes a matrix.

This matrix has shape (3, 2): there are three students, each with two features, study time and quiz score. The second student’s quiz score is 6, in row 2, column 2. Adding one student gives shape (4, 2); adding one feature gives (3, 3).

Checklist¶

• Can you explain a scalar as one number?
• Can you explain a vector as an ordered list of values?
• Can you explain a matrix as a numeric array with rows and columns?
• Can you explain that shape can affect whether a calculation is possible and whether an error occurs?
• Can you explain that the meaning of rows and columns depends on the problem definition?
• Can you read one sample as a vector and several samples as a matrix?
• Can you explain why scalar, vector, and matrix reappear in embedding, feature, batch, loss, and parameters?
• Can you explain that vectors and matrices with the same shape can be added position by position, and that scalar multiplication multiplies the same number into each value?
• When the shape of X, the number of samples, and the number of features become mixed up, can you separate them again through shape intuition?

Sources and References¶

• Marc Peter Deisenroth, A. Aldo Faisal, Cheng Soon Ong, Mathematics for Machine Learning, Cambridge University Press, 2020, checked 2026-07-19.
• Ian Goodfellow, Yoshua Bengio, Aaron Courville, Deep Learning, MIT Press, 2016, checked 2026-07-19.
• Charles R. Harris et al., Array Programming with NumPy, Nature, 2020, checked 2026-07-19.
• NumPy Developers, numpy.ndarray.shape, NumPy User Guide, checked on 2026-07-19. This official reference supports reading shape as array dimension information and checking it in code.
• scikit-learn developers, Glossary of Common Terms and API Elements, scikit-learn User Guide, checked on 2026-07-19. This is a reference for the convention of reading X as an input data matrix and y as the target.