NumPy Data Types (dtypes)

What is a dtype?

A dtype (data type) tells NumPy what kind of values an array contains, such as:

• integers (int32, int64)
• floats (float32, float64)
• booleans (bool)
• strings (<U...) and bytes (|S...)

Because NumPy uses a single dtype for the entire array, it can store values efficiently and run fast computations. A dtype name is really just a hint about memory layout: a type name (int, float) followed by the number of bits per element. A standard double-precision float takes 8 bytes (64 bits) — hence float64.

diagram How NumPy infers a dtype mermaid

  flowchart TD
  A["np.array(data)"] --> B{"All values
the same kind?"}
  B -- "All ints" --> C["dtype: int64"]
  B -- "Any float present" --> D["dtype: float64"]
  B -- "Any string present" --> E["dtype: string / object"]
  B -- "Explicit dtype= given" --> F["Use that dtype
(cast or error)"]

numpy.array walks the input data once and picks the narrowest common dtype that can hold every value.

Checking dtype

check

import numpy as np
 
arr = np.array([1, 2, 3])
print(arr.dtype)

Common numeric dtypes

Integers

ints

a = np.array([1, 2, 3], dtype=np.int32)
b = np.array([1, 2, 3], dtype=np.int64)
print(a.dtype, b.dtype)

Floats

floats

a = np.array([1.5, 2.0, 3.25], dtype=np.float32)
b = np.array([1.5, 2.0, 3.25], dtype=np.float64)
print(a.dtype, b.dtype)

Note

There are both signed and unsigned integer types. A signed integer (like int8) can represent negative and positive numbers (-128 to 127). An unsigned integer (uint8) only represents 0 and up (0 to 255) — the same 256 possible values, just shifted.

Memory usage and dtype

Smaller dtypes use less memory. Every element in a float32 array takes 4 bytes; every element in a float64 array takes 8 bytes — double the space for the same number of values.

memory

import numpy as np
 
arr32 = np.ones(1_000_000, dtype=np.float32)
arr64 = np.ones(1_000_000, dtype=np.float64)
 
print("float32 bytes:", arr32.nbytes)
print("float64 bytes:", arr64.nbytes)

sketch Memory layout: int8 vs int64 p5.js Pause Restart

Each block is one byte. A single int64 element takes as much memory as eight int8 elements.

Type conversion

Using .astype()

astype

import numpy as np
 
arr = np.array([1, 2, 3])
arr_f = arr.astype(np.float64)
print(arr_f, arr_f.dtype)

Note

.astype() always creates a new array (a copy), even if you cast to the same dtype it already has.

Safe conversion (avoid overflow)

Converting large values into a smaller dtype can overflow.

overflow

import numpy as np
 
arr = np.array([300], dtype=np.int16)
print(arr.astype(np.uint8))  # wraps around in many cases

Dtype pitfalls in data analytics

1) Missing values

NumPy numeric arrays can’t store NaN in integer dtype.

nan-int

import numpy as np
 
# This will upcast to float automatically because of np.nan
arr = np.array([1, 2, np.nan])
print(arr)
print(arr.dtype)

2) Mixed types

If you mix strings and numbers, dtype may become object or strings.

mixed

import numpy as np
 
arr = np.array([1, "two", 3])
print(arr)
print(arr.dtype)

Arrays with dtype=object are slower for numerical operations.

From the book

Be careful with NumPy’s fixed-size string type (numpy.string_ / "S10"). It truncates input silently if a string is longer than the declared size — no error, no warning. Pandas handles non-numeric data far more intuitively than raw NumPy.

Next

Continue to: Indexing and Slicing Arrays to learn how to select, filter, and extract parts of arrays.

🧪 Try It Yourself

Exercise 1 – Check and Set dtype

pch.exerciseTag pch.exerciseTitle

Exercise 2 – Convert with astype()

pch.exerciseTag pch.exerciseTitle

Exercise 3 – Memory Cost of a dtype

pch.exerciseTag pch.exerciseTitle