Table of Contents
NumPy Data Types
NumPy’s power stems from its efficient ndarray (N-dimensional array) object. Unlike Python lists, NumPy arrays are homogeneous; all elements share the same data type. This homogeneity allows for optimized vectorized operations, significantly boosting performance. NumPy offers a wide variety of data types, many mirroring those in C and Fortran, each with a shorthand code:
| Data Type | Description | NumPy Type Code |
|---|---|---|
| Integer | Signed integer | int8, int16, int32, int64 |
| Unsigned Integer | Unsigned integer | uint8, uint16, uint32, uint64 |
| Floating-Point | Single and double precision floating-point | float16, float32, float64 (commonly float) |
| Complex Floating-Point | Single and double precision complex numbers | complex64, complex128 (commonly complex) |
| Boolean | True/False values | bool |
| String | Fixed-length strings | str (or string_ with length, e.g., string_10) |
| Unicode | Fixed-length Unicode strings | unicode_ (e.g., unicode_10) |
| Object | Arbitrary Python objects | object |
Specifying Data Types:
When creating an array, you can explicitly set the data type using the dtype argument:
import numpy as np
arr_int = np.array([1, 2, 3, 4], dtype=np.int32)
print(arr_int.dtype) # Output: int32
arr_float = np.array([1.1, 2.2, 3.3], dtype=np.float64)
print(arr_float.dtype) # Output: float64
arr_mixed = np.array([1, 2.5, 3]) # dtype will be upcast to float64
print(arr_mixed.dtype) # Output: float64
Careful data type selection is crucial for memory efficiency and performance. Smaller types (e.g., int32) save memory but risk overflow if values exceed their range.
Data Type Conversion
Converting between data types is common. NumPy provides several methods:
1. Using astype():
astype() creates a copy with the new data type:
arr_int = np.array([1, 2, 3], dtype=np.int32)
arr_float = arr_int.astype(np.float64)
print(arr_float) # Output: [1. 2. 3.]
print(arr_float.dtype) # Output: float64
2. Implicit Type Conversion:
NumPy sometimes implicitly converts types during operations:
arr_int = np.array([1, 2, 3])
arr_float = np.array([1.1, 2.2, 3.3])
result = arr_int + arr_float
print(result) # Output: [2.1 4.2 6.3]
print(result.dtype) # Output: float64
3. Using view() (Use with Caution!):
view() creates a new array sharing the same data but with a different type. Changes to one view affect the other. This is memory-efficient but risky:
arr_int = np.array([1, 2, 3], dtype=np.int32)
arr_view = arr_int.view(np.float32) # Potentially dangerous!
print(arr_view) # Output: [1. 2. 3.]
arr_view[0] = 10.5 # Modifies the original as well!
print(arr_int) # Output: [10 2 3]
Understanding NumPy’s data types and conversion is vital for writing efficient and reliable code. Choose appropriate types, and use view() cautiously, mindful of potential data loss or unexpected behavior.