Data types in NumPy

In this topic, we will become acquainted with NumPy data types. You have already learned some basic data types provided in Python, such as string for text data and float for floating-point numbers. NumPy and Python data types share some common features, but NumPy data also has some unique characteristics. We will discuss the basic NumPy data types and methods for working with them.

Types of data in NumPy

Imagine that you strive to know everything about all data types in NumPy and you wonder how to do it. In this case np.sctypeDict can help you. It prints a dictionary in which all the keys are different data types.

print(np.sctypeDict)
# {'?': <class 'numpy.bool_'>, 0: <class 'numpy.bool_'>, 'byte': <class 'numpy.int8'>, 'b': <class 'numpy.int8'>, 1: <class 'numpy.int8'>, 'ubyte': <class 'numpy.uint8'>, ...}

Of course, such a dictionary is not the most convenient way to explore the data types. The summary of the main types with their brief descriptions is shown below. The most widely-used types are the ones that represent integers and floating numbers, but there are other types, too. It is worth mentioning that NumPy data types are the same as in C programming language because NumPy itself is a wrapper on C, which makes NumPy operations so fast.

1. Integer data

For integer numbers, there are a lot of specific data types in NumPy.

Mind different ranges of numbers. As we have mentioned, NumPy specifies the size of the data: the thing is, one byte (or 8 bits) is required to decode integers from -128 to 127. If we take only non-negative ones, the range is from 0 to 255, respectively. For larger integers, we need two bytes (16 bits), four bytes (32 bits), or eight bytes (64 bits).

The range thresholds are determined by different powers of two. In case of int8, it is $2^{8-1} = 2^7 = 128$. One bit is reserved for the negative / positive sign, so it's 27 rather than 28. Also, one place is taken by zero, so the upper boundary is 128 - 1 = 127. Similarly, for the upper limit for int64, 9 223 372 036 854 775 807 = $2^{64-1} - 1$.

It is important to note that -3, for instance, can have either int8, int16, int32, or int64 data type. On the other hand, 3 can represent any of int and uint formats, it all depends on what you are working with.

2. Float data

For float values, there are the following data types:

It is necessary to clarify the types of precisions. Half-precision floating-point format provides results that are not that accurate. If you want to obtain more decimal spaces, you need to use the single-precision, double-precision, or quadruple-precision formats. It is shown in the example below.

n = 22/7
print(np.float16(n))  # 3.143
print(np.float32(n))  # 3.142857
print(np.float64(n))  # 3.142857142857143

For other purposes, you can use boolean, complex, or string data. bool_ is for boolean data in NumPy, it works the same way as the boolean type in Python. For complex numbers, there are several data types: complex64, complex_ (complex128), and complex256. String data is represented by str_ (standard Python string) or unicode_(Unicode string), which is an alias of str_, and bytes as bytes_ (a sequence of encoded bytes which is ready to be stored in the memory of a program). You can read more about these data types in the official NumPy documentation.

Specifying data types

In NumPy, the np.dtype class stores different objects of the data type. Below there are some sample queries:

print(np.float64)  # <class 'numpy.float64'>
print(np.str_)     # <class 'numpy.str_'>

To find out the data type of an existing array, use array.dtype:

array_1 = np.array([[5, 6], [15, 0]])
array_2 = np.array([[2.5, 5.7], [1.2, 6.9]])
array_3 = np.array([[2, 34], [78.9, 4.5]])
print(array_1.dtype)  # int64
print(array_2.dtype)  # float64
print(array_3.dtype)  # float64

Be aware of the array_3 variable. It contains an array with different types of data — integers and floats. When combining data, NumPy will convert all the data into one type, so that all the integers become floating numbers.

When creating an array, you can specify the type of data in the array with the help of the dtype argument:

array_4 = np.array([[1, 2], [3, 4]], dtype=np.float32)
print(array_4)
# [[1. 2.]
#  [3. 4.]]

array_5 = np.arange(7, dtype=np.float64)
print(array_5)  # [0. 1. 2. 3. 4. 5. 6.]

array_6 = np.arange(7, dtype=np.int64)
print(array_6)  # [0 1 2 3 4 5 6]

Changing data types

To transform one data type into another, use the np.<type> function:

print(np.bool_(1))  # True
print(np.float64(35))  # 35.0

If you want to do it with an array rather than with single values, use the array.astype() function. It copies the array and converts it into the specified type.

array_7 = np.array([[-129, 45], [34, 129]], dtype=np.int64)
print(array_7)
# [[-129   45]
#  [  34  129]]

print(array_7.astype(np.float64))
# [[-129.   45.]
#  [  34.  129.]]

print(array_7.astype(np.str_))
# [['-129' '45']
#  ['34' '129']]

print(array_7.astype(np.bool_))
# [[ True  True]
#  [ True  True]]

In the example above, you can see that we can easily transform our array_7 into float, string, or boolean data. However, if we try to change the type to int8, a problem will arise:

print(array_7.astype(np.int8))
# [[ 127   45]
#  [  34 -127]]

As you can remember, int8 can store integers from -128 to 127 that take up to one byte. When we try to convert into int8 numbers that are not in its range, NumPy will not display a memory error, but instead, it just starts counting from the very beginning. -129 was changed to 127, the first positive integer. Similarly, 129 was changed to the second negative integer.

Conclusion

NumPy provides a great number of data types compared to basic Python data types. In this topic we've learned:

• main types of data in NumPy;

• how to specify data types when you create arrays;

• how to print the data type of an existing array;

• the way of changing data types in arrays.

The next step is practicing your skills!