Being a linear sequence generator, the `numpy.arange()`

function is used to generate a sequence of numbers in linear space with a uniform step size.

This is similar to another function, numpy.linspace(), which also generates a linear sequence with a uniform step size.

Let’s understand how we can use this function to generate different sequences.

Table of Contents

## Syntax

Format:

```
array = numpy.arange(start, stop, step, dtype=None)
```

Here,

`start`

-> The starting point (*included*) of the range, which is set to**0**by default.`stop`

-> The ending point () of the range**excluded**`step`

-> The step size of the sequence, which is set to 1 by default. This can be any real number**except**Zero.`dtype`

-> The type of the output array. If`dtype`

is not given (or provided as`None`

), the datatype will be inferred from the type of other input arguments.

Let us take a simple example to understand this:

```
import numpy as np
a = np.arange(0.02, 2, 0.1, None)
print('Linear Sequence from 0.02 to 2:', a)
print('Length:', len(a))
```

This will generate a linear sequence from **0.2** (included) until **2** (excluded) with a step size of **0.1**, so there will be (2 – 0.2)/0.1 – 1 = **20 **elements in the sequence, which is the length of the resulting numpy array.

**Output**

```
Linear Sequence from 0.02 to 2: [0.02 0.12 0.22 0.32 0.42 0.52 0.62 0.72 0.82 0.92 1.02 1.12 1.22 1.32
1.42 1.52 1.62 1.72 1.82 1.92]
Length: 20
```

Here is another line of code which generates the numbers from 0 to 9 using `arange()`

, using the default step size of 1:

```
>>> np.arange(0, 10)
array([0, 1, 2, 3, 4, 5, 6, 7, 8, 9])
```

If the step size is provided as 0, this is not a valid sequence, since a step of 0 implies you are dividing the range by 0, which will raise a `ZeroDivisionError`

Exception.

```
import numpy as np
# Invalid Step Size!
a = np.arange(0, 10, 0)
```

**Output**

```
ZeroDivisionError: division by zero
```

**NOTE**: This function is a bit different from `numpy.linspace()`

, which, by default, includes both the starting and the endpoints for the sequence calculation. It also does not take the step size as an argument, but rather takes only the number of elements in the sequence.

### A simple example

Let’s now put all of this together into a simple example to demonstrate the linearity of the sequences generated by `numpy.arange()`

.

The following code plots 2 linear sequences between `[0, 20]`

and `[0, 10]`

using `numpy.arange()`

to show that there is uniformity generated by the sequence, so the resulting arrays are linear.

```
import numpy as np
import matplotlib.pyplot as plt
y = np.zeros(5)
# Construct two linear sequences
# First one has a step size of 4 units
x1 = np.arange(0, 20, 4)
# Second one has a step size of 2 units
x2 = np.arange(0, 10, 2)
# Plot (x1, [0, 0, ..])
plt.plot(x1, y, 'o')
# Plot (x2, [0.5, 0.5, ..])
plt.plot(x2, y + 0.5, 'o')
# Set limit for y on the plot
plt.ylim([-0.5, 1])
plt.show()
```

**Output**

As you can see, the orange dots represent a linear sequence from 0 to 10 having a step size of 2 units, but since 10 is not included, the sequence is `[0, 2, 4, 6, 8]`

. Similarly, the blue dots represent the sequence `[0, 4, 8, 12, 16]`

.

## numpy.arange() vs range()

The whole point of using the `numpy`

module is to ensure that the operations that we perform are done as quickly as possible, since `numpy`

is a Python interface to lower level C++ code.

Many operations in `numpy`

are *vectorized*, meaning that operations occur in parallel when `numpy`

is used to perform any mathematical operation. Due to this, for large arrays and sequences, `numpy`

produces the best performance.

Therefore, the `numpy.arange()`

is much faster than Python’s native `range()`

function for generating similar linear sequences.

### Performance Test

We *should not* interleave `numpy`

‘s vectorized operation along with a Python loop. This slows down performance drastically, as the code is iterating using native Python.

For example, the below snippet shows how you should **NOT** use numpy.

```
for i in np.arange(100):
pass
```

The recommended way is to directly use the `numpy`

operation.

```
np.arange(100)
```

Let’s test the difference in performance using Python’s `timeit`

module.

```
import timeit
import numpy as np
# For smaller arrays
print('Array size: 1000')
# Time the average among 10000 iterations
print('range():', timeit.timeit('for i in range(1000): pass', number=10000))
print('np.arange():', timeit.timeit('np.arange(1000)', number=10000, setup='import numpy as np'))
# For large arrays
print('Array size: 1000000')
# Time the average among 10 iterations
print('range():', timeit.timeit('for i in range(1000000): pass', number=10))
print('np.arange():', timeit.timeit('np.arange(1000000)', number=10, setup='import numpy as np'))
```

**Output**

```
Array size: 1000
range(): 0.18827421900095942
np.arange(): 0.015803234000486555
Array size: 1000000
range(): 0.22560399899884942
np.arange(): 0.011916546000065864
```

As you can see, `numpy.arange()`

works particularly well for large sequences. It’s almost **20** times (!!) as fast as the normal Python code for a size of just **1000000**, which will only scale better for larger arrays.

Therefore, `numpy.arange()`

should be the unanimous choice among programmers when working with larger arrays.

For smaller arrays, when the difference in performance isn’t that much, you could use among either of the two methods.

Your timing test is testing different things for range and arange. For range you run ‘for i in range(1000000): pass’. The creates the range object and then iterates over it. For arange you simply create the array, but you don’t iterate over it. When I perform the apples-to-apples test using “timeit.timeit(‘for i in np.arange(1000000): pass’, number=100, setup=’import numpy as np’)”, I get opposite results from yours. The python range ran in 1/4 the time of the np.arange in a fair test. Also, since range is a generator in python 3 (like xrange in python 2), it doesn’t allocate the memory for the array.

To summarize, in a fair test, range is faster and uses less memory than arange for loop indexing. arange has uses outside of simple loop indexing, but for loop indexing it is worse in both speed and size than the standard python range object.