As we’ve seen so far, every sorting is done using the ascending order, which is the default behavior. However, we often want to have the data sorted by a descending order. We can take advantage of the <code>ascending </code>parameter. What should we do if we sort by multiple columns and have different ascending requirements for these columns? In this case, we can pass a list of boolean values with each corresponding to one column.

5. Sort by Descending Orders

In this article, we’ll be using the flights dataset, which records the monthly passenger numbers from 1949 to 1960. For the purpose of this tutorial, we’ll select a random subset, as shown below. When we want to sort the data by a single column, we specify the column name directly as the function call’s first parameter. As a side note, you may see me use <code>head </code>a lot, just to show you the top values without wasting the space.

1. Sort by a Single Column

Apparently, the sorted data isn’t something that we expect — the months are not in the desired order. To make this happen, we can take advantage of the <code>sort_method </code>taking a <code>key </code>parameter, to which we can pass a custom function for sorting, just like Python’s built-in <code>sorted </code>function. A possible solution is shown below.<ul><li>The <code>key </code>takes a callable, and we use a custom function here. Besides, this parameter is only available with pandas 1.1.0+.</li><li>Unlike the <code>key </code>parameter used in <code>sorted() </code>, the <code>key </code>function applies to each of the sorting columns in the <code>sort_values </code>method. </li></ul>

6. Sort by Custom Functions

It’s important to remember that your datasets can always contain NANs. Unless you’ve examined your data quality and know that there are no NANs, you should pay attention to that. When we sort values, these NANs are placed behind all the other valid values, by default. If we want to change this default behavior, we set the <code>na_position </code>parameter. <ul><li>We first inject one NAN into the <code>DataFrame </code>object. </li><li>When we do nothing with the <code>na_position </code>, the NAN value is placed at the end of the sorting group. </li><li>When we set <code>“first” </code>to <code>na_position </code>, the NAN value appears at the top. </li> </ul>

8. Don’t Forget about NANs

The above sorting using the <code>key </code>parameter can be confusing to some people. Is there a cleaner way? Pandas is arguably the most versatile library for data processing, and you can expect that there is something neat to solve this relatively common problem — converting these lexicographically unordered columns to categorical data. <ul><li>We define a <code>CategoricalDtype </code>by specifying the order of the months. </li><li>We cast the month column to the new defined category. </li><li>When we sort the month, it will use the order of the months in the category data definition. </li> </ul>

7. Sort Lexicographically Unordered Columns After Casting to Categorical

We don’t always need one column for sorting. In many cases, we need to sort the data frame by multiple columns. It’s also simple with sort_values because <code>by </code>doesn’t only take a single column but also a list of columns without any special syntax.

4. Sort by Multiple Columns

In the previous sorting, you may notice that the index goes with each sorted row, which puzzles me sometimes, when I want the sorted DataFrame has an ordered index. In this case, you can either reset the index after sorting, or simply take advantage of the <code>ignore_index </code>parameter, as shown below.

3. Reset Index After Sorting

In the previous sorting, one thing you may have notices is that the <code>sort_values </code>method will create a new <code>DataFrame </code>object, as shown below. To avoid creating a new DataFrame, you can request the sorting to be done inplace by setting the <code>inplace </code>parameter. When you do that, note that calling <code>sort_values </code>will return <code>None </code>.

2. Sort Values Inplace

When we deal with data, sorting is an important preprocessing step to visually examine the quality of your data. With pandas, although sometimes we may use a related method — sort_index, we sort data…