Here, you'll learn all about merging pandas DataFrames. You'll explore different techniques for merging, and learn about left joins, right joins, inner joins, and outer joins, as well as when to use which. You'll also learn about ordered merging, which is useful when you want to merge DataFrames whose columns have natural orderings, like date-time columns.
import pandas as pd
Suppose your company has operations in several different cities under several different managers. The DataFrames
managers contain partial information related to the company. That is, the rows of the
city columns don't quite match in
managers (the Mendocino branch has no revenue yet since it just opened and the manager of Springfield branch recently left the company).
revenue = pd.read_csv("revenue.csv") managers = pd.read_csv("managers.csv") print(revenue) print(managers)
The DataFrames have been printed in the IPython Shell. If you were to run the command
combined = pd.merge(revenue, managers, on='city'), how many rows would combined have?
combined = pd.merge(revenue, managers, on="city") print(combined)
Correct! Since the default strategy for
pd.merge() is an inner join,
combined will have 2 rows.
The merge command is the key learning objective of this tutorial. The merging operation at its simplest takes a left dataframe (the first argument), a right dataframe (the second argument), and then a merge column name, or a column to merge “on”. In the output/result, rows from the left and right dataframes are matched up where there are common values of the merge column specified by “on”.
An inner merge, (or inner join) keeps only the common values in both the left and right dataframes for the result.
You expect your company to grow and, eventually, to operate in cities with the same name on different states. As such, you decide that every branch should have a numerical branch identifier. Thus, you add a
branch_id column to both DataFrames. Moreover, new cities have been added to both the
managers DataFrames as well.
revenue = pd.read_csv("revenue_branch_id.csv") managers = pd.read_csv("managers_branch_id.csv") print(revenue) print(managers)
pd.merge(), merge the DataFrames
managers on the
'city' column of each
merge_by_city = pd.merge(revenue, managers, on="city") print(merge_by_city)
Merge the DataFrames
managers on the
'branch_id' column of each.
merge_by_id = pd.merge(revenue, managers, on="branch_id") print(merge_by_id)
Well done! Notice that when you merge on
'city', the resulting DataFrame has a peculiar result: In row 2, the city Springfield has two different branch IDs. This is because there are actually two different cities named Springfield - one in the State of Illinois, and the other in Missouri. The
revenue DataFrame has the one from Illinois, and the
managers DataFrame has the one from Missouri. Consequently, when you merge on
'branch_id', both of these get dropped from the merged DataFrame.
We continue working with the
managers DataFrames from before. This time, someone has changed the field name
'branch' in the
managers table. Now, when you attempt to merge DataFrames, an exception is thrown:
revenue = pd.read_csv("revenue_branch_id_2.csv") managers = pd.read_csv("managers_branch_id_2.csv") print(revenue) print(managers)
pd.merge(revenue, managers, on='city') Traceback (most recent call last): ... <text deleted> ... pd.merge(revenue, managers, on='city') ... <text deleted> ... KeyError: 'city'
Given this, it will take a bit more work for you to join or merge on the city/branch name. You have to specify the
right_on parameters in the call to
combined = pd.merge(revenue, managers, left_on="city", right_on="branch") print(combined)
Great work! It is important to pay attention to how columns are named in different DataFrames.
Another strategy to disambiguate cities with identical names is to add information on the states in which the cities are located. To this end, you add a column called
state to both DataFrames from the preceding exercises.
Our goal in this exercise is to use
pd.merge() to merge DataFrames using multiple columns (using
'state' in this case).
revenue = pd.read_csv("revenue_branch_id.csv") managers = pd.read_csv("managers_branch_id.csv") # Add 'state' column to revenue revenue["state"] = ["TX", "CO", "IL", "CA"] # Add 'state' column to managers managers["state"] = ["TX", "CO", "CA", "MO"] print(revenue) print(managers)
# Merge revenue & managers on 'branch_id', 'city', & 'state' combined = pd.merge(revenue, managers, on=["branch_id", "city", "state"]) print(combined)
There are three different types of merges available in Pandas. These merge types are common across most database and data-orientated languages (SQL, R, SAS) and are typically referred to as “joins”. If you don’t know them, learn them now.
merge type to use is specified using the
how parameter in the merge command, taking values
inner (default), or
Venn diagrams are commonly used to exemplify the different merge and join types.
We now have, in addition to the
managers, a DataFrame
sales that summarizes units sold from specific branches (identified by
state but not
sales with a right merge, we can identify the missing
revenue values. Here, we don't need to specify
right_on because the columns to merge on have matching labels.
managers = pd.read_csv("managers_branch_id_2.csv") managers["state"] = ["TX", "CO", "CA", "MO"] sales = pd.read_csv("sales.csv") print(sales)
revenue_and_sales = pd.merge(revenue, sales, how="right", on=["city", "state"]) print(revenue_and_sales)
managers with a left merge, we can identify the missing manager. Here, the columns to merge on have conflicting labels, so we must specify
right_on. In both cases, we're looking to figure out how to connect the fields in rows containing
sales_and_managers = pd.merge( sales, managers, how="left", left_on=["city", "state"], right_on=["branch", "state"] ) print(sales_and_managers)
Well done! This is a good way to retain both entries of
The merged DataFrames contain enough information to construct a DataFrame with 5 rows with all known information correctly aligned and each branch listed only once. We will try to merge the merged DataFrames on all matching keys (which computes an inner join by default). We can compare the result to an outer join and also to an outer join with restricted subset of columns as keys.
merge_default = pd.merge(sales_and_managers, revenue_and_sales) print(merge_default)
merge_outer = pd.merge(sales_and_managers, revenue_and_sales, how="outer") print(merge_outer)
revenue_and_sales only on
['city','state'] using an outer join.
merge_outer_on = pd.merge( sales_and_managers, revenue_and_sales, how="outer", on=["city", "state"] ) print(merge_outer_on)
Fantastic work! Notice how the default merge drops the
Springfield rows, while the default outer merge includes them twice.
This exercise uses DataFrames
houston that contain weather data from the cities Austin and Houston respectively.
Weather conditions were recorded on separate days and we need to merge these two DataFrames together such that the dates are ordered. To do this, we'll use
pd.merge_ordered(). Note the order of the rows before and after merging.
austin = pd.read_csv("austin.csv") houston = pd.read_csv("houston.csv") print(austin) print(houston)
Perform an ordered merge on
tx_weather = pd.merge_ordered(austin, houston) print(tx_weather)
Perform another ordered merge on
This time, specify the keyword arguments
suffixes=['_aus','_hus'] so that the rows can be distinguished.
tx_weather_suff = pd.merge_ordered( austin, houston, on="date", suffixes=["_aus", "_hus"] ) print(tx_weather_suff)
Perform a third ordered merge on
This time, in addition to the
suffixes parameters, specify the keyword argument
fill_method='ffill' to use forward-filling to replace
NaN entries with the most recent non-null entry
tx_weather_ffill = pd.merge_ordered( austin, houston, on="date", suffixes=["_aus", "_hus"], fill_method="ffill" ) print(tx_weather_ffill)
Well done! Notice how after using a fill method, there are no more
Hurray! You have come to the end of the tutorial. In this tutorial, you learned to merge DataFrames using the
merge() function of pandas library. Towards the end, you also practiced the special function