Intro to Pandas for Data Analytics

P ROFESSOR L ESLIE A LBERT

S AN J OSE S TATE U NIVERSITY

In this tutorial, you will learn how to use Python Pandas library to explore data. Participants will need a Gmail account to access
Google Colaboratory. You’ll need a computer and browser, a GMAIL account, and the provided data: TableauSalesDala.xlsx.

INTRO TO PANDAS FOR DATA ANALYSIS PLAN

1. Learn how to open Google Colab, upload a data file, import Pandas and data
2. Introduce project goals: investigate underperforming sub-categories in terms of profits for Office Solutions, an online office
supplies retailer.
3. Preview data and columns, use the .describe method to see data descriptive statistics
4. Use the .unique method to see values inside Sub-Category columns
5. Determine Sub-Category Profits and identify those with negative Profits
6. Look at individual products within the Tables Sub-Category and determine their Profits and Sales
7. Look at regional and customer Segment effects on Table Profits and Sales
8. Look at average Discounts on products in the Tables Sub-Category along with Profits
9. Look at the correlation between Discounts and Sales, Discounts and Profits
10. Look at yearly differences in Tables Profits and Sales
11. Summarize findings, make a recommendation

BACKGROUND: An office supply retailer, Office Solutions, has recently hired you to help analyze their sales data. Your boss,
Natasha, is looking to boost profits and sales. One suggestion to improve profits is to get rid unprofitable or low profit products. As
such, she has asked you to determine low performing sub-categories of merchandise and has given you a spreadsheet,
TableauSalesDala.xlsx, with 4 years of sales data to analyze. As you go through the exercises below, look beyond the code and pay
attention to how I’m drilling down into the data, exploring it from many different angles to build a “story,” a logical flow of data
insights that builds a case for my recommendation.

PART 1A: IF YOU’VE NEVER USED GOOGLE COLABORATORY (GOOGLE COLAB).

Before you can use Colab you need to add it to your Google Drive.
1. Navigate to: https://colab.research.google.com
2. Open a Colab Notebook by clicking Google Drive and selecting New Notebook.
3. Once the notebook is open, go back to File and select Save the Notebook to Drive, this will create a copy of the file stored in your
own Google drive. You will now see Colab as an option under the New button under the More option (see Part 1B below)
4. If you don’t see Colab listed under the New Button you may need to refresh your Drive window,
5. The next time you want to use Colab, skip this step and start with Part 1B, below.

PART 1B: GETTING STARTED WITH COLAB

1. Download the Excel file “TableauSalesData.xlsx” from seminar
website
2. Open your Gmail account and under “Apps” select Google Drive
3. In Drive, click the “New” button on the left, scroll down to “More”
and select “Google Colaboratory.

8/26/2021
PART 2: IMPORTING DATA
1. In Colab, click the folder icon on the top left, then click “Upload.” Browse to where you saved the tableauSalesData.xlsx file
and click “Open”
2. Right click on the TableauSalesData.xlsx file and “copy path” (this is the name and location of the file).
3. Add a code cell by clicking the + Code at the top of the page and type the code below into it. Replacing the location & file
name in the example below with your file path, be sure to keep it in quotes. Click the arrow to run your code. Please note
that every time you end then restart a Colab session you’ll need to re-upload your data file...it’s not saved but your code is.
You may also need to rerun code cells that build dataframes you want to use.

FIGURE 1: IMPORT DATA

4. So what’s happening in the code in Figure 1 above? We are using the Python pandas library to import data from an Excel
spreadsheet and create a dataframe. Note the abbreviation of pandas as pd…this is convention and the abbreviation is
called an “alias.” The next line, we create xl and use it to tell pandas where are data it. The easiest way to get the file path is
to right click on your file icon on the left (TableauSalesData.xlsx) and select copy path then control + V to paste it in the
parentheses. Don’t forget the quotes. Note you could combine lines 2 and 3 in the above if you wish. We are reading in our
data with ExcelFile but there are other ways to import data including pd.read_excel() or converting our Excel file into a .csv
and using read_csv() though the latter could alter your data. You may import the data however you like as long as you end
up with a dataframe we’re good.
5. Now let’s print the first 10 rows of our data to take a look at it by adding the code print(SalesData.head(10)) and clicking the
run arrow – see Figure 2 below. The .head() prints from the top row of our dataframe. The default is 5 rows but we can
specify more or less. If you’ve never used Python before you should also know, it is case sensitive. This means Python sees
ExcelFile and Excelfile as different words. And if you write Print instead of print for a print statement you’ll get an error, so
be careful when reproducing the code provided. Also, you can name your dataframe something else but it can help to have
names that tell you what’s in your dataframes and Python won’t allow spaces in variable names.

FIGURE 2: PRINTING THE FIRST 10 ROWS IN OUR DATAFRAME

You can’t see all of the columns but you can get an idea of the data in our dataframe. A dataframe is a Pandas Data Structure but
you can think of it like a spreadsheet. It has columns of data (called series) with column labels and rows identified in this case by
index numbers there on the far left. A better option for viewing dataframes is the display function, illustrated below in Figure 3. My
screen capture couldn’t show all of it but display shows more of the data and formats it much more nicely than does print.
Sometimes print is better, sometimes display is better. I’ll use both in this tutorial.
8/26/2021
 FIGURE 3: DISPLAYING FIRST 5 ROWS

PART 3: LEARNING ABOUT THE DATA

1. LIST OF COLUMNS. Add a new code cell then type the one line of code below to see the column names. When you use
the arrow to run your cell you should see output like Figure 4, below. Seeing these is really helpful because we’ll be pulling
data from our dataframe using column names…these names must be exact, even capitalized the same as they are in these
results:

FIGURE 4: .COLUMNS() FUNCTION

2. .DESCRIBE FOR DATA DESCRIPTIVE STATISTICS: Add a new code cell then type and run the following code:
SalesData.describe() and you’ll see the output in Figure 5.

FIGURE 5: .DESCRIBE()

8/26/2021
 The .describe() is a predefined function that runs descriptive statistics on the numerical values in your dataframe. But not all of
it makes sense. For example, what is the average of postal codes? Pandas see postal codes (zip codes) as numbers and treats
them as such. Now this could be in issue in other applications of this data. There are states with postal codes that begin with
zeros, like those in the northeast US like Connecticut, Massachusetts, Maine. When we treat a postal code like a number,
they’re really text with no numerical value, that leading zero is dropped. It’s unlikely that postal codes without leading zeros will
cause issues in our project but you should be aware of them.

3. .DESCRIBE() FOR SPECIFIC COLUMNS: If we are interesting in just

seeing specific columns we can specify them after the dataframe
name as I did in Figure 6. If you have more than 1 column you want
to describe, please be sure to include two sets of square brackets or
you’ll get an error. The inner set of [] is because we are passing the
function a list rather than a single value…more on that later.

4. .UNIQUE() TO GET A LIST OF UNIQUE VALUES IN A COLUMN. FIGURE 6: .DESCRIBE ON SPECIFIC COLUMNS

We’ve been given the challenge of identifying poor performing sub-

categories so let’s look at the types of sub-categories we have. I’ve created a new code cell, see Figure 7. In it I’ve asked
Pandas to use the .unique() method. This is a predefined function that returns all the unique values in the column we
specify and saves them in a numpy array, I’ve called SubCats. Numpy is another Python library that Pandas is built on top of.
If you’ve never heard of an array, don’t worry, it’s like a list, just less flexible - all of the elements in an array have to be of
the same data type. I have also printed my SubCats array so I can see all the sub-categories.

FIGURE 7: .UNIQUE TO IDENTIFY ALL THE UNIQUE VALUES IN A COLUMN

Be careful that “Sub-Category” is spell correctly in your code…you’ll get the KeyError below if it’s not (you can see that I misspelled
Sub-Category as Sub-Categroy in Figure 8, below). This error happens because Pandas is looking for an exact column name in the
dataframe and cannot find Sub-Categroy in the list of columns.

FIGURE 8: KEY ERROR FROM MISSPELLED COLUMN NAME

Also, Python is case sensitive to subcats is not the same as SubCats. The SubCats array name must match in the unique() and print
statements below.

5. MODIFYING THE .UNIQUE() CODE TO SEE VALUES IN OTHER COLUMNS. You can also modify the correct code to
see the unique values in other columns like I did below in Figure 9 for “Category” and for “Segment” which are customer
types.

8/26/2021
 FIGURE 9: .UNIQUE LOOKING AT OTHER COLUMNS

PART 4: GROUP, SUM, & SORT

In this section we’ll first determine those sub-categories that aren’t helping Office Solutions meet its financial goals.

1. WEEDING OUT UNNEEDED DATA: The first thing we need to do to understand Sub-Category Profits is to tell Pandas to
only look at certain columns in our dataframe and to disregard the others. Much of the work in analyzing data with Pandas
is getting rid of data we don’t need and that could complicate our analyses. In the code in Figure 10 below, I am naming a
reference to three columns in my SalesData dataframe, Sub-Category, Profit, and Sales. It looks like we are creating new
dataframe called SubCatProfits with only three columns. That’s not what’s really happening…instead we are telling Pandas
just to look at these columns when we write SubCatProfits. If you are new to Python and Pandas, this can be confusing so
don’t worry. We are going to treat our new SubCatProfits like a new dataframe with just three columns in it in our next
step. Try adding the code below. The results below is just a partial image of the output. You might notice that the
Accessories Sub-Category appears twice. That’s because we haven’t done any grouping and these are just the first 10 rows
of sales data in my dataframe. NOTE: Be sure there are two sets of square brackets around Sub-Category, Profit, and Sales!

FIGURE 10: PICKING ONLY THE COLUMNS WE WANT TO WORK WITH

2. .GROUPBY, .SUM(), AND .SORT_VALUES(): What we really want to see is the total of all Profits for each Sub-
Category so we want Pandas to group records by Sub-Category and total up the Profits and Sales. That’s just what the next
code does in Figure 11. I’m also going to sort the resulting values by Profit so I can quickly see lowest to highest. With
sort_values, sorting values in ascending order is the default so strings will appear in alphabetical order, our Profits, below,
are smallest to largest. Also, groupby will automatically sort by your first column so if you just want Sub-Categories sorted
in alphabetical order, you don’t need sort_values.

8/26/2021
 FIGURE 11: GROUPBY, SUM, AND SORT_VALUES

So what’s happening in the code above? First, I ask Pandas to look through all 9994 rows of Sub-Category data, group them together
and total up their Profits and Sales, the assign a references to these with a new dataframe name TotalSubCatProfits. Of course, I still
need a print or a display statement to see my results. As you can see above, Tables, Bookcases, and Supplies are all negative! These
aren’t just underperforming; they are hurting the Office Solution’s total profits. Now your instinct may be to say, “Let’s get rid of
these sub-categories!” but we don’t know much about them. What if these sub-categories are much better performing in some
geographical Regions than others or maybe there are some highly profitable products in these sub-categories but they are
overshadowed by highly negative products when we total them up. Let’s take a look in there to see what might be going on.

3. Let’s look at another example of grouping. Maybe instead of looking at the Sub-Category level I want to look at Category Profits
and Sales. In Figure 12 you’ll see this code plus how you can condense the code above into a single line. I also use the display here
instead of print but when you are looking at small amounts of output the difference between the two isn’t so great. Note you can
add another argument to sort_values to reverse the sort order: .sort_values(by= “Profit”, ascending = False)

FIGURE 12: CATEGORY PROFIT AND SALES

4. Multi-level grouping. We can also group by more than one value. In Figure 13 you can see how I grouped first by Category and then
by Sub-Category. You need to use the [] as you are passing a list to the groupby function, not just the single value it is expecting. Also,
you can see that I split the column selection and grouping into two lines…this is mainly for legibility. I also dropped the sort_values
and let the group by order the output based on category. I chose print instead of display too. Display is prettier but print is more
compact. Be sure that if you are modifying the code in Figure 12 to make the code in Figure 13 than you are printing the correct
dataframe, printing the second dataframe in Figure 12 TotalCatSubCat, not the first, CategoryProfitsSales. I make this mistake all the
time! You can group on as many columns as you like but it does have its limits – groupby needs repeated values to work. If you try
grouping on a column with many unique values, like Profit, there isn’t much this function will be able to do for you.

8/26/2021
 FIGURE 13: MULTILEVEL GROUPING CATEGORY AND SUB-CATEGORY

PART 5. USING THE .LOC FUNCTION TO FIND ROWS CONTAINING SPECIFIC VALUES PLUS A
FILTER EXAMPLE
1. To dig into the data analyses I’m going to focus on just one of the negative sub-categories we found previously: Tables. I’ll
show you how to modify the code to see Bookcases in a bit. The first thing I’m going to do is to ask Pandas to find me all the
rows in SalesData, my original dataframe, where the value in the Sub-Category column is equal to Tables. For this we are
going to use the .loc method in Figure 14, below. We will use the dataframe JustTables created below in several exercises
so you’ll want to reproduce it carefully.

FIGURE 14: LOCATING JUST THOSE ROWS WITH TABLES IN THE SUB-CAT COLUMN

What we’ve done in this code is to start over with our original full set of data in the SalesData dataframe, then located only those
rows with sales data about Tables. I’m using the .head() to limit my output to print only the first 10 rows. That’s because it without
doing this, the results will be 9994 rows and our code cell won’t print all that…it’s just too much data. However, 21 columns are
there noted at the bottom of the output. Be sure you add both == in the first line about. This is how Python makes a comparison, like
A equals B. A single = is an assignment statement were we assign values like we did for JustTables. If you want to look at Bookcases
instead of Tables the code would like that in Figure 15. You can modify this code for any Sub-Category you want to look at.

8/26/2021
 FIGURE 15: LOCATING JUST BOOKCASES

2. TOTAL PROFITS AND SALES FOR EACH PRODUCT IN THE TABLE SUB-CATEGORY. Now I need a new dataframe
that only looks at my three columns of interest, Product Name, Sales, and Profit. You can see below that we are doing
something similar to before, only we have 3 columns we are working with instead of two and we are grouping by Product
Name rather than Sub-Category. You may be wondering why we didn’t include Sub-Category as a column. Well do we really
need it? We already weeded out all the Sub-Categories that weren’t Tables so we could have a column that listed table in
each row but it wouldn’t add much to our output. Also, you’ll see that I didn’t specify which columns I wanted .sum() to add
up. We don’t need to, .sum() will total up any and all numerical data for us – even if it doesn’t make much sense, like postal
code, which is why we want to remove any columns we don’t need as we play with the data. I’ve combined the column
selection and groupby / sum into one line in Figure 16 but you can do it in two steps if you prefer.

FIGURE 16: A FEW OF THE NEGATIVE PROFIT TABLE PRODUCTS

Figure 16 above doesn’t show all the output from this code but in your code cell you’ll see that many of the table products are
negative and those that aren’t really aren’t making the company much money. Even the best performing table product at the end of
our list only made Office Solutions about $600 in 4 years!

What if you wanted to find only those rows that contain two or more specific values, for example Table sales to Corporate
customers in just one line of code? For this we can build a statement containing multiple criteria for or .loc. Here’s an example of
what that would look like, note how we use the & to tell the .loc that both stated conditions needs to be met for the row to be
selected:

FIGURE 17: MULTI-CRITERIA LOC

3. FINDING ONLY NEGATIVE TABLE PRODUCTS. If we are only interested in finding negative profit table products, we
can use a filter as shown in Figure 18. Note that I group and sum before filtering so I’m only finding those tables products
with total negative profits over 4 years, not each instance where a table sold for a negative profit. I only provide a portion
of the output in Figure 18.

8/26/2021
 FIGURE 18: NEGATIVE TABLES ONLY

PART 6. DIGGING DEEPER TO UNDERSTAND TABLE PROFITS AND SALES

1. At this point you may be thinking, “Tables are a really underperforming Sub-Category. Let’s discontinue Tables! While it is true that
OS is losing money on almost all of its Table products, we need to look further into the data to understand what’s going on and why
so many Tables are unprofitable. There can also be repercussions to discontinuing whole Sub-Categories. For example, what if our
best customers by lots of tables from us? Do we want to lose their business if they make us money on other products? Is there a
difference in table profitability depending on whom we sell them to or where we sell them? We don’t have a full picture yet so before
we advise Natasha, we should dig deeper into the data. Let’s look at two columns in particular: Segment (i.e. customer segment) and
Region. Before we do that though, keep in mind that we are still working with the JustTables dataframe we created earlier. If you
stopped working through this tutorial and are now in a new Colab project then you’ll need to upload your data and import it and then
run this code again:

Below in Figure 19 is the code for Segment and Regional analyses of the Table Sub-Category’s Profit and Sales. You may be
wondering, why look at regional effects or consumer Segment effects? Before we make a recommendation, we want to understand
everything we can about the data. What if we lose money on Tables in three regions but make a ton of money in the fourth…profits
that were obscured at a higher level of analysis by other regions’ losses? If so, that could change a recommendation from
discontinue Tables to only marketing Tables in the one profitable region. Alternatively, what if sell a lot of Tables in one Segment and
that Segment is one we really care about, like the Corporate Segment? This is something Natasha may want to consider before
discontinuing Tables

FIGURE 19: TABLE PROFITS AND SALES BY CUSTOMER SEGMENT & REGION

8/26/2021
Again, we are only working with the table data but what we can see from above is that we lose money on Tables in each Segment,
though we lose the most to individual Consumers, and we lose money on Tables in all the Regions but the West.

2. T OTAL TABLE PROFITS AND SALES FOR EACH CUSTOMER SEGMENT IN JUST THE WEST REGION. We can combine
these to investigate further to see which, if any, of our customer Segments makes a profit on Tables in the West Region. For that, we
will use the .loc method again. It can get tricky though as we create a new dataframe reference on each line, building on the work
we did on the previous line. Track your dataframes carefully to make sure you are using / printing the right one. Note we could have
used a multi-criteria .loc statement like we did back in Figure 17 to find all the Tables in the Western Region. However, we’ve already
been working with the JustTables dataframe for a while so now we just need to isolate those records in the West.

FIGURE 20: BREAK DOWN OF SEGMENT PROFITS AND SALES IN THE WESTERN REGION

Let’s break down what’s happening above. First, we are starting with the JustTables dataframe…it only has those rows that have
Tables in the Sub-Category column but it has all the original SalesData columns. Next, we create a new dataframe reference
TableSegRegion that only looks at 4 of the columns in our JustTables dataframe. In line two we use the .loc method again to find
only those rows that have West in the Region column and to ignore the other rows. Once we have a dataframe, WestTableSegment,
that has only Table Sales in the West Region, we grouped by Segment and totaled up our Sales and Profits. As you can see from the
output, we do make some money on Tables in the Consumer and Corporate Segments in the West Region, not a lot but it’s
important for us to give Natasha as a complete a picture of table performance as we can. Maybe the evidence we have is enough for
her to want to discontinue Tables…or drastically reduce the Table products OS offers. But we still don’t know why we lose so much
on Tables. Looking at our columns of data, we don’t have cost of goods sold but we do have data on discounts. Maybe there’s a lead
there.

3. .MEAN TO SEE AVERAGE DISCOUNTS ON TABLES. Below, in Figure 21, is code that will help us see what the average
overall Discount is on Tables and what the Discount is on some of our table products. The .mean() is similar to the .sum() we used
before and only works on numbers.

FIGURE 21: CALCULATED THE AVERAGE (MEAN) DISCOUNT ON TABLES

Below is a part of the output from this code. At the top, we see that the average Discount on Tables is high, 26%, even while the
average Profit is negative. Next, we see the most discounted Tables some of which are discounted almost 50%! Also, notice how the
Profit on all these highly discounted table products are negative! Looks like we might have found at least one issue with our poorly
performing Tables Sub-Category!

8/26/2021
 FIGURE 22: OUTPUT OF MEAN TABLE DISCOUNT

PART 7. C ORRELATION BETWEEN TABLE DISCOUNTS AND SALES & DATES (.DT)
1. Companies typically give discounts to drive up sales or to clearance out products. Sometimes, companies will even lose money to
attract customers. An example of this are loss leaders…products discounted deeply to get you into the store to spend your money on
other products – think Door Busters on Black Friday. However, OS is losing money on most of their Table products so this probably
isn’t what’s happening. Let’s see if there’s any relationship (correlation) between Discounts and Table Sales. If the discounts are
driving sales as we hope, there should be a moderate to strong positive correlation between the two. In other words, when
discounts increase, sales also increase. If sales decrease and discounts increase, that would be a negative correlation and
certainly not what we want as discounts can drive down profits. Below, in Figure 23, is the code to determine this correlation.
Note I had to import a new library, scipy.stats.

FIGURE 23: CORRELATION BETWEEN DISCOUNTS & SALES

So what’s happening in the code? First, we need an x and a y to correlate. Yes, we could write these into our spearmanr code
but by assigning them to variables, we can change them easily if we want. We also just want to look at the correlation between
sales and discounts for Tables only so I’m using my JustTables dataframe. GetCorrelation is a numpy array that holds the results
of this calculation as two values, the correlation in the first slot and the p value in the second (pandas is built on top of numpy
so we don’t have to import that library). I’ve rounded the values for readability. For correlation, two decimal places is enough,
for p-values, they can get quite small so I rounded to five decimal places. After that, we just need to convert to strings to print
plus an explanation of what we are seeing. Our results tell us there is a moderate negative correlation between discounts and
sales. The p values is significant but given how many records we have that’s probably not meaningful – when you have lots of
data even small, un-meaningful relationships can appear significant. However, what is important to know is that OS’s current
approach to discounts on Tables is backfiring and hurting, not helping, sales.

2. .DT (DATE TIME) AND CREATING A NEW COLUMN TO SEE ANNUAL SALES OF TABLES. For the last analysis we’ll do
together, let’s see if there’s a year-over-year trend in table Sales. This is a bit more complicated so we’ll do it in two steps.

8/26/2021
STEP 1: We have a column called Order Date – the date when the order was place. Before we can group by year, we first need to
get the Order Date’s year all by itself. Originally, it was in a day, month, and year format. So we create a new column in our
JustTablesYears dataframe called Year then use the .dt method (date time) to pull the year out of each Order Date and put it in the
Year column. You can replace .dt.year with .dt.month or .dt.quarter etc. to see other periods. Using .dt.quarter will break the year
into 4 quarters that align with the seasons (Q1 = Jan., Feb., March = winter) if you are interested in looking at seasonal effects. You
can see our new year column in our list of columns, in Figure 24 below.

FIGURE 24: ADDING A NEW SERIES (COLUMN) TO HOLD THE YEAR PORTION OF ORDER DATE

STEP 2: Now we can use the same groupby and sum methods we’ve used previously. From the output below we can see Table
Sales are negative each year and there doesn’t seem to be a trend over the 4 years in terms of Profits but it looks like total Sales of
Tables is on the decline (Figure 25).

FIGURE 25: GROUPING & SUMMIING ANNUAL TABLE SALES

PART 8: COMBINING SEVERAL LEVELS OF A NALYSIS AT ONCE

We’ve looked at the Tables data in quite a few different ways and could probably recommended the company discontinue Tables
sales, or perhaps only sell them to Corporate accounts. But let’s look at just a couple more analyses. The embedded green comments
explain each line.
1. AVERAGE TABLE DISCOUNTS BY REGION.

FIGURE 26: AVERAGE TABLE DISCOUNTS BY REGION

8/26/2021
2. AVERAGE TABLE DISCOUNTS BY REGION FOR EACH YEAR (ADDING A LOOP).

FIGURE 27: LOOPING THROUGH YEARS TO SEE ANNUAL REGIONAL DISCOUNTS ON TABLES

FIGURE 28: OUTPUT OF AVERAGE TABLE DISCOUNTS BY REGION FOR EACH YEAR

8/26/2021
3. A VERAGE TABLE D ISCOUNTS BY CUSTOMER S EGMENT FOR EACH YEAR. You’ll notice the code is a combination of what
we’ve done before. We are still looping through the years but we’ve swapped the Region column for Segment.

FIGURE 29: ANNUAL SEGMENT TABLE DISCOUNTS CODE

FIGURE 30: ANNUAL SEGMENT TABLE DISCOUNTS OUTPUT

8/26/2021
4. TOTAL TABLE PROFITS AND S ALES FOR EACH CUSTOMER S EGMENT BY R EGION LOOP . We could also look at all
Segments in all Regions at the same time but that requires a loop too.

FIGURE 31: SEGMENT TABLE PROFITS & SALES BY REGION

FIGURE 32: SEGMENT TABLE PROFITS & SALES BY REGION OUTPUT

5. AVERAGE ANNUAL TABLE DISCOUNTS FOR EACH SEGMENT BY REGION. This requires a loop within a loop, in Figure
33. In our first loop, we isolate data for 1 year. Once we have that we loop through all of the four regions for that year in the inner
loop before hopping back to the outer loop to grab the data from our second year. A partial output follows in Figure 34.

8/26/2021
 FIGURE 33: AVERAGE ANNUAL TABLE DISCOUNTS FOR EACH SEGMENT BY REGION CODE

FIGURE 34: AVERAGE ANNUAL TABLE DISCOUNTS FOR EACH SEGMENT BY REGION OUTPUT

6. COMBINING ANALYSES WITH .GROUPBY

If you want to do both .sum and .mean in the same output, things can get a bit trickier. In the figure below I use .agg to allow me to
sum Profit while also take the mean of Discounts. The first 5 records of the output are also included.

8/26/2021
 FIGURE 35: .AGG TO COMBINE SUM AND MEAN

PART 9: BASIC GRAPHS

In this section, I will show you how to create two basic graphs. Be sure to think about the story you are trying to tell when selecting a
graph, chart, or table…they each have their strengths and limitations.

1. Importing the needed libraries. Here are two very helpful libraries for
graphs and charts, matplotlib and seaborn. Matplotlib is a very powerful
library that will let you control many aspects of your graphs. Seaborn
layers on top of matplotlib and reduces the amount of code you need to
write to create a graph. The first thing we need to do is import both of
these as shown in Figure 36. Make sure you run this code cell after you
FIGURE 36: IMPORTING MATPLOTLIB & SEABORN
add your import statements.

2. I’m going to use part of the yearly profit and sales code we created previously to create a pie chart of annual table sales. If you
haven’t used your Colab file in a while, you will need to reload your data file, run the first code cell wherein we created the
SalesData dataframe, as well as the cells that create the
JustTables dataframe and the JustTableYears dataframes.
Once that’s done, we need can determine the total Tables
FIGURE 37: CREATING THE JUSTTABLESYEARS DATAFRAME sales by year, using groupby and sum as we’ve done several
times. In order to graph this data though we will need to
reset our dataframe’s index using .reset_index(). Why? Because of the way groupby works. Look at the output in Figure 38 below. In
the first output you’ll see what how the JustTableYearsSales dataframe looks. Do you see how the Year column label is offset from
the Sales column label and how there are no index numbers? This is because the groupby uses the Year column values as indices. In
other words, if you wanted to look at the first row of the dataframe you’d look at row 2016, not row 0.

FIGURE 38: BEFORE & AFTER RESET INDEX

8/26/2021
Now check out the second print that runs after we reset the index. Do you see the index column has returned and how Year is a proper
column (i.e. a series) again? This is what our chart and graphs will need to work. I’ll show you the error you’ll get if you skip this step
in a moment. First, let try a simple pie chart using matplotlib. Add the code in the figure below. Green comments provide explanations.
You’ll see it takes very few lines of code with matplotlib to create a very professional looking pie chart.

FIGURE 39: BASIC PIE CHART

Here’s what happens when you don’t reset the index, see Figure 40…matplotlib can’t find the column Year (because it’s an indexing
column thanks to groupby) and throws an attribute error. If you don’t reset_index(), you’ll get a value error:

FIGURE 40: ATTRIBUTE ERROR FROM MATPLOTLIB FROM NOT USING RESET INDEX

8/26/2021
3. So maybe you like the pie chart but think a bar chart would be more effecting. For our bar chart, let’s use seaborn. I’ve included the
groupby, sum and reset index code in the figure below for completeness but if you’ve already run that code in a different cell, perhaps
in creating your pie chart, you need not repeat it.

FIGURE 41: BASIC BAR CHART WITH SEABORN

There are many other graphs available through these libraries and you can customize them by size, color, orientation, etc.

Now let’s wrap everything up and make a data-justified business recommendation (see next page).

8/26/2021
PART 10: MAKING A RECOMMENDATION – ONE EXAMPLE YOU CAN FOLLOW
Natasha has stated a goal of increasing profits by 10% in the next year. One way to accomplish this is to minimize losses by reducing
or discontinuing underperforming products. We present our analyses of Office Solutions’ (OS) underperforming Sub-Category,
Tables, below.

• First, we calculated the profits and sales for all sub-categories. The Tables Sub-Category has the highest negative profit of all
sub-categories, with total losses over 4 years of $17,725. Sales are strong however, totaling over $207,000. The Bookcases
and Supplies Sub-Categories are also unprofitable at (-$3,473 and -$1,189, respectively).
• We then investigated the profitability of individual table products and found most have negative profits. Only 14 table
products are profitable and these barely so with the best performing table product earning only $600 in profit in 4 years.
• We also examined how the Tables Sub-Category performs in different segments and regions:
o Tables are unprofitable in all three customer segments, earning -$9,728 in the Consumer Segment, -$4,907 in the
Corporate Segment, and -$3,091 in the Home Office Segment.
o The Tables Sub-Category is unprofitable in all sales regions (East -$11,035, South -$4,623, Central -$3,560) but the
West where it generated $1,483 in profits.
o In the West Region specifically, tables generated profits of $522 in the Consumer Segment and $1,074 in the
Corporate Segment but lost $114 in the Home Office Segment.
o We also looked into segment effects in the other Regions finding that Corporate customers also generate $700 in
profits on tables in the Central Region. Table sales to the Home Office Segment are unprofitable in all regions. All
Tables sales in the South and East Regions return negative profits. TABLE 1: REGIONAL TABLE
DISCOUNT AVERAGES
• To try to understand why tables are unprofitable, we investigated discounts given in this sub-
category. Discounts on tables are high, averaging 26% and reaching almost 50% on some
highly unprofitable products. Table 1 provides the average discounts given on tables for each
region. Further, discounts exhibit a moderate negative correlation with sales suggesting that
the current discount strategy decreases sales rather than increasing them.
• After conducting a year-over-year analysis of both table profits and sales we find OS lost
money on tables each year though no annual profit trend is apparent. Sales of tables have declined over the 4 years, as
maybe seen in Figure 1, suggesting waning demand for OS table products.

FIGURE 1: TABLES SALES BY YEAR

Recommendation: Based on the analyses above, we recommend OS reduce discounts offered on tables until they are profitable or,
given the decline in sales, clearance table inventory and discontinue all products in this sub-category.

8/26/2021