Heads up! To view this whole video, sign in with your Courses account or enroll in your free 7-day trial. Sign In Enroll
Preview
Start a free Courses trial
to watch this video
Discover the capabilities of ChatGPT's "Code Interpreter" model in this AI workshop. It's a secure, sandboxed Python interpreter that revolutionizes Python code testing, debugging, data analysis, and visualization. Ideal for developers, data analysts, and AI enthusiasts eager to enhance their Python programming skills and data interpretation proficiency.
Announcement
As of 08/28/2023, the ChatGPT code interpreter has been renamed to ChatGPT Advanced Data Analysis. Please be assured that aside from this name change, no other modifications were made. Therefore, you should be able to follow along with this video without any issues.
Resources
Open AI's blog post on Code Interpreter
ChatGPT Conversation about Electric Vehicle Data: Link to Chat
ChatGPT Conversation about Amazon Customer Behavior Data: Link to Chat
Kaggle: https://www.kaggle.com/datasets
Related Discussions
Have questions about this video? Start a discussion with the community and Treehouse staff.
Sign upRelated Discussions
Have questions about this video? Start a discussion with the community and Treehouse staff.
Sign up
[MUSIC]
0:00
Hey there.
0:09
I'm Daniel,
a content producer here at Treehouse.
0:10
And in this short workshop we'll go
over ChatGPT's code interpreter model.
0:12
This model is super powerful.
0:17
It's a secure sandbox Python
interpreter perfect for testing code.
0:19
Debugging, data analysis,
and visualization.
0:23
This model is ideal for developers,
data analysts, and anyone eager to enhance
0:26
their Python programming skills and
data interpretation proficiency.
0:31
Let's hop in and get started.
0:36
To use the Code Interpreter model,
you must first be a ChatGPT Plus member.
0:37
Once you're signed into an account
with a Plus membership,
0:42
you will go down to the bottom
left-hand corner of your screen and
0:45
click on the three dots adjacent
to your profile picture and name.
0:48
After clicking on the three dots,
you will click on Settings.
0:52
Once you're in the Settings menu,
0:55
click on Beta features from
the menu on the left-hand side.
0:56
Now ensure you have the Code Interpreter
toggle enabled, perfect.
0:59
You're now ready to get started
with the Code Interpreter.
1:03
Warning, ChatGPT vs code interpreter has
the same limit as the general GPT for
1:06
model, which is 25
messages in three hours.
1:11
Because this model works so
well in a conversational format,
1:15
you can quickly hit this limit when
working with the code interpreter.
1:17
I just wanted you to keep this
in mind as we get started.
1:21
Before we begin,
first we'll need some test data.
1:24
I'm getting mine from kaggle.com.
1:27
Let's use this electric vehicle
data from 1997 to 2024.
1:30
We'll upload this data to ChatGPT
using the plus icon to the left.
1:40
Please note that this icon for
1:44
uploading is only available in
the Code Interpreter model.
1:46
Let's ask ChatGPT to analyze the data
we provided and to give us a summary.
1:50
I'll click on the toggle for
ChatGPT to show its work, and
1:59
we'll see the Python it's
producing in real time.
2:02
Now ChatGPT begins to provide an overview
of what type of data belongs to each
2:05
column.
2:10
We can see the data includes the county,
city, state, and
2:11
postal code to show where these
electric vehicles are located
2:16
as well as the make, model,
year of release and tons more.
2:22
ChatGPT continues on by providing
a summary of the data overall.
2:28
It tells us the data has
over 134,000 entries.
2:32
Holy smokes, that's one large doc.
2:36
The data shows there
are 39 unique counties,
2:38
the most frequent being King County.
2:41
As well as 454 unique cities,
with Seattle being the most frequent.
2:44
There are also 125 unique
models of electric vehicles,
2:49
with the Tesla Model 3 being the most
frequent model, interesting.
2:52
Now I'll ask ChatGPT what else
it can tell us about the data.
3:01
I'll tell it to include trends,
patterns and overall interesting results.
3:05
Let's see what it comes up with.
3:09
It recommends a few subsets of data
such as vehicle count by make,
3:11
vehicle count by model, year, and more.
3:15
I'll continue to have
ChatGPT show its work so
3:18
we can see how it uses Python to
plot out these visualizations.
3:20
Just like that we have our visualizations.
3:42
As you can see,
3:46
each of the five visualizations produced
comes with detailed explanations.
3:47
This is incredibly handy.
3:52
Now let's review the results.
4:06
Our first chart shows the top 10 vehicle
makes visualized with a bar chart,
4:08
followed by the vehicle
count by model year.
4:12
This is the amount each model sold
in the year it first debuted.
4:15
Next we have battery-powered vehicles
versus the plug-in hybrid models.
4:22
We can clearly see the battery version is
a favorite amongst consumers according to
4:26
this data.
4:30
Our next graph shows the leaders
in range sorted by make.
4:31
Tesla seem to have the most
range at the moment,
4:35
which could be a contributing
factor in its popularity.
4:37
Lastly, we have the top 10 counties
with the most electric vehicles.
4:40
All of these visualizations could
be useful in the right situation,
4:44
and they were literally done for
us in minutes.
4:48
And for our next example,
we'll head back to kaggle.com/datasets.
4:52
We'll click on All datasets to get
a list of the available datasets.
4:56
Let's use this Amazon
consumer behavior dataset.
5:04
We'll download the zip
file to our computer.
5:07
I'm gonna name the file Amazon data for
our purposes.
5:10
Now back to ChatGPT.
5:14
We'll toggle GPT-4 and
click on the Code Interpreter model.
5:15
Now we can upload our data.
5:19
I'm going to use the prompt,
please analyze the data and
5:23
generate a comprehensive summary.
5:26
Let's see how ChatGPT
handles the data this time.
5:28
First ChatGPT takes
a moment to unzip the file.
5:36
It extracts a CSV file named
Amazon Consumer Behavior Survey,
5:39
it then loads the CSV into a panda's
data frame to examine the data.
5:43
ChatGPT then starts off its analysis by
explaining there are 23 different columns,
5:50
such as timestamp or
when the survey was conducted, age and
5:54
gender of the respondent,
purchase frequency, purchase categories,
5:58
browsing frequency, and much more.
6:02
Now I wanna see if we are able
to conduct a sentiment analysis.
6:08
Let's see what ChatGPT says.
6:11
So the data we have is in the incorrect
format for proper sentiment analysis.
6:19
However, it seems we can run an indirect
form of sentiment analysis using
6:24
the service appreciation and
improvement area columns.
6:28
I'll take a quick look over the available
columns to ensure ChatGPT shouldn't
6:32
include any other columns.
6:35
Okay, I think those two are fine for
our test case.
6:51
I'll ask ChatGPT to run
the sentiment analysis.
6:54
It confirms and warns that the dataset
doesn't contain the needed fields, but
7:03
it will run the indirect analysis.
7:07
However, this time it suggests two
new columns to determine the positive
7:09
sentiment indicators,
the shopping satisfaction scores and
7:13
areas of service that
customers appreciate.
7:16
For the negative sentiment,
7:19
it still suggests using
the improvement areas column.
7:21
We'll let ChatGPT just use these columns
in the data to see what it comes up with.
7:24
ChatGPT starts by
producing two bar graphs.
7:29
The first was created using
the shopping satisfaction score, and
7:32
the second is based on the service
appreciation responses.
7:35
Next, it creates one more bar graph
this time using the improvement area
7:38
responses column.
7:42
Now that it's done, let's take
a better look at the visualizations.
7:46
We can see in our first graph that the
average shopping satisfaction score given
7:50
was a 3 out of 5.
7:54
This is interesting because in
the summary below, ChatGPT said it was 4.
7:55
This is why we have to be sure to fact
check our ChatGPT results as the chatbot
8:00
isn't always 100% accurate.
8:04
Our second graph seemed to have a wide
variety of answers, so much so some of
8:07
the response themes are overlapping as
there isn't enough room for everything.
8:10
Here, we could ask ChatGPT for something
like the top five service appreciation
8:14
responses for a cleaner visualization.
8:18
In the summary ChatGPT explains
that competitive prices is the most
8:20
common service appreciation response
next to wide product selection,
8:25
and from the graph,
we can see this is true.
8:29
In our last graph we can see the most
common improvement area responses which
8:32
were product quality and accuracy,
as well as reducing packaging waste.
8:36
All right, now we'll ask ChatGPT for
a breakdown of the audience
8:47
by requesting both age and gender
statistics as it pertains to our data.
8:50
We get two visualizations side-by-side.
9:02
The age distribution plot on the left
shows that the largest segment of our
9:04
respondents were in their mid-20s, while
the data on the right shows a majority of
9:08
our respondents identified as female.
9:12
Again, pretty interesting.
9:15
Now, I think we'll finish
by just asking for
9:21
ChatGPT to provide some general insights
from the data, both positive and
9:23
negative, as well as how
customers feel overall.
9:27
I'll let ChatGPT produce the results
first, and then we'll review them.
9:30
All right, let's begin at the top, ChatGPT
starts off with an audience breakdown that
9:44
explains the majority of survey
respondents are in their early to mid 20s.
9:49
As well as there are a majority
of female survey respondents.
9:53
This adds up with what
we reviewed earlier.
9:57
Next, it touched on our
indirect sentiment analysis,
9:59
starting with the positive sentiments.
10:02
ChatGPT explains respondents generally
have a positive shopping experience on
10:05
Amazon and they appreciate
Amazon's competitive pricing and
10:09
wide product selection.
10:12
For the negative sentiments, ChatGPT
mentions that the main areas customer seek
10:14
improvements are in product quality,
and accuracy.
10:19
Reducing package waste is also a concern
for a significant number of customers.
10:23
It finished the overview by saying the
sentiment seems to be generally positive,
10:26
however, there are areas
that could use improvement.
10:30
And again, it mentions we didn't have
the correct data for a proper analysis and
10:34
instructs us on steps we can
take to improve our results.
10:38
Even with the occasional error thrown by
ChatGPT, it was generally accurate and
10:41
useful in helping us understand the data.
10:45
I have one more idea to
really push ChatGPT.
10:48
Let's see if we can request two
visualizations, one showcasing the top
10:50
five positive responses and one showcasing
the top five negative responses.
10:54
Except this time make the graphs in an
HTML doc that will allow us to hover over
10:58
each bar for a pop up with more
information on that data point.
11:02
ChatGPT reminds us we don't have any
free-form text fields that could include
11:06
customer reviews or feedback since
the survey wasn't structured in this way.
11:10
It says we can provide the data now,
or proceed by generating
11:15
the visualizations using aggregated
responses from the survey.
11:18
Let's see how this will look.
11:22
Okay, wow.
11:31
So from the looks of things ChatGPT
has two HTML files we can download.
11:32
I'll download each to my computer and
open them up so we can take a look.
11:36
First we have the top five
service appreciation responses.
11:41
Product recommendations was the top
result, followed by competitive prices and
11:44
wide product selection and so on.
11:48
But this time if we
hover over each result,
11:51
we get the exact count of respondents
who had a similar answer.
11:53
With product recommendations being
mentioned by 185 respondents,
11:56
competitive price by 182,
wide product selection had 150,
12:01
80 respondents mentioned Amazon
having a friendly site/app.
12:06
And this last result has one user,
and the text label is a dot.
12:10
So this could be an error or
an incorrectly filled out submission.
12:14
Let's check out our second visualization.
12:18
Okay, we have a similar plot, but
this time for the top five improvement
12:21
areas which can also be seen as
negatives Amazon needs to work on.
12:24
And just like before we can hover
to get the count of respondents.
12:27
Customer support responsiveness
has the most counts at 217.
12:31
Product quality and accuracy has 159.
12:35
Followed by reducing
package waste with 133,
12:39
79 respondents mentioned shipping speed
and reliability, and the least mentioned
12:42
of the top five was quality of product
is poor according to the big offers.
12:47
I'm not certain, but I think this means
respondents felt highly discounted and or
12:52
promoted products weren't as
high quality as advertised.
12:56
ChatGPT was able to produce HTML coded
visualizations that we could embed in
13:00
a project, talk about amazing.
13:04
And there you have it, folks.
13:06
That's how powerful and versatile
ChatGPT's Code Interpreter model can be.
13:08
From in depth data analysis and
visualization to sentiment analysis and
13:13
creating interactive HTML
based visualizations,
13:16
we've just scratched
the surface of what's possible.
13:19
So remember, as a developer or data
analyst, the sky's the limit in leveraging
13:23
this model to simplify your work and
achieve faster, more insightful results.
13:27
This has been Daniel,
your content producer at Treehouse.
13:32
Keep experimenting, keep coding, and
let's unlock more insights together.
13:35
See you in the next workshop.
13:39
You need to sign up for Treehouse in order to download course files.
Sign upYou need to sign up for Treehouse in order to set up Workspace
Sign up