About the Authors

Thomas Carroll is currently a professional software developer. After spending

almost a decade working as a teacher in a prestigious school in his hometown, he
decides to devote herself to his greatest passion, programming. Having dedicated
his initial working life to teaching, he decided, in addition to teaching through
private courses, to write some programming books for beginners. If you want to
know more about programming and machine learning I can help you.

Table of Contents

Table of Contents
About the Authors
Table of Contents
Automate ChatGPT Prompts for Data Science with Python
Chapter 1: ChatGPT Prompts for Data Science (Tried, Tested, and Rated)
Automate data science tasks with ChatGPT
How to Read this
Table of Contents
Write python
Explain code
Optimize code
Format code
Translate code from one language to another
Explain concepts
Suggest ideas
Troubleshoot problem
 Write SQL
Write other Code
Miscellaneous
Chapter 2: WRITE PYTHON
1. Train Classification Model
2. Automatic Machine Learning
3. Tune Hyperparameter
4. Explore Data
5. Generate Data
6. Write Regex
7. Train Time Series
8. Address Imbalance Data
9. Get Feature Importance
10. Visualize Data with Matplotlib
11. Visualize Image Grid Matplotlib
12. Explain Model with Lime
13. Explain Model with Shap
14. Write Multithreaded Functions
15. Compare Function Speed
16. Create NumPy Array
17. Write Unit Test
18. Validate Column
Chapter 3: EXPLAIN CODE
19. Explain Python
20. Explain SQL
21. Explain Google Sheets Formula
Chapter 4: OPTIMIZE CODE
22. Improve Code Speed
23. Optimize Pandas
24. Optimize Pandas Again
25. Optimize Python
26. Optimize SQL
27. Simplify Python
Chapter 5: FORMAT CODE
28. Write Documentation
29. Improve Readability
30. Format SQL
Chapter 6: TRANSLATE CODE
31. Translate Between DBMS
32. Translate Python to R
33. Translate R to Python
Chapter 7: EXPLAIN CONCEPTS
34. Explain to Five-Year-Old
35. Explain to Undergraduate
36. Explain to Professor
37. Explain to Business Stakeholder
38. Explain Like StackOverflow
Chapter 8: SUGGEST IDEAS
39. Suggest Edge Cases
40. Suggest Dataset
41. Suggest Portfolio Ideas
42. Suggest Resources
43. Suggest Time Complexity
44. Suggest Feature Engineering
45. Suggest A/B Testing Steps
46. Career Coaching
Chapter 9: TROUBLESHOOT PROBLEM
47. Correct Own ChatGPT Code
48. Correct Python Code
49. Correct SQL Code
50. Troubleshoot PowerBI Model
Chapter 10: WRITE SQL
51. Create Running Average
52. Solve Leetcode Question
Chapter 11: WRITE OTHER CODE
53. Write Google Sheets Formula
54. Write R
55. Write Shell
56. Write VBA
Chapter 12: MISCALLANEOUS
57. Format Tables
58. Summarize Book
59. Summarize Paper
60. Provide Emotional Support (My Favourite)
Closing

Automate ChatGPT Prompts for Data

Science with Python
Chapter 1: ChatGPT Prompts for
Data Science (Tried, Tested, and
Rated)
Automate data science tasks with ChatGPT

I rated 60 ChatGPT functions for Data Science. Use these prompts and ask
ChatGPT to write, and explain code, optimize data science code. It can also
explain data science concepts, suggest ideas, and troubleshoot problems.

How to Read this

The table of content contains all 60 prompts. Click on the links in the table of
content to navigate to the prompt, ChatGPT’s response and the rating.

Table of Contents
Write python
Train Classification Model
Automatic Machine Learning
Tune Hyperparameter
Explore Data
Generate Data
Write Regex
 Train Time Series
Address Imbalance Data
Get Feature Importance
Visualize Data with Matplotlib
Visualize Image Grid Matplotlib
Explain Model with Lime
Explain Model with Shap
Write Multithreaded Functions
Compare Function Speeds
Create NumPy Array
Write Unit Test
Validate Column

Explain code
Explain Python
Explain SQL
Explain Google Sheets Formula

Optimize code
Improve Code Speed
Optimize Pandas
Optimize Pandas Again
Optimize Python
Optimize SQL

Format code

Simplify Python
Write Documentation
 Improve Readability
Format SQL

Translate code from one language to another

Translate Between DBMS
Translate Python to R
Translate R to Python

Explain concepts
Explain to Five-Year-Old
Explain to Undergraduate
Explain to Professor
Explain to Business Stakeholder
Explain Like Stackoverflow

Suggest ideas
Suggest Edge Cases
Suggest Dataset
Suggest Portfolio Ideas
Suggest Resources
Suggest Time Complexity
Suggest Feature Engineering
Suggest A/B Testing Steps
Career Coaching

Troubleshoot problem
Correct ChatGPT Code
Correct Python Code
 Correct SQL Code
Troubleshoot PowerBI Model

Write SQL
Create Running Average
Solve Leetcode Question

Write other Code

Write Google Sheets Formula

Write R
Write Shell
Write VBA

Miscellaneous
Format Tables
Summarize Book
Summarize Paper
Provide Emotional Support
Chapter 2: WRITE PYTHON
1. Train Classification Model

Prompt: I want you to act as a data scientist and code for me. I have a dataset
of [describe dataset]. Please build a machine learning model that predict [target
variable].
ChatGPT is able to provide satisfactory code that:

provide the regular preprocessing (scaling the data and encoding

categorical variable),
drop the testing column
provide the train-test split
evaluate the model

This is good because it provides enough code to start off a model. It is also
technically sound and correct. However, it’s lacking because it only uses
accuracy for evaluation.

2. Automatic Machine Learning

Prompt: I want you to act as an automatic machine learning (AutoML) bot using
TPOT for me. I am working on a model that predicts [...]. Please write python
code to find the best classification model with the highest AUC score on the test
set.
ChatGPT provided good code (it runs!) that:

provide the train-test split

instantiated the AutoML classifier
fit it to the AutoML classifier
export the classifier

I’m not at all surprised that ChatGPT excels here since the code is pretty much
only boilerplate code. This can be copied and pasted from the documentation as
is.

3. Tune Hyperparameter

Prompt: I want you to act as a data scientist and code for me. I have trained
a [model name]. Please write the code to tune the hyper parameters.
For a decision tree classifier, ChatGPT wrote code that:

performed a grid search

suggested the best parameters.

Not bad at all, and the code runs! Grid search is the most popular method for
hyperparameter tuning, so it’s no surprise that it came up on top.

However, note that grid search goes through the search space exhaustively. The
larger the search space, the longer time it will take. An alternative is the
randomized search, which implements a randomized search over parameters.
Once assigned a budget, randomized search takes the same amount of time
regardless of the search space.

4. Explore Data

Prompt: I want you to act as a data scientist and code for me. I have a dataset
of [describe dataset]. Please write code for data visualisation and exploration.
For an exploration, ChatGPT wrote code that:

used df.describe() to provide descriptive statistics

created count plots for visualizing the distribution of categorical
variable
created a heatmap to determine the between numerical variables
determined the relationship of numerical variables and the target
variable with box plots of numerical columns
determined the relationship of categorical variables and the target
variable with barplot.

The above choices of charts are indeed correct.

5. Generate Data

Prompt: I want you to act as a fake data generator. I need a dataset that has x rows
and y columns: [insert column names]
ChatGPT produced code that generated the data as per my requirements
completely. It only used the pandas library and the inhouse random library,
making it portable. I actually did use the synthetic dataset for other tasks.

6. Write Regex

Prompt: I want you to act as a coder. Please write me a regex in Python

that [describe regex]
Here, ChatGPT provided code that matched a regex as per requirements in
Python.

However, note that I have cherrypicked this example. Before this, I had tried
asking ChatGPT to create more complex other regex patterns but it failed. It even
sounded confident when it explained the regex, but the regex it produced didn’t
work.

7. Train Time Series

Prompt: I want you to act as a data scientist and code for me. I have a time series
dataset [describe dataset]. Please build a machine learning model that
predict [target variable]. Please use [time range] as train and [time range] as
validation.
ChatGPT provided code that:

preprocessed the data using standard scaling

Created lag variables
Trained a RandomForestRegressor for the evaluation
Used Mean Absolute Error for evaluation

Alright, here’s a controversial one. ChatGPT is not able to properly write code for
time series validation.

Here’s why: Without the prompt “please use data from A to B as training. Please
use data from Y to Z as validation”, ChatGPT uses sklearn’s train_test_split to
randomly splits the data points into train and validation. This is a huge no-no
because the train_test_split function randomly splits the data with no regard for
the temporal relationship between the datasets. Instead, it should have used all
data points from an earlier timeframe as train and later timeframe as test.

8. Address Imbalance Data

Prompt: I want you to act as a coder. I have trained a machine learning model on
an imbalanced dataset. The predictor variable is the column [Insert column
name]. In python, how do I oversample and/or undersample my data?
ChatGPT used the imblearn library to write boilerplate code that randomly under
and oversamples the dataset.

The code is sound, but I would nitpick on its understanding of over and
undersampling.

Undersampling and oversampling should only be done on the train dataset. It

should not be done on the entire dataset, which includes the test dataset.
Otherwise, the evaluation will be done on a resampled test set that no longer
reflects the actual distribution of the real dataset.

Thus, it should have attempted to split the dataset into train and test sets first,
before performing the under and oversampling.

9. Get Feature Importance

Prompt: I want you to act as a data scientist and explain the model’s results. I
have trained a decision tree model and I would like to find the most important
features. Please write the code.

ChatGPT correctly provided the code for plotting the feature importance of a
decision tree. It even ranked the feature importances from the most important to
the least important. Kudos!

10. Visualize Data with Matplotlib

Prompt: I want you to act as a coder in python. I have a dataset [name] with
columns [name]. [Describe graph requirements]
ChatGPT could provide the correct code for plotting visualizations in Matplotlib.

I am particularly impressed because I provided very long and rather convoluted

instructions, and ChatGPT followed them to a tee. Here is the output from the
code:
11. Visualize Image Grid Matplotlib

Prompt: I want you to act as a coder. I have a folder of images. [Describe how
files are organised in directory] [Describe how you want images to be printed]
Again, ChatGPT can provide the correct code for plotting grid of images with
Matplotlib.

Again, I provided convoluted and long instructions, but ChatGPT could follow
them to a tee. (I personally loathe creating charts with Matplotlib. It is extremely
flexible but writing code in it is rather painful.)

12. Explain Model with Lime

Prompt: I want you to act as a data scientist and explain the model’s results. I
have trained a [library name] model and I would like to explain the output using
LIME. Please write the code.
ChatGPT provided code for explaining the model with LimeTabularExplainer
and explained an instance with the correct commands.

Personally, I find this helpful because I do not need to refer to the documentations
for the boilerplate code. Overall, ChatGPT provided accurate code that does not
violate any principles.

13. Explain Model with Shap

Prompt: I want you to act as a data scientist and explain the model’s results. I
have trained a scikit-learn XGBoost model and I would like to explain the output
using a series of plots with Shap. Please write the code.
14. Write Multithreaded Functions
Prompt: I want you to act as a coder. Can you help me parallelize this code across
threads in python?
I’m pleasantly surprised to find that this code actually works without any
modifications. I can foresee myself doing this often in the future.

15. Compare Function Speed

Prompt: I want you to act as a software developer. I would like to compare the
efficiency of two algorithms that performs the same thing in python. Please write
code that helps me run an experiment that can be repeated for 5 times. Please
output the runtime and other summary statistics of the experiment. [Insert
functions]
ChatGPT is able to write code that compares two function speeds with built-in
packages.

The code runs and is correct. Though, I would rather ChatGPT use
the %%timeit magic in Python instead of writing verbose code.

16. Create NumPy Array

Prompt: I want you to act as a data scientist. I need to create a numpy array. This
numpy array should have the shape of (x,y,z). Please initialize the numpy array
with random values.

This simple command results in a simple function. Here, ChatGPT demonstrated

its ability to translate human language into an appropriate function and parameter.

The code is correct, and I have zero complaints here.

17. Write Unit Test

Prompt: I want you to act as a software developer. Please write unit tests for the
function [Insert function]. The test cases are: [Insert test cases]

18. Validate Column

Prompt: I want you to act as a data scientist. Please write code to test if that my
pandas Dataframe [insert requirements here]
Chapter 3: EXPLAIN CODE
19. Explain Python

Prompt: I want you to act as a code explainer. What is this code doing? [Insert
code]

I am extremely pleased with ChatGPT’s response with this. Not only is it able to
explain concepts for each function. It is also able to guess why the programmer is
performing a certain action. For example, it explained the
line np.clip(reconstructed, 0, 1) by saying that “this is likely done to ensure that
the values are within the valid range for pixel intensity value”, which is
absolutely correct.

ChatGPT’s ability to understand code is pretty good and robust, in my opinion.

For that, I would strongly suggest any coders who face problems understanding a
code use ChatGPT as a complement to the API’s documentation.

20. Explain SQL

Prompt: I want you to act as a data science instructor. Can you please explain to
me what this SQL code is doing? [Insert SQL code]
Again, a correct and an awesome explanation. Much like its explanations for
Python, ChatGPT is able to understand and explain SQL code.

I like its explanation of the line SUM(Revenue) OVER (ROWS BETWEEN

UNBOUNDED PRECEDING AND CURRENT ROW). Note that it’s not only
able to provide a high level explanation of what the line is doing, but also provide
a breakdown of what the OVER(),UNBOUNDED PRECEDING and CURRENT
ROW clauses are doing.

21. Explain Google Sheets Formula

Prompt: I want you to act as a Google Sheets formula explainer. Explain the
following Google Sheets command. [Insert formula]
ChatGPT is able to explain the FILTER function well. I like that it provided the
generalize code snippet like one would see in the documentation, and proceeded
to explain the general snippet with the specific arguments provided by the users.

In particular, the line the "Filter" function is being used to filter the range
"C2:C12" based on the condition "A2:A12=F2" is an excellent explanation is
contextualized to the user’s arguments.
Chapter 4: OPTIMIZE CODE
22. Improve Code Speed

Prompt: I want you to act as a software developer. Please help me improve the
time complexity of the code below. [Insert code]

A correct optimization made by ChatGPT from O(n) to O(1). However, note that
this is a simple example. You might want to test it on more complex cases to see
if this holds up.
23. Optimize Pandas

Prompt: I want you to act as a code optimizer. Can you point out what’s wrong
with the following Pandas code and optimize it? [Insert code here]
I personally think these suggestions are particularly salient. It pointed out that
there is nothing wrong with the code (a correct observation) and moved to to
suggest good code practices.

24. Optimize Pandas Again

Prompt: I want you to act as a code optimizer. Can you point out what’s wrong
with the following Pandas code and optimize it? [Insert code here]
Of course ChatGPT is able to catch iterrow as an anti-pattern in Pandas. Love this
optimization.

25. Optimize Python

Prompt: I want you to act as a code optimizer. The code is poorly written. How
do I correct it? [Insert code here]
26. Optimize SQL

Prompt: I want you to act as an SQL code optimizer. The following code is slow.
Can you help me speed it up? [Insert SQL]
The first two suggestions here are valid suggestions for optimization. Howeve,
the third tip is incorrect. In fact, DATEPART is actually faster than EXTRACT.
For that reason, I’ll have to dock stars off for this category.

27. Simplify Python

Prompt: I want you to act as a code simplifier. Can you simplify the following
code?
If you’ve written spaghetti code in Python, ChatGPT might be able to help you
simplify it.
Chapter 5: FORMAT CODE
28. Write Documentation

Prompt: I want you to act as a software developer. Please provide documentation

for func1 below. [Insert function]

ChatGPT could provide a correct, nicely formatted docstring. It also provides a

one-liner that captures the essence of the function (“calculates the sum of the
integers”). This is neat and I foresee myself doing this a lot.
29. Improve Readability

Prompt: I want you to act as a code analyzer. Can you improve the following
code for readability and maintainability? [Insert code]

I am slightly disappointed here because I had hope ChatGPT would help me

remove the redundant variables in the function. That said, its suggestions already
did help with readability.

30. Format SQL

Prompt: I want you to act as a SQL formatter. Please format the following SQL
code. Please convert all reserved keywords to uppercase [Insert
requirements]. [Insert Code]
The code is formatted nicely and neatly with the correct indentation and casing.
Chapter 6: TRANSLATE CODE
31. Translate Between DBMS

Prompt: I want you to act as a coder and write SQL code for MySQL. What is the
equivalent of PostgreSQL’s DATE_TRUNC for MySQL?

While ChatGPT is able to translate the DATE_TRUNC function in this example,

it failed on other instances. In particular, it failed to translate the Redshift’s
PIVOT into a query that contains a series of CASE WHEN. Thus, this translation
might work well only for simple use cases.
32. Translate Python to R

Prompt: I want you to act as a code translator. Can you please convert the
following code from python to R? [Insert code]

This is a correct translation! However, note that this is a very straightforward

function. ChatGPT might fail on on more complex code blocks.

33. Translate R to Python

Prompt: I want you to act as a code translator. Can you please convert the
following code from R to python? [Insert code]
Again, a correct translation here. However, note that this is a very straightforward
function. ChatGPT might fail on on more complex code blocks.
Chapter 7: EXPLAIN CONCEPTS
Overall, I love ChatGPT’s ability to explain concepts correctly with an awareness
of the audience. Through the examples below, you can see how ChatGPT is able
to explain different concepts to audiences of different caliber. When explaining a
difficult concept to a child, it uses an appropriate analogy. When explaining the
concept to a professor, it uses the appropriate tone and language. When
explaining the concept to a business stakeholder, it drew parallels to the business
jargons and contexts. five stars for each of this.

34. Explain to Five-Year-Old

Prompt: I want you to act as a data science instructor. Explain [concept] to a five-
year-old.
35. Explain to Undergraduate

Prompt: I want you to act as a data science instructor. Explain [concept] to an

undergraduate.
36. Explain to Professor

Prompt: I want you to act as a data science instructor. Explain [concept] to a

professor.
37. Explain to Business Stakeholder

Prompt: I want you to act as a data science instructor. Explain [concept] to a

business stakeholder.
38. Explain Like StackOverflow

Prompt: I want you to act as an answerer on StackOverflow. You can provide

code snippets, sample tables and outputs to support your answer. [Insert technical
question]
Not only is the answer correct and succinct, it also came with relevant
information (the different types of constraints) and the code snippets.

One tip that I have here is to explicitly prompt ChatGPT for code snippets,
sample tables and outputs for a more comprehensive answer.
Chapter 8: SUGGEST IDEAS
39. Suggest Edge Cases

Prompt: I want you to act as a software developer. Please help me catch edge
cases for this function [insert function]
For a relatively simple function, ChatGPT is able to correctly and exhaustively
suggests the edge cases. It exhaustively considered all cases without being
prompted on what the cases are.

40. Suggest Dataset

Prompt: I want you to act as a data science career coach. I want to build a
predictive model for [...]. At the same time, I would like to showcase my
knowledge in [...]. Can you please suggest the five most relevant datasets for my
use case?
I like ChatGPT’s dataset suggestions because they are customized to my use case.
I have also tested this prompt for different personas and use cases, and the
suggestions it gives are consistently excellent.

However, note that I have only tested relatively simple use cases. For more
complex use cases that requires niche datasets, ChatGPT might not do well.

41. Suggest Portfolio Ideas

Prompt: I want you to act as a data science coach. My background is in [...] and I
would like to [career goal]. I need to build a portfolio of data science projects that
will help me land a role in [...] as a [...]. Can you suggest five specific portfolio
projects that will showcase my expertise in [...] and are of relevance
to [company]?

This is one of my favourite prompts. Its suggestions are relevant not only to
myself, but also to the industry that I am gunning for, and also for my dream
company. It does not suggest any generic projects like the Titanic or Iris project.
Instead, it makes salient suggestions based on its knowledge of the company
(Tesla, in this case), the industry (data science), and my background in
engineering.

I love this feature and I hope more people will try it out!

42. Suggest Resources

Prompt: I want you to act as a data science coach. I would like to learn
about [topic]. Please suggest 3 best specific resources. You can include [specify
resource type]

This one is for those looking to learn a new skill. I find that the resources that it
has suggested are not only relevant, but also critically acclaimed. It also does not
provide an overwhelming list of resources and instead provide only a few. I like
this because it certainly provides a good starting point for anyone hunting for new
resources.

43. Suggest Time Complexity

Prompt: I want you to act as a software developer. Please compare the time
complexity of the two algorithms below. [Insert two functions]
For this very simple example, ChatGPT is able to deduce the time complexity.

44. Suggest Feature Engineering

Prompt: I want you to act as a data scientist and perform feature engineering. I
am working on a model that predicts [insert feature name]. There are
columns: [Describe columns]. Can you suggest features that we can engineer for
this machine learning problem?
I absolutely loved ChatGPT’s reponse here. The features it suggested are sound
and actually useful for the use case.

45. Suggest A/B Testing Steps

Prompt: I want you to act as a statistician. [Describe context] Please design an

A/B test for this purpose. Please include the concrete steps on which statistical
test I should run.
While ChatGPT’s steps for A/B testing is correct. I did find it a requirement to
explicitly ask it to include the statistical test required.

46. Career Coaching

Prompt: I want you to act as a career advisor. I am looking for a role as a [role
name]. My background is [...]. How do I land the role and with what resources
exactly in 6 months?
I have no surprises that ChatGPT did well here. It is able to regurgigate oft-
repeated advice for career changers. While this is technically correct, I didn’t find
it overwhelming impressive.

Side note: One can combine this tip with Tip 41 (Suggest portfolio ideas) and 42
(Suggest dataset) when building one’s portfolio.
Chapter 9: TROUBLESHOOT
PROBLEM
47. Correct Own ChatGPT Code

Prompt: Your above code is wrong. [Point out what is wrong]. Can you try again?
ChatGPT is quick to admit its mistakes and correct its code. In the above
example, I asked ChatGPT to create a shell command. I realized that it does not
actually run, so I pointed out what was wrong. ChatGPT is able to provide the
correct code on the second attempt.

This is good enough if you know where the bug is exactly. But in cases where
you do not know where the bug is, ChatGPT might not be able to pinpoint where
the fault is. The onus is still on the coder to figure things out.

48. Correct Python Code

Prompt: I want you to act as a software developer. This code is supposed

to [expected function]. Please help me debug this python code that cannot be
run. [Insert function]
I provided ChatGPT with a simple incorrect function and it is able to tell me why
it is wrong correctly.

49. Correct SQL Code

Prompt: I want you to act as a SQL code corrector. This code does not run
in [your DBMS, e.g. PostgreSQL]. Can you correct it for me? [SQL code here]
A correct explanation for a simple query. However, upon further testing, I did
find ChatGPT’s understanding of complex SQL query to be somewhat lacking.
It’s not able to debug more complex SQL queries containing multiple tables.

50. Troubleshoot PowerBI Model

Prompt: I want you to act as a PowerBl modeler. Here is the details of my current
project. [Insert details]. Do you see any problems with the table?
ChatGPT is able to suggest relevant improvements to the PowerBI model. It also
admitted its shortcoming (“It’s difficult to determine if there are any problems
without more context”)

Mathias Halkjær Petersen had a good chat with ChatGPT on this topic. He
showed that ChatGPT is able to troubleshoot PowerBI models effectively when
more context is provided. It is also able to write the models relatively
independently.
Chapter 10: WRITE SQL
51. Create Running Average

Prompt: I want you to act as a data scientist and write SQL code for me. I have a
table with two columns [Insert column names]. I would like to calculate a running
average for [which value]. What is the SQL code that works for PostgreSQL 14?
ChatGPT did provide a good response to the requesst here! It correctly provided
the output for creating running averages in SQL. It also provided a correct
explanation and breakdown.

One tip that I have is to provide ChatGPT with the exact version and flavour of
SQL. This allows it to create versions of the data.

I would encourage practitioners to create test ChatGPT with their own SQL
questions and see if it creates a correct response.

52. Solve Leetcode Question

Prompt: Assume you are given the tables… with the columns… Output the
following…
ChatGPT could solve an introductory Leetcode question (Difficulty: Simple) on
SQL. The above code is sound and answers the question correctly.

However, it failed to produce runnable SQL code for a Leetcode question with
medium difficulty when I tested it.
Chapter 11: WRITE OTHER CODE
53. Write Google Sheets Formula

Prompt: I want you to act as a bot that generates Google Sheets formula. Please
generate a formula that [describe requirements]
ChatGPT excels at acting as a lookup table for all commands. As you can see, I
asked it for a function that pulls one sheet and puts it into another. It faithfully
provided the correct importrange.

However, again, I cherrypicked this example. ChatGPT fails when asked to

provide more complex formulas. In particular, I asked it to provide a formula that
outputs the number of Mondays in 2023. It failed to produce a formula that
works. Yet, it still confidently explained it as though it performed splendidly.

54. Write R

Prompt: I want you to act as a data scientist using R. Can you write an R script
that [Insert requirement here]
ChatGPT sure excelled at this basic R command. This code runs and indeed
creates the plot as requested.

However, do note that I did not test ChatGPT on R coding extensively. I can
imagine it failing or producing incorrect codes for more complex cases.

55. Write Shell

Prompt: I want you to act as a Linux terminal expert. Please write the code
to [describe requirements]
These are the correct commands for the requested instructions. It even provided a
line for granting the necessary permission to the file without me explicitly asking
it to

However, note that I cherrypicked this example out of all the prompts for writing
shell commands. In particular, when I asked ChatGPT to copy files from a folder
recursively, it gave an incorrect output that could not be executed. In its defense,
it’s able to correct its own code (See “47. Correct Own ChatGPT Code” above.)

56. Write VBA

Prompt: I want you to act as an Excel VBA developer. Can you write a VBA
that [Insert function here]?
Chapter 12: MISCALLANEOUS
57. Format Tables

Prompt: I want you to act as a document formatter. Please format the following
into a nice table for me to place in Google Docs? [insert text table here]

The output speaks for itself. It is able to convert a table that I copied from the
internet into a nicely formatted table.
However, ChatGPT cannot offer any customization to the table. I asked it to
change the colors and font, but it was not able to.

58. Summarize Book

Prompt: I want you to act as a technical book summarizer. Can you please
summarize the book [name] with 5 main points?

Knaflic’s book is a gold mine of data storytelling tips. I personally feel that the
summary was too generic and did not capture the tips sufficiently.

59. Summarize Paper

Prompt: I want you to act as an academic. Please summarise the paper [...] in
simple terms in one paragraph.

I asked ChatGPT to summarize a relatively old paper “Distilling the knowledge

in a neural network”. It is able to explain the most important concepts in a
relatively straightforward and jargon-free manner.

However, note that this is possible because enough people have written reviews
or summaries of the above paper. It will not be able to summarize papers that are
not as popular well.

Also, ChatGPT does not have an understanding of the world beyond 2021. Thus,
it will not be able to help you understand new papers.

60. Provide Emotional Support (My Favourite)

Prompt: I want you to provide emotional support to me. [Explain problem here.]
Alright I’m no psychologist, so I do not think I am qualified to judge this. That
said, I personally think that this is a response that is both empathetic and
practical. It also showed a good understanding of the data science process.

Closing
ChatGPT performs well in explaining things and providing boilerplate code.
However, it falters when it comes to debugging or creating complex code
snippets.

Will ChatGPT replace data scientists? No, I don’t think so. At least not yet.
However someone using ChatGPT might.