This repository contains my submission for Research Task 05: Descriptive Statistics and LLMs. The goal of this task was to analyze a small dataset using descriptive statistics and compare that with how a Large Language Model (ChatGPT) interprets and answers natural language questions based on the same data.
| File Name | Description |
|---|---|
Premier_League_bar_Goals_Home.png |
Bar chart of "Goals Home" column (top 10 values) |
Premier_League_bar_Away_Goals.png |
Bar chart of "Away Goals" column |
Premier_League_bar_home_possessions.png |
Bar chart of "home_possessions" column |
Premier_League_bar_away_possessions.png |
Bar chart of "away_possessions" column |
prompt_log.txt |
Log of LLM prompts and responses |
qa_results.xlsx |
Manual evaluation comparing LLM answers with Python-generated answers |
questions.txt |
10 natural language questions used to prompt the LLM |
summary_without_grouping_Premier_League.xls |
Descriptive statistics summary (first 10 rows of dataset, no grouping) |
Task_05_Descriptive_stats.ipynb |
Python scripts used (first 10 rows of dataset, no grouping) |
- Dataset Used: Premier League 2022–23 season (not uploaded per assignment instructions)
- Scope: Used only the first 10 rows to fit within LLM input limits
- Grouping: No grouping used in this task; grouping by
stadiumwill be explored in Task 06 - Model Used: ChatGPT (GPT-4)
- Descriptive Statistics:
- Calculated mean, min, max, standard deviation, count, and most frequent value
- Generated bar plots for 4 selected numeric columns
- Prompt Design:
- Created 10 questions across 5 categories (Easy to Coaching Insight)
- Used Python for validating answers to questions 5–8
- Questions 9–10 are narrative and answered only by the LLM
- Answers to questions 1–4 were verified using the output from the summary file.
- Python scripts were written to calculate true answers for questions 5–8.
- LLM responses were evaluated manually and logged in
qa_results.xlsx. - Narrative insight questions (9–10) were not evaluated numerically.