Application projects
Application projects.
Application projects take you from a question to a model, then to a web application. This page introduces the first project in the learning path; more projects will follow. You can start without knowing how to program.
- 4 stages
- Python · VS Code · Jupyter
- Beginner level
The question
First project: estimating an exam score
To learn AI without losing your way, the learning path follows one specific question. Each concept meets a project need: read data, understand it, build an estimate, then check how reliable it is.
We will start with information available before an exam: study time, attendance and previous results. The score to estimate is kept separate from these inputs.
Can we estimate an exam score from study habits and the learning context?
The destination
What we will build
A small web application: you enter information such as study time, attendance and previous results. The model returns an estimated score with its limitations.
Information before the exam
The model’s response
The factors with the greatest influence
Educational demonstration: do not use it to make decisions about a real person.
The path
Four stages, one piece at a time
Each stage adds a piece to the project. Lessons are published one at a time, in order.
Understand AI and set up your tools
Core concepts, Python, VS Code and Jupyter, followed by data exploration.
The project gains: data opened and described in a notebook.
In progress The curriculumA working model
Prepare inputs, train, compare with a baseline, measure errors and examine bias.
The project gains: an evaluated model saved to a file.
Coming soonThe web application
A first application in Python, then an API and a web page that calls it.
The project gains: the model responds through a web page.
Coming soonDeep learning
A neural network trained on the same data and compared with the first models.
The project gains: a second model to compare.
Coming soon
The data
The project data
The Student Performance Factors workbook contains 6,607 rows and 20 columns. Each row combines factors related to study habits and the learning context with an exam score, Exam_Score. A second sheet describes the fields.
Reading the workbook comes after Python basics, in the lesson about Excel files.
Source: Student Performance Factors, lainguyn123, on Kaggle. Column names are kept in English. Some variables contain missing values; the lessons will explain how to handle them and discuss the data’s limitations.
- Hours_Studied
- Input: study hours per week
- Attendance
- Input: percentage of classes attended
- Previous_Scores
- Input: previous exam results
- Gender
- Sensitive attribute: reserved for examining bias
- Exam_Score
- Target: exam score to estimate
The tools
Three free tools
They are installed once, in lesson 2, then used throughout the learning path.
- 1
Python
The language used in the examples. Libraries are introduced when needed: pandas, matplotlib and scikit-learn.
- 2
VS Code
The code editor, with Microsoft’s Python and Jupyter extensions.
- 3
Jupyter
Notebooks
.ipynbopened in VS Code: text and executable code, one cell at a time.
Key ideas
The essentials, before you write code
These ideas return in every lesson. They give you a critical starting point for reading an AI result.
The learning path’s ten concepts
- Data is read as a table: one row per observation, one column per piece of information.
- Distinguish the model’s inputs from the target: what you want to predict.
- Training a model is different from using it to predict.
- Evaluate a model on data it has never seen.
- Compare it with a simple baseline, such as the mean score calculated using training data only.
- A predicted score is an estimate, subject to errors and uncertainty.
- Information from the future must never enter the model.
- Different errors have different costs.
- A model can be unfair: check for this.
- A trained model is a file that a web application can use.
The project rules
- Inputs contain only information available before the exam; Exam_Score is the target.
- Evaluation data stays separate during training and parameter selection.
- Gender, family income and learning disabilities are reserved for examining bias, and are not used as application inputs.
- No result is reported before it has been calculated.
- The application is a demonstration, not a decision-making tool.
Ready to begin?
The first lesson introduces the language of AI without installing anything. The second prepares Python, VS Code and Jupyter.