Machine learning project

Total 20pts.

Tasks:

1. Write a project proposal in ProjectProposal.md
2. Complete the project. In ProjectReadme.md include:

• Background on your code files
• How to run your code, guide to install any additional packages
• Results, interpretation and reflection

1. Proposal (4pts)

1. Why: Question/Topic being investigated 1pt
2. How: Plan of attack 1pt
3. What: Dataset, models, framework, components 2pts

Example 1

1. Why: Question being investigated 1pt

I would like to understand better how the decision tree algorithm works.

2. How: Plan of attack 1pt

In insert reference here a guide to implement decision trees from scratch is available. I will download the code, run with the tutorial dataset (very small) and then apply it to a real dataset (see below) and compare to scikit-learn. Furthermore, I will try and implement early stopping by introducting max_depth parameter. This is not part of the tutorial code and an extension.

3. What: Dataset, models, framework, components 2pts

• UCI dataset insert url here (13 features, 300 samples)
• Code from insert code github url
• Scikit-learn DecisionTreeClassifier

Example 2

1. Why: Question being investigated 1pt

I would like to see what it takes to participate in a Kaggle machine learning competition.

2. How: Plan of attack 1pt

I have created an account on Kaggle and they recommend starting with the Titanic classification problem. I will follow these guidelines to prepare a notebook and submit it. I will use the Lab3 notebook as a template.

If time premits, a second competition will be selected and prepared.

3. What: Dataset, models, framework, components 2pts

• Kaggle competition url insert url
• Scikit-learn classifiers: LogisticRegression, RandomForest, GradientBoosting, SVC
• Possibility to explore XGBoost library as an alternate classifier.

Example 3

1. Why: Question being investigated 1pt

I am interested in learning more about putting a model into production.

2. How: Plan of attack 1pt

Initial research showed, that mlflow insert url is a framework that allows for training and deploying machine learning models. On their website insert url here there are numerous tutorials. I am planning to follow the following:

• Tutorial 1 add description
• Tutorial 2 add description

Subsequently, I will adapt the tutorial code to create a website that allows entering Iris flower sepal and petal measurements, and the classifier displays the predicted type of Iris flower. To demonstrate this pipeline, only one classifier will be trained.

3. What: Dataset, models, framework, components 2pts

• mlflow url here with submodules module 1 module 2
• mlflow to serve the model as a RESTapi
• Scikit-learn Iris dataset
• Scikit-learn classifiers: LogisticRegression
• Flask framework to setup up webserver.

The Flask server will provide the front-end for the website allowing the user to enter Iris measurements. Upon submission of the measurements, Flask will call the mlflow RESTapi to obtain the prediction results. Flask then displayes the results as: predicted class, probability and a sample image of the Iris class.

2. Final report (16pts)

1. Code 6pts

• Runs 2pts
• Complete 2pts
• Organized 2pts

2. Results 4pts

• Data summarized/visualized 2pts
• Model selection organized and explained 2pts

3. Interpretation 3pts

• As described in the proposal, was the question answered/topic investigated and how?

4. Reflection 3pts

• Where there deviations from proposal? Explain why (or why not).

3. Project ideas

Gaining more confidence

Use lab2 or lab3 as a template and select a different dataset. Run through the steps that make sense for the data, add new steps if necessary, and show your solution.

Becoming competitive

Select a kaggle.com competition and try to put together a solution for the problem.

Implementing from scratch

Browse Data Sceince from Scratch for an algorithm of interest. Or use any other blog that guides through an implementation from scratch for your favourite algorithm. Demonstrate your working code on a different dataset than the guid/book used.

Machine learning history

Write about the history of machine learning and algorithms. Here it would be important to provide a new perspective. For example, can the sequence of algorithms reported be correlated with popularity of these algorithms?

Investigating a machine learning library or framework

If you find an interesting library or framework, you would follow the intoroduction tutorial and try to adapt it to new data.

Machine learning theory

Write about the mathematics used in one of your favourite algorithms. Here it would be important to connect any equation to code. The goal would be to make theory more accessible to others.

Machine learning productizing

Describe a problem that would need a frontend or app to bring a machine learnign model to users. Design the system, select frameworks, maybe build a small prototype.