Task 1: Topic Approval

Choosing a Topic

The approval form ensures you start in the right direction before investing time and effort into task 2. Evaluators look for our (instructors’) signature, and we look for the following:

1. An application of machine learning (ML).

2. An organizational need for your ML application to help solve.

3. A basic outline of your implementation plan.

What is machine learning?

It’s a new field, and answers vary depending on who you ask. But for the purposes of this project, ML is the application of an algorithm to data. Also see FAQ: What is ML?.

The “organizational need” requirement gives your project a purpose and audience. From an assessment perspective, it has little other value, so you need not worry about your project’s profitability or practical impact. Such criteria are not assessed. This is a computer science project, not a business or software project.

Data

ML applies an algorithm to data, and you can’t apply an algorithm to data -without data. So finding data should be an early step.

A machine learning application is the hard requirement of the capstone, but you can’t apply an algorithm to data -without data. Moreover, most datasets provide a question to answer, e.g., predictions, classifications, or recommendations for which an “organizational need” can always be found.

• Kaggle.com

• OpenML. Importing OpenML data

• Google Dataset Search

• Data.gov

• More here and here

• Simulated data

• Python library built-in datasets:

  • sklearn’s data sets (these can be imported directly into your code)

  • PyTorch’s built-in datasets: images, texts, and audio

  • Tensorflow datasets

Note

No minimal data complexity or processing is required. Choosing data which needs less processing or simplifying a dataset (you don’t have to use it all) can make the project technically more accessible.

Machine Learning Algorithms

Classification & Regression Overview

Supervised Learning

If the answers for what you are trying to predict (the dependent variable; also called tagged or labeled) are in the data, then a supervised method is the way to go. Using the other features (independent variables) as input, a supervised algorithm can train a function to predict a dependent variable for new unseen dependent variables. For predicting numbers, e.g., profit, temperature, etc., a regression method should be used. For predicting categories, e.g., yes/no, blue/red, spam/ham, etc., a classification method should be used.

• Regression algorithms predict numbers. Examples include linear regression, polynomial regression, decision tree regression, random forest regression, and many more.

• Classification algorithms predict categories. Examples include logistic regression (see the margin), naive Bayes, support vector machine, decision tree, and many others.

Logisitc regression: regression or classification?

Really both, but most often, it’s used for classification. Logistic regression uses input variables to predict the probability of an outcome, a number between 0.0 and 1.0 -hence “regression.” However, using that probability to predict whether an outcome occurs (yes/no) is classification.

Unsupervised Learning

If what you are trying to find is not directly in the data, then an unsupervised method might be used. Unsupervised algorithms identify patterns in the data. Common approaches include clustering (e.g., k-means, expectation-maximization distribution, and agglomerative hierarchical), dimensionality reduction (e.g., PCA and LDA), and anomaly detection (e.g., outlier factor and isolation forest).

Reinforced Learning

Sometimes solutions need to adapt to particular situations. That is, the algorithms need to learn how to make decisions. For example, a robot needs to navigate a never before seen maze. By awarding (or punishing) a robot’s state after decisions, an algorithm can be progressively trained to predict decisions maximizing results. Reinforced methods can be more challenging to develop than supervised or unsupervised methods.

Note

Need help determining which particular algorithm will work best? That’s OK! You won’t know until you’ve spent time investigating and experimenting. But that’s not the point of Task 1. For now, you only need to identify the type, i.e., supervised, unsupervised, or reinforced. Any ML algorithm given in Task 1 appropriate for your data will be accepted, and you might do something different in task without needing to revise Task 1 (see the FAQ below).

Where to start? Data or the ML algorithm?

It’s best to consider both. In a “real-world” scenario, you would likely be given data to analyze and a problem to solve. So starting with the data, and then determining the proper ML tools is most natural. However, you get to choose the problem and hence the tools needed to solve it -provided you have data fitting that problem. For example, someone interested in image recognition might restrict their data search to sets of labeled images. To simplify coding steps, they might look at images with standard sizes or take a subset of the images limiting the number of classifications. At any point, the problem or chosen algorithm can be revised to better fit what you want or can do (even after graduation 😄). A suggested approach:

1. Find some data to match your preferred ML method.

2. Find an ML algorithm you can apply to that data.

3. Formulate an “organizational need” helped by that application.

Unsure about which ML method to use?

Look for data which has mutually exclusive categories which can used as a dependent variables,e.g., yes/no, alive/dead, blue/red/yellow, etc. Our supervised classification example, uses mutually exclusive flower types. This type of supervised classification method most easily fits all the rubric requirements.

Examples

Topic Summary Examples

Example 1

Data: Petal dimensions and species of Iris Samples.

ML application (non-descriptive method): Classify an Iris’s species using its petal dimensions to train a supervised calssification model, say SVN or Logistic regression.

Descriptive method: Histograms and barplots showing distributions of different flower features and confusion matrix illustrating the accuracy of the classification model.

Organizational Need: Grow Fast Ferilizer Inc. needs to help customers identify their flowers.

Example 2

Data: Housing data

ML application (non-descriptive method): Predict the selling price of a house using a supervised regression model, say random forest or linear regression.

Descriptive method: Histograms showing distributions of data features and scatterplots exploring possible data correlations.

Organizational Need: The realty firm, We Sell em Fast!, needs a tool to estimate housing prices.

Example 3

Data: Movies, e.g., budget, genre, starring actors, etc.

ML application (descriptive method): Use unsupervised clustering, say K-means, to identify groups of similar movies.

Non-descriptive method: Using the clusters and user input, recommend a movie.

Organizational Need: The streaming service, InterWebFlixs, needs to help users pick a movie.

Example 4

Data: Images of dogs and cats.

ML application (non-descriptive): Classify an image as a dog or cat using a supervised neural network (classification model), say CNN.

Descriptive method: Histograms showing distributions of image features and a confusion matrix illustrating the accuracy of the classification model.

Organizational Need: We Love Pets Inc. wants to use customer-uploaded images to market the correct type of pet food.

Completed Examples

These are (slightly modified) examples of approved topics. All of which went on to become passing capstone projects.

Completed Task 1 Ex. A

Also see: task 2 document examples

Completed Task 1 Ex. B

Also see: task 2 example B

WGU Capstone Excellence Archive

The Capstone Excellence Archive includes a wide range of completed projects. Reviewing them may help get ideas, provide inspiration, and deepen understanding of the requirements. However, keep in mind that they all are above and beyond the requirements. Therefore, don’t use these as examples of what’s needed to meet the requirements. For more down-to-earth examples of what’s required, see the examples above.

Topic Approval

Once you’ve decided on a topic, complete the approval form following the template below and email it to your C964 assigned instructor for approval.

Topic Approval Form Template

Note

The approval form only needs to be a rough outline of a passing project. Changes from task 1 to task 2 are allowed and expected. Determining the finer details of a complex project takes time and effort, which you won’t invest until task 2.

Project topic and purpose: Describe the problem (the “organizational need”) your project will solve.

Non-descriptive method(s): These methods infer from the data, i.e., make predictions or prescriptions. Examples include classification models, regression, image recognition, etc. Typically, but not necessarily, this is where ML is applied.

Descriptive methods: These methods describe and help understand the data, e.g., mean, median, bar plot, scatterplot, k-means clustering, etc. Three visualizations are required. The visualizations are typically descriptive and can count as your descriptive method.

Note

You must identify a machine learning application in either the descriptive or non-descriptive section. Most often, the non-descriptive method uses ML, e.g., a classification model allowing the user to provide input to an app that returns a prediction.

Directly emailing your assigned course instructor (identified on your C964 COS page; or find them here) is typically the fastest and best way to get a signature. Whether emailing ugcapstoneit@wgu.edu or your CI directly, always practice professional communication:

• Use your WGU email (non-WGU emails may be categorized as spam).

• Provide a subject, your capstone course (we support all IT college capstones), and your program mentor’s name (if not in your signature).

• Clearly state your questions or requests.

Warning

The submitted topic approval form must be signed by a C964 course instructor. Forms without a signature are automatically returned without further review.

Waiver Form

Note

the waiver form is only required if your project is based upon or includes restricted information. If no waiver form is submitted, Task 1 B: Capstone Release Form, passes automatically.

In most cases, obtaining authorization can be avoided by fabricating or masking identifying information. But if you choose to move forward using restricted information, you must obtain documented permissions and submit them along with a waiver form to Assessments.

FAQ

Do I need to set up an appointment to get approval?

No. Usually, students email the approval form to their assigned instructor. We then sign it or follow up with questions. However, if you have questions about the requirements or need help choosing a topic, you are encouraged to set up an appointment. A 15-30 minute phone call can address most questions or concerns.

I have questions. Should I email the question or set up an appointment?

It’s up to you. But for specific questions, emails are usually best and almost always faster. Appointments are great for general discussions or when you’re unsure what to ask. For technical coding questions or if you are unsure who to contact, see the course faculty page.

What if I start working on task 2 and want to change things? Do I need to resubmit task 1?

No, not unless it’s an entirely different topic. Minor changes from task 1 to task 2 are expected and allowed without updating the approval form. Evaluators will not rigorously compare tasks 1 and 2. Task 2 is where the work is, and even with complete topic changes, at most, you’ll be asked to revise the approval form (if at all). So never let task 1 dictate what you do in task 2.

Can I use projects from other WGU courses?

You can use any of your work or academic projects (at WGU or elsewhere), provided no proprietary information is used without permission. Don’t worry about self-plagiarism, as the similarity check will identify and ignore it. Just as in reusing work projects, expect to modify and remold past academic assignments to meet the project requirements.

Can I use my C950 project for C964?

Yes. You can use any of your own academic or professional work for C964 including the C950 project (Data Structures & Algorithms II). Though the document (Task 2 parts A, B, and D) will need some adjustment, the coding portion of C950 almost meets all the requirements of the C964 application (Task 2 part C) -it only lacks visualizations. Referring to the Task 2 part C page, the C964 application needs the following:

1. Data → ML model: C950 applies a reinforced learning algorithm to the distance and package data.

2. Accuracy Metric: The total miles. The maximum allowed miles for C950 is 120, which WGU Assessment Department has already determined to be “efficient.”

3. Visualizations: This will need to be added, but any three pictures will meet the requirements.

4. User Application: The console user interface required for C950 allows the user to provide input and apply the algorithm toward solving the problem.

So only the three images will need to be added. Furthermore, you are free to adjust the distance and package data as desired. For example, dropping some delivery restrictions requiring different trucks or certain packages to be delivered together, and will be easier to apply a more sophisticated algorithm. Say a Monte Carlo method such as Simulated Annealing.

[Wik18]

Can I use my C951 task 3? Should I use it?

You can reuse anything you’ve of your own academic or professional work, including copying verbatim from C951 task 3. If it’s convenient, feel free to do it. But at best, the time saved is little. At worst, you might get bogged down trying to work on two projects simultaneously and going with an unnecessarily complex C964 topic. If you have time, consider completing C964 first, as parts A and B of task 2 can always be used verbatim for task 3 of C951.

Here are some points to consider:

• C951.3 is just a written project, typically around five pages (I’m guessing; ask your C951 instructor), and can be completed in a single afternoon. Comparatively, C964 requires a working machine learning application and accompanying documentation, typically around 20 pages.

• C951.3 only relates to parts A and B of C964.2. These parts are just a framework for providing a general audience and purpose for the ML application. If present, these parts almost always pass. Furthermore, they’ll have to be at least partially rewritten anyway. Parts C and D of C964 are what evaluators care about, but C951.3 has no corresponding parts C and D.

• Rewriting C951.3 content for a different C964 topic takes little additional work.

• As it’s just a written project, students often pick a complex topic for C951.3. But then they feel pressured to use the same complex topic for C964 and struggle with creating the app.

• Trying to comprehend two projects at once is just more difficult.

Whatever you do for C964 can meet the requirements of C951 task 3. If you have plenty of time, completing C964 first might be the best option.

The rubric and task directions cover items not on this website, e.g., “security features,” “interactive queries,” “monitoring tools,” etc. Do these need to be included?

The official rubric and directions were written to map the project’s elements to specific competencies. However, to allow for a broad range of projects its language is necessarily also broad. For more specifics, we recommend referring to the guidelines on this webpage and the Task 2 template. Following the official directions will meet all the requirements, but be aware that some items might not be applicable to your project or be inherently met by other items. For example, in part C:

• The descriptive and nondescriptive requirement is met by the visualization and decision support functionality respectively.

• Visualization functionality and monitoring tools are inherently part of tools used to create visualizations and code.

• A solution using a mathematically based algorithm to data both suffices as ML and provides a “non-descriptive” method.

The guidance on this website serves to provide some specificity to help students overworking or misinterpreting requirements. For more details of the Task 2 requirements see:

• Part C requirements

• Parts A, B, and D requirements

For the documentation, preserve the template’s section titles, and order, and submit all four parts as a single document (preferably a .pdf). With a long, complicated document, aligning content to competencies presents a challenge. Don’t make things difficult for the evaluator by spreading the content over several documents in an unfamiliar format. Your C964 course instructor team has collaborated with evaluators to ensure the explanations on this webpage align with how the assessment’s requirements are assessed. If anything needs further explanation, please ask us! Contact your C964 course instructor.

The official learning resource seems to include documentation items not included on this webpage. Which should I follow?

Either will meet the requirements. However, the template on this webpage, is more succinct and was developed in collaboration with the capstone evaluators to specifically align with versions SIM3 and SIM2. Hence, you can be ensured that following this website’s template will meet all the requirements.

The content for the official learning resource (LR) was largely copied from an older version of this website which aligned to the previous template written for SIM2. When C964 was updated to SIM3, we updated this website and the template accordingly. Hence, the discrepancy. As the actual requirements for SIM2 and SIM3 are the same, following the official LR or this website should be fine.

Help! I’ve never coded a machine learning project. For C951 task 3, I only had to write about Machine Learning. Where do I learn this?

WGU provides access to a very good AI textbook which includes a Machine Learning section. However, it is conceptually focused and includes very little application or practical examples. Furthermore, reading this text might require mathematics not provided in WGU’s BSCS curriculum.

If you have time, Udemy offers some ML courses](https://wgu.udemy.com/course/machinelearning/learn/lecture/14473662#overview). Maybe the fastest way to get started is with the video and examples included on this website. Though a minimally passing C950 project (applying a greedy algorithm to hand-picked truckloads) would not be consider ML by many, it meets the criteria for this project as it is an algorithm applied to data.

Do I need an “electronic signature?”

No, you can type in your name, use a “fancy” font, or insert an image of your signature.

What are the common reasons for task 1 being returned?

1. No instructor signature on the approval form. You need to send it to us and get a signature before submitting it to Assessments.

2. Both or neither box is marked on the waiver form. Mark one and only one box. See the waiver form instructions

Note, the waiver form is only required if your project is based upon or included restricted information. Task 1 B: Capstone Release Form, passes automatically if no waiver form is submitted, i.e., the waiver only needs to be submitted if it’s needed for permissions.

How many attempts are allowed for each task?

You have unlimited attempts for both tasks 1 and 2. However, incomplete submissions or submissions significantly falling short of the minimum requirements may be locked from further submissions without instructor approval. Furthermore, such submissions do not receive meaningful evaluator comments. It’s in your best interest to the complete minimal requirements to the best of your understanding.

What is a descriptive method?

Anything that describes the data. Histograms, scatter plots, pie charts -all the familiar descriptive statistics techniques are included. ML methods such as k-means clustering can also be descriptive. Whether a method is descriptive or non-descriptive is determined by its use. For example, using a regression line to describe the relationship between variables is descriptive, but using the line to predict a variable or claim a correlation between the variables exist is inferential (non-descriptive).

Descriptive and non-descriptive methods do not need to be explicitly identified in task 2. The visualizations and requirements of the user application will satisfy these parts respectively. The distinction is only made in the approval form so we can easily identify where and how ML will be applied.

What is a non-descriptive method?

Anything that infers from the data, e.g., making predictions, recommendations, identifying correlations, inferring from correlations, etc. Also, see the comments above.

Descriptive and non-descriptive methods do not need to be explicitly identified in task 2. The visualizations and requirements of the user application will satisfy these parts respectively. The distinction is only made in the approval form, so we can easily identify where and how ML will be applied.

What is machine learning (ML)?

That depends on who you ask! But for this project, it is an algorithm applied to data.

For computer science, Machine learning is a subfield of artificial intelligence (a subfield of mathematics), broadly defined as the development of machines capable of self-adjusting behavior based on data. However, from the data science perspective, machine learning is generally defined as using algorithms to identify patterns, make predictions, etc., from data. That is, machine learning is a goal, not a technique. So, for example, a data scientist (and the evaluators) consider linear regression machine learning -when it’s used as a prediction model. However, a mathematician would politely (or not so politely) disagree with a 19^th-century equation being classified as ML.

Can I use libraries outside the standard (Python, Java, etc.) installation?

Yes. Unlike C950 (Data Structures & Algorithms II), you are allowed and encouraged to use outside libraries. All the major languages, but particularly Python, have a wide array of highly developed ML tools. The C964 capstone is about applying these tools -not their development.

What language, libraries, and platforms should I use?

You can use whatever you like. However, we recommend using Python and the scikit-learn (aka sklearn) library. In addition to having an extensive collection of ML-specific tools and tutorials, WGU has better faculty support for these. Jupyter Notebook is a great place to start for the app front end. Passing applications are often submitted as the Notebook file (.ipynb) and data files. Jupyter Notebooks are a great way to present code and information together, but they can also progressively be developed into a more polished product. Students are often tempted to use Java because of their JavaFX experience in software II (C195), but a GUI is not required, and Python is overall better suited for everything which is required. Furthermore, implementing a GUI with Python or Jupyter Notebooks is not difficult (see these widget examples).

A development path might look like the following:

Python IDE → Jupyter notebook → notebook with widgets → hosted notebook with widgets → web app.

Provided the minimal app criteria are met, submitting at any point along this path will pass part C.

What sort of user interface do I need? Do I need a GUI?

No, a GUI is not required. Your app must be usable by the “client” to solve the proposed problem. If the evaluators can run your app as intended, playing the role of the “client,” following your user guide, then your app will be considered to have a user-friendly interface. This can be done through a GUI and widgets, but using the command line or reading user data from a local directory will also suffice.

Are there length requirements for the documentation?

No. What you see in the examples and template are just guidelines. The individual evaluator determines what qualifies as “sufficient detail,” which further varies depending on the project and writing style. If you feel you’ve met the requirements, simply move on to the next section. Upon submission, it will pass, or they will request more details. In the latter case, you can then focus on revising the more narrow scope as directed by the evaluator’s comments, which is generally more efficient than overworking the entire project.

I’ve completed the task 2 documentation. Should I send it to my course instructor for review?

If you have specific questions or concerns -yes. However, in most cases, it’s best just to submit. What suffices as “sufficient detail” is highly subjective. We can always tell you to add more, but if you’ve done your best to fulfill the requirements, submit it and let them tell you which (if any) parts need revision. At best, it passes; at worst, we address the issues cited by the evaluator -and then it passes. Responding to the more narrow focus of the evaluator’s comments is generally easier than overworking the entire project.

You have unlimited submissions but limited time. So typically this is the best and most efficient approach.

I only have a Linux (or Mac) machine. Will evaluators be able to run my code?

Technically (and unfortunately), we are a “Windows” university, and all submitted projects should be able to run in Windows. However, being Windows-compatible is nowhere specifically required in the C964 rubric, and doing so would be a little silly for a computer science program. That said, WGU evaluators are only issued Windows 10 machines, and they may have difficulty running a Linux, Android, or Mac app without special instructions. Therefore for non-Windows apps, we recommend providing explicit instructions in the user guide for a Windows 10 user to run your code, such as using a virtual machine, a remote machine, or using a Linux subsystem.

How complex does my data, algorithm, or model need to be?

It must be complex enough to meet the needs of your project. There is no explicit minimal complexity for any of these items. However, the model must meet the needs of the “organizational need” and the data must be appropriate for developing the model which could indirectly impose a minimal complexity. For example, a supervised model requires two variables.

Are there any restrictions on which datasets I can choose?

Only that data must be legally available to use and share with evaluators. For example, using data belonging to a current employer would require submitting a waiver form.

• You can use any dataset found on kaggle.com.

• You can use simulated data.

• You can use data used for previous projects (submitted by you or others).

• You only need to apply for IRB review if you are collecting data involving human participants (this is very rare). Otherwise, your project is in IRB compliance.

Questions, comments, or suggestions?