Portfolio
Individualized Learning with Bayesian Knowledge Tracing
Bayesian Knowledge Tracing (BKT) is a probabilistic model used in educational data mining (EDM). The BKT model is a Hidden Markov Model that estimates learners’ knowledge over time. Data is gathered from learners interacting with a system, included whether they get a problem correct or not. BKT empowers teachers to identify struggling students, design personalized interventions, and improve learning outcomes. In this example, we implement Bayesian Knowledge Tracing with the pyBKT Python library, described in the paper by Bulut et al.
Generating Music with Deep Learning
We develop a model to improvise percussion music (code download or display notebook without output. The entire file is too large to display on Github). Since music is sequential, we use a specialized Recurrent Neural Network (RNN) called a LSTM model (Long short-term Memory) Model to learn the patterns of musical sequences. We then use these learned patterns to generate new music.
The type of music depends on a collection of music files in MIDI format. Each MIDI file corresponds to a musical piece, which is a series of notes over time. In this example, we use the Groove Dataset from TensorFlow. However, in the appendix, we show how to import files from a url.
Titanic Kaggle Competition
Ranked in the top 7% of the Kaggle Competition for prediction which passengers survived the Titanic. My submission included data imputation for missing data, creating new features, comparing different models, and creating ensemble models. An outline of my submission gives an idea of my thinking process.
Homeschooling Data Visualization
Data Visualization using R and RStudio (code). A variety of graphs illustrate different aspects of homeschooling families.
Prompt Engineering Example – ChatGPT and Mad Libs
Demonstrated the difference a prompt can make with ChatGPT. Used the example of generating Mad Libs with different prompts.
NLP Word Embeddings
In Natural Language Processing (NLP), Word Embeddings represent words with real-valued vectors in sucha way that word embedding vectors that are close correspond to words that are similar in meaning. In this project, we use Linear Algebra to
- Solve Word Analogies
- Discover bias in the word embeddings
- Neutralize bias in words like “receptionist”
- Apply equalization to pairs of words like “actor” and “actress” on either side of a bias to make them reflections over the bias vector.
Student Dashboard
My homeschooling students used student dashboards to measure their quarterly progress.