Jared Mlekush's Projects
USF Distributed Data System Class Example Repository
USF MSDS691 (Distributed Data Systems) Class Examples
High level overview of what Bias and Variance is - and a description of what is meant by "Bias-Varaiance Tradeoff"
USF MSDS 610 Communications for Analytics Course
USF MSDS 501 Computation for Analytics Course
Reimplemented Deep Learning project using a Deep Convolutional Generative Adversarial Network (DCGAN) to produce anime faces. Experimented with different loss functions, transformations, and learning rates.
Overview of some important aspects of feature importance. Topics such as: Correlation, general importance strategies, and model-based importance strategies are discussed. Visualizations are used to display importances and compare strategies. Automatic feature selection algorithm implemented.
Fraud Detection Project
Overview of how to deal with imbalanced data. Metrics used, how to solve for said metrics, and how to address the very real problem of imbalanced data
My personal repository
Walk through of K-Means clustering. From scratch implementation shown with an explanation of different initialization techniques, what K-Means is good for, and examples of it in action, utilizing images of yours truly.
I created a Supervised Machine Learning problem. This was done by extracting a numerical column from the data and using it as the target variable. That is, I wanted to see if the other columns of the dataframe (Name, Attack, Speed, Defence, etc.) had "clues" within it that would allow me to predict the "Generation" the Pokemon was from.
Machine learning lab course (USF's MSDS 699)
Course notes for MSDS501, computational boot camp, at the University of San Francisco
MSDS593 -- Exploratory data analysis (EDA) at the University of San Francisco
Course notes for MSDS621 at Univ of San Francisco, introduction to machine learning
Course syllabus, notes, projects for USF's MSDS689
MSAN692 Data Acquisition
Performed in depth analysis of data that describes miRNA features with a target cancer stage. Followed with EDA steps to decide how to structure our model. After doing the necessary analyses, implemented XGBoost and LightGBM, which perform particularly well on tabular data. Afterwards, Optuna framework was used to optimize the hyperparameters of the selected model. Then presented the associated results for both the base model and hyperparameter optimized model.