Using a State-Space Time Series model to answer the question, how much revenue can we attribute to the latest product release?
This project is an example of using an Agile stack to develop a data-driven app from scratch. The entire workflow comprises data ingestion, serialization, transformation, data analysis, storage, modeling, predictions, visualization and recommendations. Inspired by the book Agile Data Science 2.0 by Russel Journey.
Tutorial / Notes on the Seminar: Spark with Scala by Elephant Scale at Intel, Santa Clara.
Hypotheses-driven exploration and modeling of online shopping-time. Feature Engineered and fitted a Lasso Regression.
Given a synthetic dataset of coordinates, closing dates and pricing, the goal is to code a k-NN model to predict home closing prices and evaluate its main trade-offs in terms of computational expense, scalability and performance.
Educational article on Exceedance Probabilities in Solar Energy Finance. Among banks and investment firms it's the staple statistical method to determine the economic risk associated with solar resource uncertainty.
Built an interactive web-app during the U.S. State Department Hackathon for market segmentation of solar users in Burma. This project involved wrangling data from many sources. The app was deployed with Bokeh.
Fitted a Non-linear model to forecast 24hr horizon of electrical loads. Built data pipeline and a cross-validating framework for time-series with Python, explored relationship of temperature, load and time with Seaborn visuals. Achieved an out-of-sample Mean Absolute Percentage Error of 1.8%
Analyzed 6GB of Ireland's domestic smart meter data to understand major sources of behavior variability. Used a finite mixture of Gaussian distributions and identified 6 distinct behavioral groups. Performed Control Trial Approach and successfully identified viable candidates for energy saving solutions.