I am Jude Capachietti. I am a creative data scientist and design thinker with experience in python, statistical analysis, machine learning, and time series analysis. With a background in math education and non-profit management, I use data storytelling to help make numbers, data, and abstract ideas make sense to people and organizations so that they can make informed data-driven decisions.
I built this webapp using Flask as a webapp server; python, numpy, pytorch for machine learning; and as a way to showcase my programming abilities. I have a text analysis classifier which shows metrics on the tone of text. These are whether the tone sounds positive or negative, whether the gender of the author sounds male, or female, and the age it sounds like the author is. I built the classiers using both Deep Averaging Neural Networks and using a Random Forest, and trained them on the Blog Authorship Corpus. I am happy with the way this classifier works, but I believe it tends to be overconfident. In the future, I plan to expand the text data I use to train these classifiers and also not to overfit the classifiers to the dataset as well. I believe they currently may be a little overfit to these specific text examples. Additionally, I have a name gender classifier. It is under portfolio, and "Name Gender Classifier." It estimates the gender of an entered name. I built this classifier using a bayesian predictive model and numpy. I have it running in python backend of my webpage. In my Jupyter Notebook it had 80% accuracy on my test set, which is 30% better than a 50/50 guess. I would love any feedback on this project. It has many parts, and I have designed the architecture and implemented it all myself. You can see the project in its current state down below!
Check the actual webpage and ML webapps here: My personal webpage
==============================================================================================
I built this webapp using a basic LAMP stack, Linux, Apache, MySQL, and PHP. I built with data vizualation and time saving in mind. I wanted to make it easy for my friends to read the surf forecasts, so I created a place where all the information they need to make a decision is in one place, and the ideal times to surf are colored in green (or yellow if the surf is really good). The goal was to minimized clicking and maximize checking speed of many different locations. This was acoomplished by having many graphs of various spots around New England in one page with color coding by surf quality. These spot names can then be clicked on for more details, and even wind maps. Godaddy says this domain name is worth about $700, so it does have some users still. Let me know if you are interested in buying it XD.
Check the site here: Surf forecasting site
==============================================================================================
This is a project I did as a Machine Learning Student at Cornell where I had to implement a neural network from scratch using NumPy. I learned the mechanics of the Relu activation function and how that leads to 'kinks' or 'hinges' in the linear approximatin of the data. I know there are other activations funvtions as well such as the sigmoid function too.
==============================================================================================
I created a google chrome extension that analyzes text in a similar manner to the text analysis classifier on my webpage. It tells the user what highlighted text sounds like in terms of the tone, gender and age of the author. I use the same RESTful API endpoint that I created for my personal webpage to do the analysis for this extension. I simply send over the highlighted text to the endpoint, then parse and display the response. I submitted this extension to the Google Chrome Webstore, and I hope they approve it so people can use this!
==============================================================================================
This notebook is a place to try out topic modeling and classification. I trained Word Embeddings with Word2Vec and used an ensemble of Random Forests Classifiers to classify the data. The reason it was necessary to use several Random Forest Classifiers is because the data set has multi label data, that samples may have more than one label. The dataset I am using is of research paper titles and abstracts and they fall into one or more of 5 categories: 'Computer Science', 'Physics', 'Mathematics', 'Statistics', 'Quantitative Biology', or 'Quantitative Finance'. Interestingly, the title seems to be a better indicator of the category of paper than the abstract. The data set can be found here.
==============================================================================================
- π Iβm currently working on fine tuning my text analysis classifiers.
- π± Iβm currently learning devops and networking things!
- π― Iβm looking to collaborate on a machine learning project
- π€ Iβm looking for help with devops and networking things!
- π Pronouns: He/Him/His
- β‘ Fun fact: I love surfing!