Brandon Rohrer

Book Project

How to Train Your Robot is a book-in-progress about applied robotics, machine learning, and software engineering.

Code

cottonwood, a teaching-focused neural network framework
dsmq, a dead simple message queue
myrtle, a workbench for real-time reinforcement learning
pacemaker, a steady time keeper for wall-clock sensitive functions
sqlogging, a SQLite-backed logging package that mimics Python's logging package

Courses

NOTE: The courses will be coming down on March 1, 2025. Here's a short post on things I learned while having them online.

Setting up Python on MacOS
Setting up Python on Windows
Intro to Python: Time tools A Python course for the absolute beginner. Zero prior knowledge assumed. Learn Python basics while building projects like a clock, a timer, and a stopwatch.
Decision Trees Code up a decision tree in python from scratch. Dynamically construct URL queries for live transit data API. Build the model into a command line application.
Time-series Prediction Build a command line weather prediction tool from a century of data. Perform data-driven deseasonalization to remove annual weather patterns. Use autocorrelation to extract predicted temperatures.
Polynomial Regression Code up a robust optimizer from scratch in python. Fit high-order polynomials to real data on dog breeds. Implement Monte Carlo cross-validation to select the best model.
k-nearest neighbors Code up the k-nearest neighbors algorithm in Python. Use k-NN for categorization, regression, and interpolation on several data sets.
Neural Network Visualization Create a custom neural network visualization in python. Learn Matplotlib tricks for making professional plots.
Build a Neural Network Framework Code up a fully connected deep neural network from scratch in Python. Extend it into a framework through object-oriented design.
Advanced Neural Network Methods Add regularization, dropout, computation graphs and optimizer options to the framework we built in Course 312. Run it on images from Mars.
Neural Network Optimization Build an autoencoder to extract basis elements of images of the Martian surface. Optimize compression performance by tuning hyperparameters.
One Dimensional Convolutional Neural Networks Build a classification model for detecting unhealthy rhythms in electrocardiography data. Electrocardiogram case study
Two Dimensional Convolutional Neural Networks Build image classification models for benchmark data sets. MNIST digits case study CIFAR-10 images case study

Data munging

Data science

Machine Learning

Matplotlib

Networking

Writing a bare-bones HTTP client in Python
Writing a bare-bones HTTP server in Python
The word "server" can mean several things
Setting up an ssh server Notes from my home hackery.

Neural Networks

Projects and case studies

Python

Data types
dsmq—a dead simple message queue
Make your code run faster
Multiprocessing for Parallelization
Multiprocessing for Real-Time Applications
Threading

Research

Evolutionary Powell's method for hyperparameter tuning
k-nearest neighbors adaptive feature weighting
k-nearest neightbors data reduction
k-sparse neural network layer
Naive Cartographer , a Markov Decision Process learner
oknn, an online variant of kNN
Pathfinder, a many-to-many shortest path algorithm
Sharpened Cosine Similarity, an alternative to convolution for learning 2D features in neural networks
Ziptie, an unsupervised feature creation algorithm