My Projects

Battleship Solvers

I leveraged 3 different methods to attempt to create optimal battleship strategies and attempted to pit them against each other.

1. Bayesian Sampling of board configurations
2. Naive Reinforcement learning with a Tensorflow Neural Network
3. Using an Advantage Actor Critic (A2C) setup to perform Deep Reinforcement Learning

Read more or Find on Github

Fake Tweet Generation

Using years worth of tweets from various news outlets (CNN, BBC, FOX), I use a Markov Chain crafted in an effort to preserve grammar to generate fake headlines that can sound anywhere from ludicrous to plausible.

Some of the more interesting examples:

Fox News

Obama announces sanctions over his son
Polygamist Sect Working on Missile Defense Shift
Giuliani offered vaccine overseas trip
Mikhail Gorbachev Looks Like Ninjas Targeted Liberals on America GOPDebate
Chinese Quake Destroyed in bathroom

CNN

Trump congratulated the worst possible graves resold plots
The President Jimmy Fallon resigns after moving out
The whistleblower says Joe Biden s first presidential inaugural speech
Sarah Sanders stands at the average resale clothing line against Patrick Mahomes' verified account
Investigators with Hillary Clinton Donald Trump ripped Sen. Joe Biden on the end to come

BBC

Pakistan overcame the first chairman Ben in war-torn areas
Bin Laden Taliban leader to buy new cap on alcohol-related anti-social behaviour
The first civil nuclear weapons seized by mistake by 2030
Police in the country 's government is five newborn babies
Venezuela's north-western Pakistani men are arrested on the Indian court

Read More or Find on Github

An LSTMozart

An RNN-LSTM model is used to learn how to play Mozart sonatas, being trained to predict the time of the beginning of a note, the duration, and the frequenc(y/ies) of the next note(s).

Built in Pytorch, it is a multi-class / multi-label model that simultaneously predicts all the next note's characteristics. Below is a very elegant schema depicting the network's architecture.

Network Architecture

                     INPUT
                       |
          ____LSTM Layers Output____
         /             |            \
     4 layers      3 layers      4 layers
     w/ Relu       w/ Relu       w/ Relu
        |              |             |
      Offset   --->  Concat  <---  Length
     Softmax           |           Softmax
        |          3 layers          |
        |          w/ Relu           |
        |              |             |
        |            Note            |
        |           Sigmoid          |
        |              |             |   
         \__________Concat__________/
                       |
                    OUTPUT

Read More Listen to an Example Song generated by my code on Github