Text generation on micro-controllers with CircuitPython and Markov Chains

This code generates random text using character level Markov Chains. It has been written for CircuitPython and should run on wide variety of boards supporting it. By default the code creates a character level Markov Chain which can be used for generating random dinosaur names. You can train a character level Markov Chain model on any type of data using this code.

Why CircuitPython and Markov Chains ?

• C++ is hard. Setting up dev environment for TFlite and board's SDK can be challenging for newbies.
• CircuitPython has many useful libraries available out of box.
• Loading TFLite models is currently not supported by CircuitPython. Checkout this open source project for progress in this direction .
• RNNs are complicated and use more memory.
• Markov Chains are simpler and lightweight !

Steps to run

• Download the file https://raw.githubusercontent.com/junosuarez/dinosaurs/master/dinosaurs.csv
• Generate a Markov Chain model from your text file using src/generate_chain.py (by default it generates the dino_chain.json file).
• Copy src/generate_text.py, src/markov_chain_parser.py and src/dino_chain.json to your CircuitPython device.
• Check output on a serial monitor.

Alternatively you can skip Steps 1 and 2, and run the 'src/text-generation-on-microcontrollers.ipynb' Jupyter notebook to generate dino_chain.json. This notebook is also available on Kaggle: https://www.kaggle.com/code/finalepoch/text-generation-on-microcontrollers

Output on serial monitor

This program (src/generate_text.py) outputs data to serial port. It uses CircuitPython's 'printf' statement to write data.

Output on OLED display

This program (src/generate_text_oled.py) requires an I2C OLED display (128x64) to be connected to the board. The example assumes Pi Pico board, so the OLED is connect to GP5(SLK) and GP4(SDA)