Simple handwritten digit recongition using TensorFlow.js.
To install dependencies:
npm install
To convert the jpg files into a csv for the neural network. Really recommend checking out the process script in ./src/process.ts
to understand what is happening more!
npm run process
To train the model:
npm run start
See the code being run in ./src/index.ts
The model is being trained on the MNIST dataset, which is a dataset of 60,000 28x28 grayscale images of the 10 digits, along with a test set of 10,000 images. These images are first converted into a single dimensional array with just the strength of the pixel being represented by a value between 0 and 255. The model is then trained on these images to predict the digit that the image represents.