This is a demo project to quickly get started with tfjsx. Check out the tfjsx project here.
$ git clone https://github.com/ModelDepot/tfjsx.git
$ cd tfjsx
$ yarn
$ yarn start
Visit http://localhost:1234 to see it running!
Inside src/index.js there's two models and some basic code to demonstrate
what you can do with tfjsx.
MyModel is a simple linear regression model, the
exact same one found
on Tensorflow.js's home page. It includes a trainDataGenerator that has
7 hard coded samples that the model will try to fit.
MnistModel is a more complex CNN found in
Tensorflow.js's MNIST tutorial.
It includes a validation data generator to show how you can visualize validation
metrics on every epoch.
The model training can be paused using the train prop for the <Train>
component.
Pressing the Predict button will output predictions for three samples (1, 3 and 9) in the JS console.
MyApp component drives the model training and holds the final trained
model state. You can swap out rendering MnistModel to MyModel if desired.