stewartyoung/tf-js-app

Making Deep Learning and Neural Networks using Tensorflow.js

Tensorflow.js Application

But why use a client side model?

• Keeps user data local
• No prerequisites or installs
• Interactive and easy to use

Dependencies for converting keras model to tf-js model

• keras
• tensorflowjs Use pip to install these
  Ran into a big problem installing tensorflowjs, so used
  pip install tensorflow==1.11.0rc2 h5py numpy keras
  pip install --no-deps tensorflowjs
  Again the error persisted, turns out needed to change windows settings to install file with path greater than 260 characters.
  https://www.howtogeek.com/266621/how-to-make-windows-10-accept-file-paths-over-260-characters/

More issues with installations

• Solved more problems with this fantastic tutorial: https://www.youtube.com/watch?v=59duINoc8GM
• Created Python 3.6 environment py36
  • Activate environment
    conda env list
conda activate py36
  • I then exported all packages in the environment to an environment.yaml file
    conda env export > environment.yaml
    which you can use to create the environment yourself using:
    conda env create -f environment.yaml

Setting up the node.js server using express

• cd into local-server and run npm init initially to set up package.json, and add express as a dependency.
• Then you need to run npm install from the root local-server, then the node_modules folder will be downloaded, and the package-lock.json will be added, which contains metadata about the package used in the project

What is the VGG16 model?

It is a pre-trained image classification model