A list of prereqs could be handy

[dsmilkov]

From @sgserg on April 2, 2018 17:16

Examples work neither on OSX 10.9 (errors out on unsupported node version)
nor in iOS 11:

Copied from original issue: tensorflow/tfjs-examples#49

[dsmilkov]

Hi @sgserg,

Can you share more info about the error you are seeing? Thanks!

[tafsiri]

Also, if you are trying to run the examples locally you should have node version 8.9.0 or greater. This is currently documented in the package.json files for each project.

[sgserg]

@tafsiri yes, trying to run them locally, seeing:

$ yarn
yarn install v1.5.1
warning You are using Node "7.8.0" which is not supported and may encounter bugs or unexpected behavior. Yarn supports the following semver range: "^4.8.0 || ^5.7.0 || ^6.2.2 || >=8.0.0"
warning ../../package.json: No license field
[1/5] 🔍  Validating package.json...
error tfjs-examples-polynomial-regression-core@0.1.0: The engine "node" is incompatible with this module. Expected version ">=8.9.0".
error An unexpected error occurred: "Found incompatible module".
info If you think this is a bug, please open a bug report with the information provided in "/Users/serg/mom/tf/tfjs-examples/polynomial-regression-core/yarn-error.log".
info Visit https://yarnpkg.com/en/docs/cli/install for documentation about this command.

Last time I tried updating node it failed too, so it could be Mavericks limitation.

[sgserg]

@dsmilkov the iOS screenshot is from iPad mini 2 with latest iOS installed. Same results in stock Safari and Firefox.

[tafsiri]

@sgserg You could look at instructions at https://nodejs.org/en/ for how to get a more recent node.js version installed.

To your question about examples on iOS it could be that your device doesn't support webgl float textures, this results in lower precision and may need more training time in order to converge. The red line that renders shows that the example runs, it just didn't learn the function effectively. Once you get it running locally, you can try tweaking the number of iterations and learning rate to see how that affects performance on that device. I've found that on a recent iphone the model learns the correct parameters fairly quickly.