I can't load model from file:// with @tensorflow/tfjs

Question

I am building an npm package that loads a pretrained tensorflow model and makes some prediction.

When i do tf.loadLayersModel() with tf = require('@tensorflow/tfjs-node') i have no problem. Unfortunately, that library seem problematic upon installation. I tried using instead tf = require('@tensorflow/tfjs). Although, i get the following error:

UnhandledPromiseRejectionWarning: Error: browserHTTPRequest is not supported outside the web browser without a fetch polyfill.

This is my code:

 // Load ML model
  tf.loadLayersModel("file://" + __dirname + "mymodel/model.json").then(
    model => {
      }
    }
  );

Is there a way i can work this around?.

@Seblor Now sure you're right about URL: https://www.tensorflow.org/js/guide/save_load#native_file_system_nodejs_only_2 Maybe you have 3 slashes in your path, but only 2 is allowed? — iofjuupasli, Apr 03 '19 at 11:54
I don't get that with the 3 slashes can you explain that a bit more? — Panagiotis Sakkis, Apr 03 '19 at 13:28
Possible duplicate of [TFJS-Node: How to load model from url?](https://stackoverflow.com/questions/52665923/tfjs-node-how-to-load-model-from-url) — edkeveked, Apr 03 '19 at 14:47

score 1 · Answer 1 · answered Apr 03 '19 at 14:47

1

UnhandledPromiseRejectionWarning: Error: browserHTTPRequest is not supported outside the web browser without a fetch polyfill.

You need a polyfill for fetch if you're using @tensorflow/tfjs instead of @tensorflow/tfjs-node

global.fetch = require('node-fetch');

answered Apr 03 '19 at 14:47

edkeveked

14,876
8
45
80

Added that and now i get this error: Error: Request for file://D:\Users\panos\Desktop\eslint-generator/mymodel/model.json failed due to error: TypeError: Only HTTP(S) protocols are supported – Panagiotis Sakkis Apr 03 '19 at 15:37
You have to serve the model files using a local server since you are not using node binding. Consider this answer : https://stackoverflow.com/questions/53639919/load-tensorflow-js-model-from-local-file-system-in-javascript – edkeveked Apr 03 '19 at 15:54

score 1 · Answer 2 · answered Jul 21 '19 at 00:28

It is better to use @tensorflow/tfjs-node when you working on Node.js. Just don't require @tensorflow/tfjs and everything will work as expected.

But if you really want to only use @tensorflow/tfjs, then just add tfjs-node-save

npm install --save tfjs-node-save

and use it in your script

require('tfjs-node-save');

I can't load model from file:// with @tensorflow/tfjs

2 Answers2