Tfjs: Saving a trained model in MacOS X problem

[ProgrammingLife]

TensorFlow.js version

[email protected]
MacOS X 10.12.5
node v9.4.0

Describe the problem or feature request

Trying to train a model and then save it with:
await model.save('file:///./model');

And I get a message:
(node:2792) UnhandledPromiseRejectionWarning: Error: Cannot find any save handlers for URL 'file:///./model'

How to properly save the trained model?

[caisq]

@ProgrammingLife Did you import tfjs-node using a line such as import '@tensorflow/tfjs-node' or require('@tensorflow/tfjs-node'), in addition importing @tensorflow/tfjs?

[nkreeger]

This is probably related to the fact you don't have a working Node.js package: https://github.com/tensorflow/tfjs/issues/593#issuecomment-412245609

[ProgrammingLife]

My source code is:

"use strict";

const tf = require("@tensorflow/tfjs");
async function go() {

    const training_data = tf.tensor2d([ [0, 0], [0, 1], [1, 0], [1, 1] ]);
    const target_data = tf.tensor2d([ [0], [1], [1], [0] ]);

    const model = tf.sequential();
    model.add(tf.layers.dense({ inputShape: 2, units: 10, activation: "sigmoid" }));
    model.add(tf.layers.dense({ units: 1, activation: "sigmoid" }));
    model.compile({ loss: "meanSquaredError", optimizer: "rmsprop" });
    for (let i = 0; i < 3; i++) {
        var h = await model.fit(training_data, target_data, { epochs: 25 });
        console.log("Loss after Epoch " + i + ", loss = " + h.history.loss[0]);
    }
    model.predict(training_data).print();

    await model.save('file:///./model');
}

go();

And an output is:

Loss after Epoch 0, loss = 0.2550252676010132
Loss after Epoch 1, loss = 0.25055640935897827
Loss after Epoch 2, loss = 0.25000935792922974
Tensor
    [[0.4813891],
     [0.5064855],
     [0.4947665],
     [0.5187985]]
(node:875) UnhandledPromiseRejectionWarning: Error: Cannot find any save handlers for URL 'file:///./model'
    at new ValueError (/Users/user/node-testes/node_modules/@tensorflow/tfjs-layers/dist/errors.js:36:28)
    at Sequential.<anonymous> (/Users/user/node-testes/node_modules/@tensorflow/tfjs-layers/dist/engine/training.js:1126:39)
    at step (/Users/user/node-testes/node_modules/@tensorflow/tfjs-layers/dist/engine/training.js:42:23)
    at Object.next (/Users/user/node-testes/node_modules/@tensorflow/tfjs-layers/dist/engine/training.js:23:53)
    at /Users/user/node-testes/node_modules/@tensorflow/tfjs-layers/dist/engine/training.js:17:71
    at new Promise (<anonymous>)
    at __awaiter (/Users/user/node-testes/node_modules/@tensorflow/tfjs-layers/dist/engine/training.js:13:12)
    at Sequential.Model.save (/Users/user/node-testes/node_modules/@tensorflow/tfjs-layers/dist/engine/training.js:1118:16)
    at go (/Users/user/node-testes/tensorflow-test.js:56:8)
    at <anonymous>

Sure I have all write permissions to the directory.

[caisq]

@ProgrammingLife So your code doesn't import @tensorflow/tfjs-node. Without importing that package, the machinery for file-system saving and loading won't be there. Another thing to point out here is: if you us relative path, you should do something like 'file://./model' (2 slashes after the colon), instead of "file:///./model" (3 slashes after the colon).

[ProgrammingLife]

Is any other way to get weights out of model? I wanna save and load it manually without using model.save() and model.load(). Is it possible?

[caisq]

@ProgrammingLife

1. In your package.json, make sure you have included the dependency such as
   "@tensorflow/tfjs-node": "^0.1.11"

2. In your .js file, do after const tf = require("@tensorflow/tfjs");: require('@tensorflow/tfjs-node');

Let us know if that works.

[ProgrammingLife]

Wow it worked! Thanks! 👍