[tfjs-react-native] Unhandled promise rejection: TypeError: undefined is not an object (evaluating 'a.substr')

System information

• OS Platform and Distribution (e.g., Linux Ubuntu 16.04): Ubuntu 20.04
• Mobile device (e.g. iPhone 8, Pixel 2, Samsung Galaxy) if the issue happens on mobile device: Pixel_3a_API_30_x86.avd
• TensorFlow.js installed from (npm or script link):
• TensorFlow.js version: npm install @tensorflow/tfjs,
• CUDA/cuDNN version:

Describe the problem

I’m trying to run some sample function found in your documentation https://js.tensorflow.org/api/latest/#tf.LayersModel.fit in a react-native-app.

if i run this example on the web executing npm run web and check the browser window it loads, it works fine.

If however i run the same example by running npm run web then scan the QR code using my android phone, or if i hit a to run a android emulator, in both cases i get the following error. [Unhandled promise rejection: TypeError: undefined is not an object (evaluating 'a.substr')]

What is it about the function do_training_example that ive introduced that it does not like when running via the QR code on my phone or via the emulator. If i remove this function the error goes.

[Unhandled promise rejection: TypeError: undefined is not an object (evaluating 'a.substr')]
at node_modules/@tensorflow/tfjs-core/dist/tf-core.node.js:3754:8 in isMobile
at node_modules/@tensorflow/tfjs-backend-webgl/dist/tf-backend-webgl.node.js:1043:17 in <global>
at http://192.168.1.104:19001/index.bundle?platform=android&dev=true&hot=false&minify=false:161631:39 in get
at http://192.168.1.104:19001/index.bundle?platform=android&dev=true&hot=false&minify=false:249141:41 in runWebGLProgram
at node_modules/@tensorflow/tfjs-backend-webgl/dist/tf-backend-webgl.node.js:5217:12 in MathBackendWebGL.prototype.uploadToGPU
at http://192.168.1.104:19001/index.bundle?platform=android&dev=true&hot=false&minify=false:249103:25 in <unknown>
at [native code]:null in map
at node_modules/@tensorflow/tfjs-backend-webgl/dist/tf-backend-webgl.node.js:5016:8 in MathBackendWebGL.prototype.runWebGLProgram
at node_modules/@tensorflow/tfjs-backend-webgl/dist/tf-backend-webgl.node.js:8060:37 in int
at http://192.168.1.104:19001/index.bundle?platform=android&dev=true&hot=false&minify=false:163816:35 in kernelFunc
at node_modules/@tensorflow/tfjs-core/dist/tf-core.node.js:3199:17 in scopedRun$argument_2
at node_modules/@tensorflow/tfjs-core/dist/tf-core.node.js:3011:12 in Engine.prototype.scopedRun
at node_modules/@tensorflow/tfjs-core/dist/tf-core.node.js:3195:12 in Engine.prototype.runKernelFunc
at node_modules/@tensorflow/tfjs/dist/tf.node.js:3548:12 in f2
at node_modules/@tensorflow/tfjs-layers/dist/tf-layers.node.js:24702:13 in tfc.tidy$argument_0
at node_modules/@tensorflow/tfjs-core/dist/tf-core.node.js:3001:126 in scopedRun$argument_2
at node_modules/@tensorflow/tfjs-core/dist/tf-core.node.js:3011:12 in Engine.prototype.scopedRun
at [native code]:null in map
at node_modules/@tensorflow/tfjs-core/dist/tf-core.node.js:3001:126 in scopedRun$argument_2
at node_modules/@tensorflow/tfjs-core/dist/tf-core.node.js:3011:12 in Engine.prototype.scopedRun
at node_modules/@tensorflow/tfjs-layers/dist/tf-layers.node.js:24838:87 in tfc.tidy$argument_0
at node_modules/@tensorflow/tfjs-core/dist/tf-core.node.js:3001:126 in scopedRun$argument_2
at node_modules/@tensorflow/tfjs-core/dist/tf-core.node.js:3011:12 in Engine.prototype.scopedRun
at http://192.168.1.104:19001/index.bundle?platform=android&dev=true&hot=false&minify=false:206783:36 in <unknown>
at http://192.168.1.104:19001/index.bundle?platform=android&dev=true&hot=false&minify=false:199674:24 in step
at node_modules/@tensorflow/tfjs-layers/dist/tf-layers.node.js:16694:22 in step
at node_modules/@tensorflow/tfjs-layers/dist/tf-layers.node.js:16694:22 in step
at node_modules/@tensorflow/tfjs-layers/dist/tf-layers.node.js:16681:40 in fulfilled
at node_modules/react-native/node_modules/promise/setimmediate/core.js:37:13 in tryCallOne
at node_modules/react-native/node_modules/promise/setimmediate/core.js:123:24 in setImmediate$argument_0
at [native code]:null in flushedQueue
at [native code]:null in invokeCallbackAndReturnFlushedQueue

Any other info / logs

You can see ive only added the function do_training_example and executing that within the componentDidMount

App.tsc

import * as tf from '@tensorflow/tfjs';
import '@tensorflow/tfjs-react-native';
import React, { Component } from 'react';
import { Text, View } from 'react-native';
export default class App extends React.Component {
    constructor(props: any) {
        super(props);
        this.state = {
            isTfReady: false,
        };
    }

    async componentDidMount() {
        // Wait for tf to be ready.
        await tf.ready();

        // Signal to the app that tensorflow.js can now be used.
        this.setState({
            isTfReady: true,
        });


        await this.do_training_example();

    }

    do_training_example = async () => {

        const model = tf.sequential({
            layers: [tf.layers.dense({ units: 1, inputShape: [10] })]
        });
        model.compile({ optimizer: 'sgd', loss: 'meanSquaredError' });
        for (let i = 1; i < 5; ++i) {
            const h = await model.fit(tf.ones([8, 10]), tf.ones([8, 1]), {
                batchSize: 4,
                epochs: 3
            });
            console.log("Loss after Epoch " + i + " : " + h.history.loss[0]);
        }

    }


    render() {
        return (
          <View style={{
              flex: 1,
              justifyContent: "center",
              alignItems: "center"
            }}>
            <Text>Hello, world!</Text>
          </View>
        );
    }
}

[MareSeestern]

Choosing a backend with

tf.setBackend('cpu')

solved the problem for me.

[carrycooldude]

Yes true this is bug , Version issue

On Sun, 14 Mar 2021, 20:38 Ralf Gutkowski, @.***> wrote:

Same here on expo while using expo-gl. It seems that tensorflow > 3.0.0 have issues with environment detection.

If you don’t want to downgrade and you can wait until this is fixed, knowing your app is only going to run on mobile, you can place this code early in your program to avoid this error:

/**

• Original functions throw an error in:
• @tensorflow/tfjs-core/dist/tf-core.node.js:3749
• This may be related to TFJS as it for some reason tries to load node webgl
• code instead of using expo-gl bridge.
• Not sure how to figure it out so here’s a workaround. */ tf.device_util.isMobile = () => true tf.device_util.isBrowser = () => false

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/tensorflow/tfjs/issues/4806#issuecomment-798923670, or unsubscribe https://github.com/notifications/unsubscribe-auth/AJZ4ZSFVH74PFPSHKEGN3HTTDTGO3ANCNFSM4ZAWKYLA .