how to label a three dimensional data in tensorflow.js - tensorflow.js

Question

I have a three dimensional data to fit a tensorflow.js model on it. But the code I have tried is not showing any value nor any error. I don't know how to exactly assign a label to a feature. How I am supposed to do this with three dimensional data ? The complete code with model is attached below
Update
import React from 'react';
import {View, Text, StyleSheet} from 'react-native';
import * as tf from '#tensorflow/tfjs';
import '#tensorflow/tfjs-react-native';
export default class App extends React.Component {
constructor(props) {
super(props);
this.state = {
isTfReady: false,
};
}
init() {
const model = tf.sequential({
layers: [
tf.layers.dense({
inputShape: [784],
units: 32,
activation: "relu"
}),
tf.layers.dense({
units: 10,
activation: "softmax"
})
]
});
model.compile({
optimizer: "sgd",
loss: "categoricalCrossentropy",
metrics: ["accuracy"]
});
featureTensor = tf.tensor2d([[1],[2],[4]])
featureTensor1 = tf.tensor2d([[2],[1],[3]])
featureTensor2 = tf.tensor2d([[3],[2],[1]])
featuresTensor = tf.stack([featureTensor, featureTensor1, featureTensor2])
const labelArray = ["standing", "sitting"]
// {standing: 0, sitting: 1}
const mapIndexLabel = Object.fromEntries(Object.entries({...labelArray}).map(([a, b]) => [b, +a]))
labelsTensor = tf.onehot([0, 1, 1], numberOfUniqueLabels) // numberOfUniqueLabels = 2 in this case
function onBatchEnd(batch, logs) {
logs.acc = parseFloat((logs.acc * 100).toFixed(2));
logs.loss = parseFloat((logs.loss * 100).toFixed(3));
console.log("Accuracy", logs.acc);
}
// Train for 5 epochs with batch size of 32.
model
.fit(featuresTensor, labelsTensor, {
epochs: 10,
batchSize: 32,
callbacks: {
onBatchEnd
}
})
.then(info => {
this.setState({info: info })
history = model.fit(featuresTensor, labelsTensor, {
});
console.log("Final accuracy", info.history.acc);
});
}
async componentDidMount() {
await tf.ready();
this.init()
// Signal to the app that tensorflow.js can now be used.
this.setState({
isTfReady: true,
});
}
render() {
const {info} = this.state;
return (
<View style={styles.output}>
{info && info.history.acc.map(i=>(<Text>{i}</Text>))}
</View>
)
}
}
const styles = StyleSheet.create({
output: {
flexDirection: 'column',
// alignItems: 'center',
marginTop: 50,
},
})

Answer 1

I have a three dimensional data to fit a tensorflow.js
It is not a three dimensional data, it is a two dimensional data. tf.tensor3d([[1],[2],[4]]) will fail in constructing a 3d tensor, thus throwing an error. To constructor the feature tensors, tf.tensor can be used - it will infer the shape from the data itself; alternatively you can use tf.tensor2d for easy readability.
featureTensor = tf.tensor2d([[1],[2],[4]])
The second thing to mention is that there needs to be a flatten layer before the last layer to allow the model output to be of one dimension.
As for the callBack, it needs to be called given a certain event.
const model = tf.sequential({
layers: [
tf.layers.dense({
inputShape: [3, 1],
units: 32,
activation: "relu"
}),
tf.layers.flatten(),
tf.layers.dense({
units: 2,
activation: "softmax"
})
]
});
model.compile({
optimizer: "sgd",
loss: "categoricalCrossentropy",
metrics: ["accuracy"]
});
featureTensor = tf.tensor2d([[1],[2],[4]])
featureTensor1 = tf.tensor2d([[2],[1],[3]])
featureTensor2 = tf.tensor2d([[3],[2],[1]])
featuresTensor = tf.stack([featureTensor, featureTensor1, featureTensor2])
const labelArray = ["standing", "sitting"]
// {standing: 0, sitting: 1}
const mapIndexLabel = {standing: 0, sitting: 1}
labelsTensor = tf.oneHot([0, 1, 1], 2) // numberOfUniqueLabels = 2 in this case
function onBatchEnd(batch, logs) {
logs.acc = parseFloat((logs.acc * 100).toFixed(2));
logs.loss = parseFloat((logs.loss * 100).toFixed(3));
console.log("Accuracy", logs.acc);
}
model.summary()
// Train for 5 epochs with batch size of 32.
featuresTensor.print()
labelsTensor.print()
hist = await model
.fit(featuresTensor, labelsTensor, {
callbacks: {onEpochEnd: onBatchEnd }
})
console.log(hist)