javascript - Uncaught (in promise) TypeError: e.iterator is not a function。 如何修复此类错误？

Question

I'm new to javascript and I'm implementing a binary classification. After converting the csv files of training and test data to arrays, I found this error: Uncaught (in promise) TypeError: e.iterator is not a function

Here is the model fitting:

await model.fitDataset(convertedTrainingData, 
   {epochs:100,
    validationData: convertedTestingData,
    callbacks:{
    onEpochEnd: async(epoch, logs) =>{
                             console.log("Epoch: " + epoch + " Loss: " + 
                             logs.loss + " Accuracy: " + logs.acc);
                                  }
                              }});

Answer 1

The error comes from convertedTestingData and convertedTrainingData . fitDataset takes as parameter a tf.data.Dataset . An instance of tf.data.Dataset has an asynchronous iterator.

if convertedTestingData (respectively convertedTrainingData ) is a js array, it needs to be converted either to a tf.tensor or to a tf.data.Dataset.

• Convert js array to tf.tensor

the method fit will be used instead of fitDataset

model.fit(tf.tensor(features), tf.tensor(labels))

• Use tf.data.Dataset

A tf.data.Dataset is created by using a generator.

function createDataGenerator(data) {
    return function* dataGenerator() {
      let index = 0;
      while (index < data.length) {
        const feature = getFeatureTensorAtIndex(i) ; // get the feature tensor at the index
        const label = getLabelTensorAtIndex(i); // get the label tensor at the index
        index++;
        yield {xs: feature, ys: label};
      }
    }
}

const training = tf.data.generator(createDataGenerator(convertedTrainingData));
const testing = tf.data.generator(createDataGenerator(convertedTestingData));

Then the model can be trained with the dataset ds

await model.fitDataset(training,
    {
        epochs: 100,
        validationData: testing,
        callbacks: {
            onEpochEnd: async (epoch, logs) => {
                console.log("Epoch: " + epoch + " Loss: " +
                    logs.loss + " Accuracy: " + logs.acc);
            }
        }
    });