Question

嗨，我正在尝试构建一个转换神经网络，但我无法对其进行训练。

代码如下：

model = tf.sequential();
model.add(tf.layers.conv2d({
    inputShape: [48, 48, 1],
    kernelSize: FILTER_SIZE,
    filters: 64,
    dataFormat: "channelsLast",
    activation: ActFunc.RELU
}));
model.add(tf.layers.maxPooling2d(maxPoolConf));
model.add(tf.layers.conv2d({
    kernelSize: FILTER_SIZE,
    filters: 128,
    dataFormat: "channelsLast",
    activation: ActFunc.RELU
}));
model.add(tf.layers.maxPooling2d(maxPoolConf));
model.add(tf.layers.conv2d({
    kernelSize: FILTER_SIZE,
    filters: 256,
    dataFormat: "channelsLast",
    activation: ActFunc.RELU
}));
model.add(tf.layers.maxPooling2d(maxPoolConf));
model.add(tf.layers.conv2d({
    kernelSize: FILTER_SIZE,
    filters: 512,
    dataFormat: "channelsLast",
    activation: ActFunc.RELU
}));
model.add(tf.layers.maxPooling2d(maxPoolConf));
model.add(tf.layers.flatten());
model.add(tf.layers.dense({units: 128, activation: 'relu'}));
model.add(tf.layers.dense({units: 256, activation: 'relu'}));
model.add(tf.layers.dense({units: 512, activation: 'relu'}));
model.add(tf.layers.dense({units: 1024, activation: 'relu'}));
model.add(tf.layers.dense({
    units: 7,
    activation: 'softmax'
}));
model.compile({
    optimizer: 'adam',
    loss: 'categoricalCrossentropy',
    metrics: ['accuracy', 'categoricalCrossentropy']
});

let image_tensor = tf.tensor4d(training_data.getInputData(), [training_data.length, 48, 48, 1]);
let correct_prediction_tensor = tf.tensor2d(training_data.getLabels(), [training_data.length, 7]);

const history = await model.fit(image_tensor, correct_prediction_tensor,
    {
        batchSize: 128,
        epochs: 10,
        shuffle: true,
        callbacks: {
            onEpochEnd: (epoch, logs) => {
                // Plot the loss and accuracy values at the end of every training epoch.
                console.log(epoch, logs);
            },
            onTrainStart: console.log("Starting Training..."),
            onTrainEnd: console.log("Training Finished!"),
        }
    });

当我运行此代码时，它会显示“正在开始培训...”，并在此之后立即显示“培训结束！”。（它甚至不训练模型），然后我的GPU上有100％的负载，直到我关闭选项卡。我不知道该怎么办。

输入数据是48x48的图像。

training_data.getInputData()返回包含来自每个图像的像素数据的平面数组，而training_data.getLabels()返回包含标签数据的平面数组。

Answer 1

model.fit()是一种异步方法。您应该使用await或then。

例如，使用await（自ES2017起）：

  const history = await model.fit(image_tensor, correct_prediction_tensor,
    {
        batchSize: 128,
        epochs: 10,
        shuffle: true,
        callbacks: {
            onEpochEnd: (epoch, logs) => {
                // Plot the loss and accuracy values at the end of every training epoch.
                console.log(epoch, logs);
            },
            onTrainStart: console.log("Starting Training..."),
            onTrainEnd: console.log("Training Finished!"),
        }
    });

或使用then：

  model.fit(image_tensor, correct_prediction_tensor,
    {
        batchSize: 128,
        epochs: 10,
        shuffle: true,
        callbacks: {
            onEpochEnd: (epoch, logs) => {
                // Plot the loss and accuracy values at the end of every training epoch.
                console.log(epoch, logs);
            },
            onTrainStart: console.log("Starting Training..."),
            onTrainEnd: console.log("Training Finished!"),
        }
    }).then(history => {
      console.log('history:', history);
    });

Tensorflow JS模型拟合立即完成而无需执行任何操作

1 个答案: