tensorflow.js基本使用 欠拟合、过拟合（五）

欠拟合 ：用简单的模型处理复杂的数据

 

$(async () => {
  const data = getData(200);
  tfvis.render.scatterplot(
    { name: '训练数据' },
    {
      values: [
        data.filter(p => p.label === 1),
        data.filter(p => p.label === 0)
      ]
    }
  );
  //欠拟合，模型简单，处理的数据过于复杂

  //设置连续模型
  const model = tf.sequential();

  //添加层，一个神经元，激活函数，sigmoid，设置张量形状
  model.add(tf.layers.dense(
    { units: 1, activation: 'sigmoid', inputShape: [2] }
  ));

  //设置损失函数，对数损失，设置优化器，adam能自动调整学习速率
  model.compile(
    { loss: tf.losses.logLoss, optimizer: tf.train.adam(0.1) }
  );

  //数据格式转换
  const inputs = tf.tensor(data.map(p => [p.x, p.y]));
  const labels = tf.tensor(data.map(p => p.label));

  //训练模型
  await model.fit(inputs, labels, {
    validationSplit:0.2,
    epochs:100,
    callbacks:tfvis.show.fitCallbacks(
      { name: '训练效果' },
      ['loss','val_loss'],
      { callbacks:['onEpochEnd'] }
    )
  });
});

过拟合：模型过于复杂，去兼容少部分数据，导致预测结果不准确

 

$(async () => {
  const data = getData(200);
  tfvis.render.scatterplot(
    { name: '训练数据' },
    {
      values: [
        data.filter(p => p.label === 1),
        data.filter(p => p.label === 0)
      ]
    }
  );

  // 过拟合，模型过于复杂
  const model = tf.sequential();

  model.add(tf.layers.dense({
    units: 15,
    inputShape: [2],
    activation: 'relu'
  }));

  model.add(tf.layers.dense({
    units:1,
    activation:'sigmoid'
  }))

  model.compile({
    loss: tf.losses.logLoss,
    optimizer: tf.train.adam(0.1)
  });

  const inputs = tf.tensor(data.map(p => [p.x, p.y]));
  const labels = tf.tensor(data.map(p => p.label));

  await model.fit(inputs, labels, {
    validationSplit: 0.2,
    epochs: 300,
    callbacks: tfvis.show.fitCallbacks(
      { name: '训练效果' },
      ['loss', 'val_loss'],
      { callbacks: ['onEpochEnd'] }
    )
  })
});

 欠拟合解决方法

1. 提高模型复杂度解决

过拟合解决方法

1. 降低模型复杂度
2. 早停法 减少训练次数
3. 权重衰减 调api
4. 丢弃法 调api

$(async () => {
  var data = getData(200);

  tfvis.render.scatterplot(
    { name: '数据训练集' },
    {
      values: [
        data.filter(p => p.label === 1),
        data.filter(p => p.label === 0)
      ]
    }
  );

  const model = tf.sequential();

  model.add(tf.layers.dense({
    units: 15,
    inputShape: [2],
    activation: 'tanh',
    //设置l2正则化权重衰减，防止过度拟合
    // kernelRegularizer: tf.regularizers.l2({ l2: 12 })//把多余的神经元减下去
  }));

  //添加丢弃层 丢弃一些神经元
  model.add(tf.layers.dropout({ rate: 0.4 }));

  model.add(tf.layers.dense({
    units: 1,
    activation: 'sigmoid'
  }));

  model.compile({
    loss: tf.losses.logLoss,
    optimizer: tf.train.adam(0.1)
  });

  const inputs = tf.tensor(data.map(p => [p.x, p.y]));
  const labels = tf.tensor(data.map(p => p.label));

  await model.fit(inputs, labels, {
    validationSplit: 0.2,
    epochs: 100,
    callbacks: tfvis.show.fitCallbacks(
      { name: '训练过程' },
      ['loss', 'val_loss'],
      { callbacks: ['onEpochEnd'] }
    )
  });
});