训练过程可视化

训练的目的是为了迭代出好的参数（自动）、超参数（人工）。在训练的过程中需要将注意力放到三个要素train_loss, train_acc, test_acc。如果能动态的观察训练过程中三个要素的变化，从宏观视角更本质的把握训练过程，则更利于超参数（人工）的选取和debug训练过程中出现的问题。

在实际中以epoch（训练轮数）横坐标，上面三个值为纵坐标进行描点画图，这里面涉及到两个通用的类Animator、Accumulator，二者是配合使用，下面具体分析这两个类的代码细节和用法。

Accumulator

class Accumulator:
    """For accumulating sums over `n` variables."""
    def __init__(self, n):
        """Defined in :numref:`sec_softmax_scratch`"""
        self.data = [0.0] * n

    def add(self, *args):
        self.data = [a + float(b) for a, b in zip(self.data, args)]

    def reset(self):
        self.data = [0.0] * len(self.data)

    def __getitem__(self, idx):
        return self.data[idx]

注意：这里add在代入参数的时候，要写成metric.add(1, 2)形式，不能写成metric.add((1, 2)).

In [26]: metric = Accumulator(2)

In [27]: metric
Out[27]: <__main__.Accumulator at 0x7f2e4827f820>

In [28]: metric[0]
Out[41]: 0.0

metric[1]
Out[44]: 0.0

In [29]: metric.add(1, 2)
metric[0]
Out[46]: 1.0

metric[1]
Out[47]: 2.0

zip() 函数用于将可迭代的对象作为参数，将对象中对应的元素打包成一个个元组，然后返回由这些元组组成的对象，这样做的好处是节约了不少的内存。

我们可以使用 list() 转换来输出列表。

如果各个迭代器的元素个数不一致，则返回列表长度与最短的对象相同，利用 * 号操作符，可以将元组解压为列表。

a = [1,2,3]
b = [4,5,6]
zipped = zip(a,b)  
# 返回一个对象   
zipped

# list() 转换为列表
list(zipped)  
[(1, 4), (2, 5), (3, 6)]
# 再次观察zipped
Out[31]: []
# 说明和迭代器一样，迭代一个，少一个元素
# 使用next再次验证
zipped = zip(a,b)
next(zipped)
Out[34]: (1, 4)
next(zipped)
Out[35]: (2, 5)
next(zipped)
Out[36]: (3, 6)

Accumulator

class Animator:
    """For plotting data in animation."""
    def __init__(self, xlabel=None, ylabel=None, legend=None, xlim=None,
                 ylim=None, xscale='linear', yscale='linear',
                 fmts=('-', 'm--', 'g-.', 'r:'), nrows=1, ncols=1,
                 figsize=(3.5, 2.5)):
        """Defined in :numref:`sec_softmax_scratch`"""
        # Incrementally plot multiple lines
        if legend is None:
            legend = []
        d2l.use_svg_display()
        self.fig, self.axes = d2l.plt.subplots(nrows, ncols, figsize=figsize)
        if nrows * ncols == 1:
            self.axes = [self.axes, ]
        # Use a lambda function to capture arguments
        self.config_axes = lambda: d2l.set_axes(
            self.axes[0], xlabel, ylabel, xlim, ylim, xscale, yscale, legend)
        self.X, self.Y, self.fmts = None, None, fmts

    def add(self, x, y):
        # Add multiple data points into the figure
        if not hasattr(y, "__len__"):
            y = [y]
        n = len(y)
        if not hasattr(x, "__len__"):
            x = [x] * n
        if not self.X:
            self.X = [[] for _ in range(n)]
        if not self.Y:
            self.Y = [[] for _ in range(n)]
        for i, (a, b) in enumerate(zip(x, y)):
            if a is not None and b is not None:
                self.X[i].append(a)
                self.Y[i].append(b)
        self.axes[0].cla()
        for x, y, fmt in zip(self.X, self.Y, self.fmts):
            self.axes[0].plot(x, y, fmt)
        self.config_axes()
        display.display(self.fig)
        display.clear_output(wait=True)

结合上图，这里仅关注 Animator的使用，而忽视其它细节：

1. figsize=(3.5, 2.5)：视觉上框图的长和宽，可以在此基础上适当调整；
2. xlabel='epoch'：x坐标轴的名称；
3. ylabel='epoch'：y坐标轴的名称（对于多个y值的，名称可以不标）；
4. xlim=[1, num_epochs]：x轴数值范围；
5. ylim=[0.3, 0.9]：y轴数值范围；
6. legend=['train loss', 'train acc', 'test acc']：图例，这里的顺序和add(self, x, y)  向量的数值一致（元素个数和意义都一致）；
7. animator.add(epoch + 1, train_metrics + (test_acc,))：每次描 len(y) 个点；

 train_plt函数

reduce_sum = lambda x, *args, **kwargs: x.sum(*args, **kwargs)
astype = lambda x, *args, **kwargs: x.type(*args, **kwargs)

def accuracy(y_hat, y):
    """Compute the number of correct predictions.

    Defined in :numref:`sec_softmax_scratch`"""
    if len(y_hat.shape) > 1 and y_hat.shape[1] > 1:
        y_hat = d2l.argmax(y_hat, axis=1)
    cmp = d2l.astype(y_hat, y.dtype) == y
    return float(reduce_sum(astype(cmp, y.dtype)))


def evaluate_loss(model, train_iter):
    # Set the model to evaluation mode
    if isinstance(model, torch.nn.Module):
        model.eval()
    metric = Accumulator(2)
    for X, y in train_iter:
        # Compute gradients and update parameters
        y_hat = model(X)
        l = loss(y_hat, y)
        metric.add(float(l.sum()), y.numel())
    return metric[0] / metric[1]


def evaluate_accuracy(net, data_iter):
    """Compute the accuracy for a model on a dataset.

    Defined in :numref:`sec_softmax_scratch`"""
    if isinstance(net, torch.nn.Module):
        net.eval()  # Set the model to evaluation mode
    metric = Accumulator(2)  # No. of correct predictions, no. of predictions

    with torch.no_grad():
        for X, y in data_iter:
            metric.add(accuracy(net(X), y), d2l.size(y))
    return metric[0] / metric[1]


def train_plt(animator, idx, model, train_iter):
    train_loss = evaluate_loss(model, train_iter)
    train_acc = evaluate_accuracy(model, train_iter)
    test_acc = evaluate_accuracy(model, test_iter)
    animator.add(idx, (train_loss, train_acc, test_acc))

 

参考资料：

1. 3.6. softmax回归的从零开始实现 — 动手学深度学习 2.0.0 documentation