[Deep Learning] Today's bug (August 2)

前言

博主主页：阿阿阿阿锋的主页_CSDN

代码来源：《动手学深度学习》

Got an error message today：TypeError: 'method' object is not iterable.
意思是：类型错误：“方法”对象不可迭代.
然后对mxnetThe understanding of automatic gradient calculation is a little clearer.

---

---

一. TypeError: ‘method’ object is not iterable

1. 错误提示 && 部分代码

错误提示：

---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
<ipython-input-40-9ae7d7a05a23> in <module>
      4 # From numeric labels to text labels
      5 true_labels = d2l.get_fashion_mnist_labels(y.asnumpy())
----> 6 pred_labels = d2l.get_fashion_mnist_labels(net(X).argmax(axis=1).asnumpy)
      7 titles = [true + '\n' + pred for true, pred in zip(true_labels, pred_labels)]
      8 

D:\anaconda\lib\site-packages\d2lzh\utils.py in get_fashion_mnist_labels(labels)
    183     text_labels = ['t-shirt', 'trouser', 'pullover', 'dress', 'coat',
    184                    'sandal', 'shirt', 'sneaker', 'bag', 'ankle boot']
--> 185     return [text_labels[int(i)] for i in labels]
    186 
    187 

TypeError: 'method' object is not iterable

Usually the error message is still very useful,It can effectively help us locatebug的位置.

部分代码段：

for X, y in test_iter:
    break

# From numeric labels to text labels
true_labels = d2l.get_fashion_mnist_labels(y.asnumpy())
pred_labels = d2l.get_fashion_mnist_labels(net(X).argmax(axis=1).asnumpy)
titles = [true + '\n' + pred for true, pred in zip(true_labels, pred_labels)]

d2l.show_fashion_mnist(X[0:9], titles[0:9])

2. 消灭bug

I am looking at this error,当时就懵了,I didn't understand what it meant.A closer inspection of the code later found that,It turned out to be calling a functionasnumpy时,函数名后面的()掉了.于是Originally I wanted to call the function,recognized as an object,passed as a parameter to another function,Then a further type error was thrown（TypeError).

plus missing()后,一切正常.程序跑起来了：

---

二. 自动求梯度,Find the value of the function？

代码：

num_epochs = 3
lr = 0.1
# 用于训练模型
def train_ch3(net, train_iter, test_iter, loss, num_epochs, batch_size, params=None, lr=None, trainer=None):
    for epoch in range(num_epochs):
        train_l_sum, train_acc_sum, n = 0.0, 0.0, 0
        for X, y in train_iter:
            with autograd.record():
                y_hat = net(X)
                l = loss(y_hat, y).sum()
            l.backward()
            d2l.sgd(params, lr, batch_size)
            y = y.astype('float32')
            train_l_sum += l.asscalar()
            train_acc_sum += (y_hat.argmax(axis=1) == y).sum().asscalar()
            n += y.size
        test_acc = evaluate_accuracy(test_iter, net)
        print('epoch %d, loss %.4f, train acc %.3f, test acc %.3f' % (epoch + 1, train_l_sum / n, train_acc_sum / n, test_acc))

I've been a little confused about this code,主要在train_l_sum += l.asscalar()这一语句.Variables are used herel的值,Used to calculate the loss of the model on the training set.但是lwhere does the value come from？

我们看下面这段代码：

%matplotlib inline
import d2lzh as d2l
from mxnet import gluon, autograd, nd

X = nd.array([2, 3, 4])
X.attach_grad()

with autograd.record():
    y = X ** 2
# y.backward()

y, X.grad

输出：

原来在使用mxnetin the process of automatically finding the gradient,在y = X ** 2这一步,It has already been requestedy的值.It is more than just a function expression that requires derivation,It is also an assignment statement.

---

小结

Spend a lot of time on some small mistakes,真的划不来.
Don't do it again next time马虎了啊.

There are often questions,想明白了之后,I just feel so stupid before.