机器学习：多类逻辑回归-使用gluton

接上一篇机器学习笔记(3)：多类逻辑回归继续，这次改用gluton来实现关键处理，原文见这里 ，代码如下：

import matplotlib.pyplot as plt

import mxnet as mx

from mxnet import gluon

from mxnet import ndarray as nd

from mxnet import autograd

def transform(data, label):

    return data.astype('float32')/255, label.astype('float32')

mnist_train = gluon.data.vision.FashionMNIST(train=True, transform=transform)

mnist_test = gluon.data.vision.FashionMNIST(train=False, transform=transform)

def show_images(images):

    n = images.shape[0]

    _, figs = plt.subplots(1, n, figsize=(15, 15))

    for i in range(n):

        figs[i].imshow(images[i].reshape((28, 28)).asnumpy())

        figs[i].axes.get_xaxis().set_visible(False)

        figs[i].axes.get_yaxis().set_visible(False)

    plt.show()

def get_text_labels(label):

    text_labels = [

        'T 恤', '长 裤', '套头衫', '裙 子', '外 套',

        '凉 鞋', '衬 衣', '运动鞋', '包 包', '短 靴'

    ]

    return [text_labels[int(i)] for i in label]

data, label = mnist_train[0:10]

print('example shape: ', data.shape, 'label:', label)

show_images(data)

print(get_text_labels(label))

batch_size = 256

train_data = gluon.data.DataLoader(mnist_train, batch_size, shuffle=True)

test_data = gluon.data.DataLoader(mnist_test, batch_size, shuffle=False)

num_inputs = 784

num_outputs = 10

W = nd.random_normal(shape=(num_inputs, num_outputs))

b = nd.random_normal(shape=num_outputs)

params = [W, b]

for param in params:

    param.attach_grad()

def accuracy(output, label):

    return nd.mean(output.argmax(axis=1) == label).asscalar()

def _get_batch(batch):

    if isinstance(batch, mx.io.DataBatch):

        data = batch.data[0]

        label = batch.label[0]

    else:

        data, label = batch

    return data, label

def evaluate_accuracy(data_iterator, net):

    acc = 0.

    if isinstance(data_iterator, mx.io.MXDataIter):

        data_iterator.reset()

    for i, batch in enumerate(data_iterator):

        data, label = _get_batch(batch)

        output = net(data)

        acc += accuracy(output, label)

    return acc / (i+1)

#使用gluon定义计算模型

net = gluon.nn.Sequential()

with net.name_scope():

    net.add(gluon.nn.Flatten())

    net.add(gluon.nn.Dense(10))

net.initialize()

#损失函数(使用交叉熵函数)

softmax_cross_entropy = gluon.loss.SoftmaxCrossEntropyLoss()

#使用梯度下降法生成训练器，并设置学习率为0.1

trainer = gluon.Trainer(net.collect_params(), 'sgd', {'learning_rate': 0.1})

for epoch in range(5):

    train_loss = 0.

    train_acc = 0.

    for data, label in train_data:

        with autograd.record():

            output = net(data)

            #计算损失

            loss = softmax_cross_entropy(output, label)

        loss.backward()

        #使用sgd的trainer继续向前"走一步"

        trainer.step(batch_size)

        train_loss += nd.mean(loss).asscalar()

        train_acc += accuracy(output, label)

    test_acc = evaluate_accuracy(test_data, net)

    print("Epoch %d. Loss: %f, Train acc %f, Test acc %f" % (

        epoch, train_loss / len(train_data), train_acc / len(train_data), test_acc))

data, label = mnist_test[0:10]

show_images(data)

print('true labels')

print(get_text_labels(label))

predicted_labels = net(data).argmax(axis=1)

print('predicted labels')

print(get_text_labels(predicted_labels.asnumpy()))

相对上一版原始手动方法，使用gluon修改的地方都加了注释，不多解释。运行效果如下：

相对之前的版本可以发现，几乎相同的参数，但是准确度有所提升，从0.7几上升到0.8几，10个里错误的预测数从4个下降到3个，说明gluon在一些细节上做了更好的优化。关于优化的细节，这里有一些讨论，供参考