mnist数据集，运用Python实现softmax

该博客利用Python的mxnet库来简单实现softmax算法，源代码如下：主要包括训练函数，网络构成函数，损失函数，和预测函数。

import gluonbook as gb
from mxnet import autograd,nd

def softmax(x):
    x_exp=x.exp()
    partition=x_exp.sum(axis=1,keepdims=True)
    return x_exp/partition

def net(X,num_inputs,w,b):
    print(b)
    return softmax(nd.dot(X.reshape((-1,num_inputs)),w)+b)

def loss(y_hat,lables):
    ls=-nd.pick(y_hat,lables).log()
    return ls

def train(train_inter,test_inter,batch_size,num_inputs,
          params=None,lr=None,trainer=None):
    num_epochs=3
    for epoch in range(num_epochs):
        ls_sum=0
        acc_sum=0
        for x,y in train_inter:
            with autograd.record():
                y_hat=net(x,num_inputs,params[0],params[1])
                l=loss(y_hat,y)
            l.backward()
            if trainer is None:
                gb.sgd(params,lr,batch_size)
            else:
                trainer.step(batch_size)
            ls_sum+=l.mean().asscalar()
            acc_sum+=accuracy(y_hat,y)
        test_acc=evaluate_accuraxy(test_inter,params)
        #print(ls_sum/len(train_inter),acc_sum/len(train_inter),test_acc)
    return params

def accuracy(y_hat,lables):
    return (y_hat.argmax(axis=1)==lables.astype('float32')).mean().asscalar()

def evaluate_accuraxy(test_inter,params):
    acc=0
    for x,y in test_inter:
        acc+=accuracy(net(x,num_inputs,params[0],params[1]),y)
    return acc/len(test_inter)

def pred(test_inter,num_inputs,params):
    for x,y in test_inter:
        break
    true_lables=gb.get_fashion_mnist_labels(y.asnumpy())
    print(type(true_lables))
    pre=net(x,num_inputs,params[0],params[1]).argmax(axis=1)
    pred_lables=gb.get_fashion_mnist_labels(pre.asnumpy())
    titles = [true + '\n' + pred for true, pred in zip(true_lables, pred_lables)]
    gb.show_fashion_mnist(x[:9],titles[:9])
    
    
if __name__=="__main__":
    batch_size=256
    train_iter,test_iter=gb.load_data_fashion_mnist(batch_size)
    #使用mnist数据集，并将其分块，
    #得到mxnet.gluon.data.dataloader.DataLoader类型的数据
    num_inputs=784
    num_out=10
    w=nd.random.normal(scale=0.01,shape=(num_inputs,num_out))
    b=nd.zeros(num_out)
    w.attach_grad()
    b.attach_grad()
    params=[w,b]
    lr=0.05
    params=train(train_iter, test_iter,batch_size, num_inputs,params, lr)
    pred(test_iter,num_inputs,params)

预测结果如下图：