智能评卷系统的第二种实现方式具体实现

文本在上文的基础上，对智能评卷系统的第二种实现方式进行代码级别的实现，以展示训练和实现的具体过程。

首先，本次的实现改进了互注意力机制的实现方式，在原来的numpy的实现方式改为了torch的实现，利用torch的softmax函数进行归一化操作，该方式的归一化提高了数据的准确性。

class ShuoFaObj(object):

    def __init__(self):
        self.ref_vector = load_data('./ref.csv')

    def get_ax(self, x: list):
        r = self.ref_vector[0]
        mat_m = torch.mm(torch.tensor(r, dtype=float), torch.tensor(x, dtype=float).t())
        mat_m_r = F.softmax(mat_m, dim=1)
        mat_m_c = F.softmax(mat_m, dim=0)
        row_avg = mat_m_r.mean(dim=0)

        param_ax = torch.mm(mat_m_c, torch.tensor([row_avg.tolist()], dtype=float).t())

        ax = param_ax.mul(torch.tensor(r, dtype=float))

        return ax.tolist()

下一步的训练也是基于torch的实现，进行了一定的改进，提高了训练的效率。

from saveVec import load_data_to_vec_list_together, load_data
from utils import reshape_into_slices, get_padded_vector, label_to_1D
import torch
import torch.nn as nn
from HenYouShuoFa import ShuoFaObj

vec, lable = load_data_to_vec_list_together('vec.csv')

sf = ShuoFaObj()
vec_ax = []
first = []
for i in vec:
    vec_ax.append(sf.get_ax(i))

vec_flat_t = torch.tensor(vec_ax).flatten(start_dim=1)

label_id = []
for li in lable:
    if li == 0:label_id.append(0)
    elif li == 1.5:label_id.append(1)
    else:label_id.append(2)

print(label_id)
label_id_t = torch.tensor(label_id, dtype=torch.long).view(300, -1)

print(vec_flat_t.size())


Model = torch.nn.Sequential(
    nn.Linear(3840, 1500),
    nn.ReLU(True),
    nn.Dropout(0.3),
    nn.Linear(1500, 500),
    nn.ReLU(True),
    nn.Dropout(0.2),
    nn.Linear(500, 3)
)
Model.train()
Model.load_state_dict(torch.load('params.pth'))
loss = nn.CrossEntropyLoss()
optimizer = torch.optim.Adam(Model.parameters(), lr=0.001)

for epoch in range(50):
    i = 0
    for flat_t in vec_flat_t.view(300, 10, -1):
        optimizer.zero_grad()
        out = Model(flat_t)
        loss_b = loss(out, label_id_t[i])
        i += 1
        loss_b.backward()
        optimizer.step()
    print('epoch:'+str(epoch)+', loss is:'+str(loss_b.item()))
    if (epoch+1) % 10 == 0:
        torch.save(Model.state_dict(), 'params.pth')

至此，第二种的训练方式介绍完成，同样会在最后一种评阅方式后进行结果的比对。