设训练集、验证集、测试集的特征数据分别为Xtrain、Xval、Xtest，独热编码后的标签分别为ytrain、yval、ytest；假设采用学习率为0.5的随机梯度下降法，进行小批量大小为20、总共10轮训练，模型名称为model_zyq

那么，可以使用以下代码训练模型：

import numpy as np
from model_zyq import Model

# 创建模型对象
model = Model()

# 定义超参数
learning_rate = 0.5
batch_size = 20
epochs = 10

# 获取训练集、验证集、测试集的数据和标签
Xtrain, ytrain = ...
Xval, yval = ...
Xtest, ytest = ...

# 进行独热编码
ytrain = np.eye(num_classes)[ytrain]
yval = np.eye(num_classes)[yval]
ytest = np.eye(num_classes)[ytest]

# 定义损失函数和优化器
loss_fn = ...
optimizer = ...

# 进行训练
for epoch in range(epochs):
    # 打乱训练集顺序
    permutation = np.random.permutation(Xtrain.shape[0])
    Xtrain = Xtrain[permutation]
    ytrain = ytrain[permutation]
    
    # 小批量训练
    for i in range(0, Xtrain.shape[0], batch_size):
        X = Xtrain[i:i+batch_size]
        y = ytrain[i:i+batch_size]
        
        # 前向传播
        y_pred = model.forward(X)
        
        # 计算损失
        loss = loss_fn(y_pred, y)
        
        # 反向传播
        grad = model.backward(loss)
        
        # 更新模型参数
        optimizer.update(model.params, grad, learning_rate)
        
    # 在验证集上评估模型性能
    y_val_pred = model.forward(Xval)
    val_loss = loss_fn(y_val_pred, yval)
    val_acc = accuracy(y_val_pred, yval)
    print('Epoch [{}/{}], Validation Loss: {:.4f}, Validation Accuracy: {:.4f}'.format(epoch+1, epochs, val_loss, val_acc))
    
# 在测试集上评估模型性能
y_test_pred = model.forward(Xtest)
test_loss = loss_fn(y_test_pred, ytest)
test_acc = accuracy(y_test_pred, ytest)
print('Test Loss: {:.4f}, Test Accuracy: {:.4f}'.format(test_loss, test_acc))

其中，`Model` 是模型类，需要自己定义；`loss_fn` 是损失函数，`optimizer` 是优化器，需要根据具体情况选择；`accuracy` 是计算准确率的函数，需要自己定义。