"name": "relu_1",
# 层类型
"type": "relu"
},
{
# 层名
"name": "fully_connected_2",
# 层类型，全连接层，
"type": "fully_connected",
# 神经元个数， 因为是10分类，所以神经元个数为10
"neurons_number": 10,
# 权重初始化方式 msra/xavier/gaussian
"weight_init": "msra"
},
{
# 层名
"name": "softmax_1",
# 层类型，分类层，最终输出十分类的概率分布
"type": "softmax"
}
]
# 定义模型，传入网络结构和配置项
AA = AADeepLearning(net=net, config=config)
# 训练模型
AA.train(x_train=x_train, y_train=y_train)
# 使用测试集预测，返回概率分布和准确率， score:样本在各个分类上的概率， accuracy:准确率
score, accuracy = AA.predict(x_test=x_test, y_test=y_test)
print("test set accuracy:", accuracy)
(x_train, y_train), (x_test, y_test) = mnist.load_data()
# 第一个维度是样本数目，第二维度是通道数表示颜色通道数，第三维度是高，第四个维度是宽
x_train = x_train.reshape(x_train.shape[0], 1, 28, 28)
x_test = x_test.reshape(x_test.shape[0], 1, 28, 28)
# 将x_train, x_test的数据格式转为float32
x_train = x_train.astype('float32')
x_test = x_test.astype('float32')
# 归一化，将值映射到 0到1区间
x_train /= 255
x_test /= 255
# 因为是10分类，所以将类别向量(从0到10的整数向量)映射为二值类别矩阵，相当于将向量用one-hot重新编码
y_train = np_utils.to_categorical(y_train, 10)
y_test = np_utils.to_categorical(y_test, 10)
# 网络配置文件
config = {
# 参数模型所在路径，不为空框架会加载模型，用于预测或继续训练
"load_model": "AA-1000.model"
}
# 定义模型，传入配置项
AA = AADeepLearning(config=config)
# 使用测试集预测，返回概率分布和准确率， score:样本在各个分类上的概率， accuracy:准确率
score, accuracy = AA.predict(x_test=x_test, y_test=y_test)
print("test set accuracy:", accuracy)
"type": "relu"
},
{
# 层名
"name": "fully_connected_2",
# 层类型，全连接层
"type": "fully_connected",
# 神经元个数， 因为是10分类，所以神经元个数为10
"neurons_number": 10,
# 权重初始化方式 msra/xavier/gaussian
"weight_init": "msra"
},
{
# 层名
"name": "softmax_1",
# 层类型，分类层，最终输出十分类的概率分布
"type": "softmax"
}
]
# 定义模型，传入网络结构和配置项
AA = AADeepLearning(net=net, config=config)
# 训练模型
AA.train(x_train=x_train, y_train=y_train)
# 画出损失曲线
AA.visualization_loss()
# 使用测试集预测，返回概率分布和准确率， score:样本在各个分类上的概率， accuracy:准确率
score, accuracy = AA.predict(x_test=x_test, y_test=y_test)
print("test set accuracy:", accuracy)
