paddle基本操作

1. Helloworld

# helloworld示例
import paddle.fluid as fluid

# 创建两个类型为int64, 形状为1*1张量
x = fluid.layers.fill_constant(shape=[1], dtype="int64", value=5)
y = fluid.layers.fill_constant(shape=[1], dtype="int64", value=1)
z = x + y # z只是一个对象,没有run,所以没有值

# 创建执行器
place = fluid.CPUPlace() # 指定在CPU上执行
exe = fluid.Executor(place) # 创建执行器
result = exe.run(fluid.default_main_program(),
                 fetch_list=[z]) #返回哪个结果
print(result) # result为多维张量

2. 张量操作

import paddle.fluid as fluid
import numpy

# 创建x, y两个2行3列，类型为float32的变量(张量)
x = fluid.layers.data(name="x", shape=[2, 3], dtype="float32")
y = fluid.layers.data(name="y", shape=[2, 3], dtype="float32")

x_add_y = fluid.layers.elementwise_add(x, y)  # 两个张量按元素相加
x_mul_y = fluid.layers.elementwise_mul(x, y)  # 两个张量按元素相乘

place = fluid.CPUPlace()  # 指定在CPU上执行
exe = fluid.Executor(place)  # 创建执行器
exe.run(fluid.default_startup_program())  # 初始化网络

a = numpy.array([[1, 2, 3],
                 [4, 5, 6]])  # 输入x, 并转换为数组
b = numpy.array([[1, 1, 1],
                 [2, 2, 2]])  # 输入y, 并转换为数组

params = {"x": a, "y": b}
outs = exe.run(fluid.default_main_program(),  # 默认程序上执行
               feed=params,  # 喂入参数
               fetch_list=[x_add_y, x_mul_y])  # 获取结果
for i in outs:
    print(i)

3. 简单线性回归

# 简单线性回归
import paddle
import paddle.fluid as fluid
import numpy as np
import matplotlib.pyplot as plt

train_data = np.array([[0.5], [0.6], [0.8], [1.1], [1.4]]).astype('float32')
y_true = np.array([[5.0], [5.5], [6.0], [6.8], [6.8]]).astype('float32')

# 定义数据数据类型
x = fluid.layers.data(name="x", shape=[1], dtype="float32")
y = fluid.layers.data(name="y", shape=[1], dtype="float32")
# 通过全连接网络进行预测
y_preict = fluid.layers.fc(input=x, size=1, act=None)
# 添加损失函数
cost = fluid.layers.square_error_cost(input=y_preict, label=y)
avg_cost = fluid.layers.mean(cost)  # 求均方差
# 定义优化方法
optimizer = fluid.optimizer.SGD(learning_rate=0.01)
optimizer.minimize(avg_cost)  # 指定最小化均方差值

# 搭建网络
place = fluid.CPUPlace()  # 指定在CPU执行
exe = fluid.Executor(place)
exe.run(fluid.default_startup_program())  # 初始化系统参数

# 开始训练, 迭代100次
costs = []
iters = []
values = []
params = {"x": train_data, "y": y_true}
for i in range(200):
    outs = exe.run(feed=params, fetch_list=[y_preict.name, avg_cost.name])
    iters.append(i)  # 迭代次数
    costs.append(outs[1][0])  # 损失值
    print("i:", i, " cost:", outs[1][0])

# 线性模型可视化
tmp = np.random.rand(10, 1)  # 生成10行1列的均匀随机数组
tmp = tmp * 2  # 范围放大到0~2之间
tmp.sort(axis=0)  # 排序
x_test = np.array(tmp).astype("float32")
params = {"x": x_test, "y": x_test}  # y参数不参加计算，只需传一个参数避免报错
y_out = exe.run(feed=params, fetch_list=[y_preict.name])  # 预测
y_test = y_out[0]

# 损失函数可视化
plt.figure("Trainging")
plt.title("Training Cost", fontsize=24)
plt.xlabel("Iter", fontsize=14)
plt.ylabel("Cost", fontsize=14)
plt.plot(iters, costs, color="red", label="Training Cost")  # 绘制损失函数曲线
plt.grid()  # 绘制网格线
plt.savefig("train.png")  # 保存图片

# 线性模型可视化
plt.figure("Inference")
plt.title("Linear Regression", fontsize=24)
plt.plot(x_test, y_test, color="red", label="inference")  # 绘制模型线条
plt.scatter(train_data, y_true)  # 原始样本散点图

plt.legend()
plt.grid()  # 绘制网格线
plt.savefig("infer.png")  # 保存图片
plt.show()  # 显示图片