tensorflow2 多项式拟合

参考博文

https://www.jianshu.com/p/37a7e9b51891

import os
import tensorflow as tf
import numpy as np
import matplotlib.pyplot as plt

os.environ['CUDA_VISIBLE_DEVICES'] = "0"  # Specify visible gpus.
tf.debugging.set_log_device_placement(True)  # Show the devices when calculating.

cont=1000
input_x = np.linspace(-3,3,cont) #创建等差数列
# input_y = 3*np.sin(input_x) + np.random.uniform(0,1,cont)
input_y =input_x * 1.0 + pow(input_x,2)*2.0+ pow(input_x,3)*1.5 + pow(input_x,4)*1.1 + 5.0  +  np.random.uniform(-2,2,cont)

#绘制出原始数据
plt.scatter(input_x, input_y)
plt.show()

initializer = tf.initializers.GlorotUniform()
# W = tf.Variable(initializer(shape=(1, ), dtype=tf.float64), name="W")
# b = tf.Variable(initializer(shape=(1, ), dtype=tf.float64), name="b")

w1 = tf.Variable(0.0,dtype=tf.float64,name='w1')
w2 = tf.Variable(0.0,dtype=tf.float64,name='w2')
w3 = tf.Variable(0.0,dtype=tf.float64,name='w3')
w4 = tf.Variable(0.0,dtype=tf.float64,name='w4')
b = tf.Variable(0.0,dtype=tf.float64,name='b')

optimizer = tf.optimizers.Adam(0.01)

def train_step(x, y_batch, epoch, batch_i):
    with tf.GradientTape() as tape:
        y_predict = tf.multiply(w1, x) + tf.multiply(w2, tf.pow(x, 2)) + tf.multiply(w3, tf.pow(x, 3)) + tf.multiply(w4,tf.pow(x, 4)) + b
        loss = tf.reduce_mean(tf.math.pow(y_predict - y_batch, 2))
    # print("loss", loss)
    train_variables = [w1,w2,w3,w4, b]
    gradients = tape.gradient(loss, train_variables)
    # print("grads", gradients)
    optimizer.apply_gradients(zip(gradients, train_variables))
    print(epoch, batch_i,loss.numpy())
    return loss


def fit():
    for epoch in range(100):
        for i in range(0, cont, 10):
            start = i
            end = i + 10
            x_batch = input_x[start:end]
            y_batch = input_y[start:end]
            loss=train_step(x_batch, y_batch, epoch, i)
            # if(loss.numpy()< 1):
            #     return

fit()

print(w1)
print(w2)
print(w3)
print(w4)
print(b)

#作图，显示线性回归的结果
plt.plot(input_x, input_y, 'bo', label='real data')
# plt.plot(input_x, input_x * w1 + pow(input_x,2)*w2+ pow(input_x,3)*w3 + pow(input_x,4)*w4 + b, 'r', label='predicted data')
plt.plot(input_x, input_x * w1 + input_x**2*w2+  input_x**3*w3 +  input_x**4*w4 + b, 'r', label='predicted data')
plt.legend()
plt.show()

 

拟合结果如下图所示：

 

另外一种写法也可以

import tensorflow as tf
import numpy as np
import matplotlib.pyplot as plt


cont=200
input_x = np.linspace(-3,3,cont) #创建等差数列
input_y =5.0*np.sin(input_x) + np.random.uniform(0,1,cont)
# input_y =input_x * 1.0 + pow(input_x,2)*2.0+ pow(input_x,3)*1.5 + pow(input_x,4)*1.1 + 5.0  +  np.random.uniform(-2,2,cont)

w1 = tf.Variable(0.0,dtype=tf.float64,name='w1')
w2 = tf.Variable(0.0,dtype=tf.float64,name='w2')
w3 = tf.Variable(0.0,dtype=tf.float64,name='w3')
w4 = tf.Variable(0.0,dtype=tf.float64,name='w4')
b = tf.Variable(0.0,dtype=tf.float64,name='b')


#绘制出原始数据
plt.scatter(input_x, input_y)
plt.show()

def model(x):
    y = tf.multiply(w1, x) + tf.multiply(w2, tf.pow(x, 2)) + tf.multiply(w3, tf.pow(x, 3)) + tf.multiply(w4, tf.pow(x, 4)) + b
    return y

opt = tf.optimizers.Adam(0.01)
# opt = tf.optimizers.Adadelta(0.01)

batchSize=20

for epoch in range(200):
    for i in range(0, cont, batchSize):
        start = i
        end = i + batchSize
        x_batch = input_x[start:end]
        y_batch = input_y[start:end]
        loss = lambda: tf.losses.MeanSquaredError()(model(x_batch), y_batch)
        opt.minimize(loss, [w1, w2, w3, w4, b])
        print(epoch,i, loss().numpy())


print(w1)
print(w2)
print(w3)
print(w4)
print(b)

#作图，显示线性回归的结果
plt.plot(input_x, input_y, 'bo', label='real data')
plt.plot(input_x, input_x * w1 + input_x**2*w2+  input_x**3*w3 +  input_x**4*w4 + b, 'r', label='predicted data')
plt.legend()
plt.show()