例2
import tensorflow as tf
import numpy as np
import matplotlib.pyplot as plt
plt.style.use('default')
#import tensorflow.contrib.eager as tfe
#from google.colab import files
#tf.enable_eager_execution()
x = np.arange(0, 5, 0.1)
y = x**3 - 4*x**2 - 2*x + 2
y_noise = y + np.random.normal(0, 1.5, len(x))
fig = plt.figure (figsize = (10,6))
plt.plot(x,y, label = "Ground truth", color = 'black')
plt.scatter(x, y_noise, label = "Ground truth + noise", color = 'black')
plt.legend(fontsize = 16, loc = 'upper left')
plt.xlabel ('x', fontsize = 16)
plt.ylabel ('y', fontsize = 16)
plt.tick_params(axis='both', which='major', labelsize=16)
# Save the file for publication
fig.savefig('ch2-michelucci-fig1.png', bbox_inches='tight', dpi = 300)
#files.download('ch2-michelucci-fig1.png')
def model(x,w):
pred=w[0] * x ** 3 + w[1] * x ** 2 + w[2] * x + w[3]
return pred
def loss(x, y,w):
err = model(x,w) - y
return tf.reduce_mean(tf.square(err))
def grad(x,y,w):
with tf.GradientTape() as tape:
loss_ = loss(x,y,w)
return tape.gradient(loss_,[w])
optimizer = tf.optimizers.Adam()
W = tf.Variable(np.random.randn(4),dtype=tf.float32)
loss_list_train = []
#model = PolyModel()
#with tf.GradientTape() as tape:
#grad = tf.implicit_gradients(loss)
#dL_dW,dL_db = tape.gradient(Loss_train,[W,b])
iters = 20001
for i in range(iters):
#optimizer.apply_gradients(grad(model, x, y))
grads=grad(x,y,W)
optimizer.apply_gradients(zip(grads,[W]))
loss_train =loss(x,y,W).numpy()
loss_list_train.append(loss_train)
if i % 2000 == 0:
print("Iteration {}, loss: {}".format(i, loss_train))
loss_hist=loss_list_train
fig = plt.figure (figsize = (10,6))
plt.plot(np.arange(0, iters),loss_hist, label = "Loss function (MSE)", color = 'black')
plt.legend(fontsize = 16, loc = 'upper right')
plt.xlabel ('Iterations', fontsize = 16)
plt.ylabel ('MSE', fontsize = 16)
plt.tick_params(axis='both', which='major', labelsize=16)
plt.xlim(0, 20000)
# Save the file for publication
fig.savefig('ch2-michelucci-fig2.png', bbox_inches='tight', dpi = 300)
#files.download('ch2-michelucci-fig2.png')
fig = plt.figure (figsize = (10,6))
plt.plot(x,y, label = "Ground truth", color = 'black')
plt.scatter(x, y_noise, label = "Ground truth + noise", color = 'black')
plt.plot(x, model(x,W).numpy(), color = 'red', ls = '--', lw = 3,
label = 'Fitted function')
plt.legend(fontsize = 16, loc = 'upper left')
plt.xlabel ('x', fontsize = 16)
plt.ylabel ('y', fontsize = 16)
plt.tick_params(axis='both', which='major', labelsize=16)
# Save the file for publication
fig.savefig('ch2-michelucci-fig3.png', bbox_inches='tight', dpi = 300)
#files.download('ch2-michelucci-fig3.png')
