带你使用 TensorFlow 进行机器学习研究

import tensorflow as tf

tf.enable_eager_execution()

# Using python state
x = tf.zeros([10, 10])
x += 2  # This is equivalent to x = x + 2, which does not mutate the original
        # value of x
print(x)

tf.Tensor（[[2。2. 2. 2. 2. 2. 2. 2. 2. 2.]
[2。2. 2. 2. 2. 2. 2. 2. 2. 2.]
[2。2. 2. 2. 2. 2. 2. 2. 2. 2.]
[2。2. 2. 2. 2. 2. 2. 2. 2. 2.]
[2。2. 2. 2. 2. 2. 2. 2. 2. 2.]
[2。2. 2. 2. 2. 2. 2. 2. 2. 2.]
[2。2. 2. 2. 2. 2. 2. 2. 2. 2.]
[2。2. 2. 2. 2. 2. 2. 2. 2. 2.]
[2。2. 2. 2. 2. 2. 2. 2. 2. 2.]
[2。2. 2. 2. 2. 2. 2. 2. 2. 2.]]，shape =（10,10），dtype = float32）

v = tf.Variable(1.0)
assert v.numpy() == 1.0

# Re-assign the value
v.assign(3.0)
assert v.numpy() == 3.0

# Use `v` in a TensorFlow operation like tf.square() and reassign
v.assign(tf.square(v))
assert v.numpy() == 9.0

class Model(object):
  def __init__(self):
    # Initialize variable to (5.0, 0.0)
    # In practice, these should be initialized to random values.
    self.W = tf.Variable(5.0)
    self.b = tf.Variable(0.0)
   
  def __call__(self, x):
    return self.W * x + self.b
 
model = Model()

assert model(3.0).numpy() == 15.0

def loss(predicted_y, desired_y):
  return tf.reduce_mean(tf.square(predicted_y - desired_y))

TRUE_W = 3.0
TRUE_b = 2.0
NUM_EXAMPLES = 1000

inputs  = tf.random_normal(shape=[NUM_EXAMPLES])
noise   = tf.random_normal(shape=[NUM_EXAMPLES])
outputs = inputs * TRUE_W + TRUE_b + noise

import matplotlib.pyplot as plt

plt.scatter(inputs, outputs, c='b')
plt.scatter(inputs, model(inputs), c='r')
plt.show()

print('Current loss: '),
print(loss(model(inputs), outputs).numpy())

Current loss: 

 7.92897

def train(model, inputs, outputs, learning_rate):
  with tf.GradientTape() as t:
    current_loss = loss(model(inputs), outputs)
  dW, db = t.gradient(current_loss, [model.W, model.b])
  model.W.assign_sub(learning_rate * dW)
  model.b.assign_sub(learning_rate * db)

model = Model()

# Collect the history of W-values and b-values to plot later
Ws, bs = [], []
epochs = range(10)
for epoch in epochs:
  Ws.append(model.W.numpy())
  bs.append(model.b.numpy())
  current_loss = loss(model(inputs), outputs)

  train(model, inputs, outputs, learning_rate=0.1)
  print('Epoch %2d: W=%1.2f b=%1.2f, loss=%2.5f' %
        (epoch, Ws[-1], bs[-1], current_loss))

# Let's plot it all
plt.plot(epochs, Ws, 'r',
         epochs, bs, 'b')
plt.plot([TRUE_W] * len(epochs), 'r--',
         [TRUE_b] * len(epochs), 'b--')
plt.legend(['W', 'b', 'true W', 'true_b'])
plt.show()

Epoch  0: W=5.00 b=0.00, loss=7.92897
Epoch  1: W=4.64 b=0.35, loss=5.61977
Epoch  2: W=4.35 b=0.64, loss=4.07488
Epoch  3: W=4.11 b=0.88, loss=3.04133
Epoch  4: W=3.91 b=1.07, loss=2.34987
Epoch  5: W=3.75 b=1.23, loss=1.88727
Epoch  6: W=3.62 b=1.36, loss=1.57779
Epoch  7: W=3.51 b=1.47, loss=1.37073
Epoch  8: W=3.42 b=1.55, loss=1.23221
Epoch  9: W=3.35 b=1.62, loss=1.13954