优化器不会优化任何变量

Question

我想解决我的优化问题，但优化器不工作。我希望得到与给定值相差最小的解的数组，但我计算的softmax函数始终是1，因为权重和偏差在迭代中不会更新。它们都是零的张量。我如何解决这个问题？

#KI-Model
x = tf.placeholder(tf.float32, [None, 5], name='input') #x_1-x_5


#Init
W = tf.Variable(tf.zeros([5,1]), dtype=tf.float32)
b = tf.Variable(tf.zeros([1]), dtype=tf.float32)

#Sigmoid
y = tf.nn.softmax(tf.matmul(x, W) + b) 

#Training
y_tensor = tf.placeholder(tf.float32, [None, 1], name='output')

loss = y-y_tensor
cost = tf.square(loss)
optimizer = tf.train.GradientDescentOptimizer(0.003).minimize(cost)


#Start
session = tf.Session() 
init = tf.global_variables_initializer()
session.run(init)


#init first 1000 training_batches
for i in range(1000):
    batch_xs.append([dataA[i], dataB[i], dataC[i], dataD[i],
              dataE[i]])
    batch_ys.append([solution[i]])

for i in range(10000):

    session.run(optimizer, feed_dict={x:batch_xs, y_tensor:batch_ys})

print(session.run(y, feed_dict={x:batch_xs, y_tensor:batch_ys}) )

Answer 1

你没有计算和应用渐变。缺少以下几行：

gradients = optimizer.compute_gradients(loss=cost)
train_step = optimizer.apply_gradients(grads_and_vars=gradients)

您还需要在每次迭代中运行训练步骤，如下所示：

session.run(train_step, feed_dict={x:batch_xs, y_tensor:batch_ys})

完整的正确代码：

#KI-Model
x = tf.placeholder(tf.float32, [None, 5], name='input') #x_1-x_5


#Init
W = tf.Variable(tf.zeros([5,1]), dtype=tf.float32)
b = tf.Variable(tf.zeros([1]), dtype=tf.float32)

#Sigmoid
y = tf.nn.softmax(tf.matmul(x, W) + b)

#Training
y_tensor = tf.placeholder(tf.float32, [None, 1], name='output')

loss = y-y_tensor
cost = tf.square(loss)
optimizer = tf.train.GradientDescentOptimizer(0.003)
gradients = optimizer.compute_gradients(loss=cost)
train_step = optimizer.apply_gradients(grads_and_vars=gradients)

#Start
session = tf.Session()
init = tf.global_variables_initializer()
session.run(init)


#init first 1000 training_batches
for i in range(1000):
    batch_xs.append([dataA[i], dataB[i], dataC[i], dataD[i],
              dataE[i]])
    batch_ys.append([solution[i]])

for i in range(10000):

    session.run(train_step, feed_dict={x:batch_xs, y_tensor:batch_ys})

print(session.run(y, feed_dict={x:batch_xs, y_tensor:batch_ys}) )