XOR MULTILAYER PERCEPTRON：如何将训练数据的子集作为参数传递以得到预测值

Question

#imports
import tensorflow as tf
#Variables
hidden_layer1_node= 2
hidden_layer2_node= 1


X = tf.placeholder('float',[8,3])
Y = tf.placeholder('float',[8,1])

#neural model
def neural_model(x):
layer1_weight = {'weight':tf.Variable(tf.random_normal([3,hidden_layer1_node])),
                'bias':tf.Variable(tf.zeros([hidden_layer1_node]))}

layer2_weight = {'weight':tf.Variable(tf.random_normal([2,hidden_layer2_node])),
                'bias':tf.Variable(tf.zeros([hidden_layer2_node]))}


zl1 = tf.add(tf.matmul(x,layer1_weight['weight']), layer1_weight['bias'])
prediction1 = tf.sigmoid(zl1)

zl2 = tf.add(tf.matmul(prediction1,layer2_weight['weight']), layer2_weight['bias'])
return tf.sigmoid(zl2)

prediction = neural_model(X)


#cost function
def cost_function():
loss = tf.reduce_mean(-1*((Y*tf.log(prediction))+((1-Y)*tf.log(1.0-prediction))))
return loss

#Optimization
loss = cost_function()
training = tf.train.GradientDescentOptimizer(0.1).minimize(loss)

#training stage
train_x = [[0,0,0],[0,0,1],[0,1,0],[0,1,1],[1,0,0],[1,0,1],[1,1,0],[1,1,1]]
train_y = [[0],[1],[1],[0],[1],[0],[0],[1]]
epoch = 10

with tf.Session() as sess:
    sess.run(tf.global_variables_initializer())

    for i in range(epoch):
        for _ in range(5000):
            sess.run(training, feed_dict={X:train_x,Y:train_y})

        print(sess.run(loss,feed_dict={X:train_x,Y:train_y}))
    print(sess.run(prediction,feed_dict={X:train_x,Y:train_y}))

Based on the network model(assuming one understands) after it has been trained, How could you pass tensors of not just [8,3] but be able to pass [1,3] such as [0,0,1] or something. I guess I'm rephrasing my question.

Answer 1

Unfortunately, TensorFlow doesn't allow the graph to change, that means that the input (and intermediate) tensors are required to have constant size. To distinguish between training and testing, you can use shared variables as explained here: https://www.tensorflow.org/guide/variables#sharing_variables