keep copying...
Keywords: one hidden layer, dropout, ReLu activation function, adaptive learning rate Adagrad, Sofmax function
Attached code:
# Load the MNIST dataset
from tensorflow.examples.tutorials.mnist import input_data
import tensorflow as tf
mnist = input_data.read_data_sets("MNIST_data/",one_hot=True)
# Create tensorflow's default Interactive Session, so that you don't need to specify a session to perform various operations later
sex = tf.InteractiveSession ()
# input node 28*28
int_units = 784
# number of hidden layer nodes
h1_units = 300
# Bias are all initialized to 0, and weights are initialized to a truncated normal distribution with a standard deviation of 0.1
w1 = tf.Variable(tf.truncated_normal([int_units,h1_units],stddev = 0.1))
b1 = tf.Variable(tf.zeros([h1_units]))
w2 = tf.Variable(tf.truncated_normal([h1_units,10]))
b2 = tf.Variable(tf.zeros([10]))
# define the placeholder for x
x = tf.placeholder(tf.float32,[None,int_units])
# The ratio of dropout is the probability of keeping the node < 1
keep_prob = tf.placeholder(tf.float32)
# define the model structure
# Hidden layer output, use relu as activation function
hidden1 = tf.nn.relu(tf.matmul(x,w1)+b1)
# dropout of the hidden layer, keep_prob means ratio<1
hidden1_drop = tf.nn.dropout(hidden1,keep_prob)
# The softmax function takes the high probability as the output
y = tf.nn.softmax(tf.matmul(hidden1_drop,w2)+b2)
# Input placeholder for true value (label)
y_ = tf.placeholder(tf.float32,[None,10])
# loss function using cross entropy
cross_entropy = tf.reduce_mean(-tf.reduce_sum(y_*tf.log(y),reduction_indices=[1]))
# Use Adagrad optimizer with learning rate set to 0.3 (along with optimizers like Adam, Adadelta, etc.)
train_step = tf.train.AdagradOptimizer(0.3).minimize(cross_entropy)
# training steps
# Initialize computational graph variables
tf.global_variables_initializer().run()
# Iterate 3000 times, a batch size of 100, a total of 300,000 samples
# 300,000 samples, each epoch has 5,500 data, which is equivalent to 5 times (300,000/5500) more epochs
for i in range(3000):
batch_xs,batch_ys = mnist.train.next_batch(100)
# keep_prob = 0.75 means keep 75% of nodes
train_step.run({x:batch_xs,y_:batch_ys,keep_prob:0.75})
# Evaluate the accuracy of the model
# tf.argmax(input, dims), returns the subscript of the maximum value, usually used with tf.equal to calculate the model accuracy
# dims = 1 operates by row, dims=0 operates by column
# tf.equal
correct_prediction = tf.equal(tf.argmax(y,1),tf.argmax(y_,1))
# tf.cast() performs format conversion, converts the bool type of correct_prediction to float32 and then averages
accuracy = tf.reduce_mean(tf.cast(correct_prediction,tf.float32))
# Because it is the prediction part, we can directly set keep_prob to 1
print(accuracy.eval({x:mnist.test.images,y_:mnist.test.labels,keep_prob:1.0}))
unfinished...