使用tf实现LR.
import tensorflow as tf import numpy as np tf.reset_default_graph() # 清空Graph
FEATURE_NUM = 8 # 特征数量 with tf.name_scope("input"): x = tf.placeholder(tf.float32, shape=[None, FEATURE_NUM]) y = tf.placeholder(tf.int32, shape=[None])
with tf.name_scope("lr"): weight_init = tf.truncated_normal(shape=[FEATURE_NUM, 1], mean=0.0, stddev=1.0) weight = tf.Variable(weight_init) bais = tf.Variable([0.0]) y_expand = tf.reshape(y,shape=[-1,1]) hypothesis = tf.sigmoid(tf.matmul(x, weight) + bais)
with tf.name_scope("loss"): y_float = tf.to_float(y_expand) likelyhood = -(y_float * tf.log(hypothesis) + (1.0 - y_float) * (tf.log(1.0 - hypothesis))) loss = tf.reduce_mean(likelyhood, axis=0)
LEARNING_RATE = 0.02 # 学习速率 with tf.name_scope("train"): optimizer = tf.train.GradientDescentOptimizer(LEARNING_RATE) training_op = optimizer.minimize(loss)
THRESHOLD = 0.5 # 判断门限 with tf.name_scope("eval"): predictions = tf.sign(hypothesis - THRESHOLD) labels = tf.sign(y_float - THRESHOLD) corrections = tf.equal(predictions, labels) accuracy = tf.reduce_mean(tf.cast(corrections, tf.float32))
init = tf.global_variables_initializer() # 初始化所有变量
EPOCH = 10 # 迭代次数 with tf.Session() as sess: sess.run(init) for i in range(EPOCH): _training_op, _loss = sess.run([training_op, loss], feed_dict={x: np.random.rand(10, 8), y: np.random.randint(2, size=10)}) _accuracy = sess.run([accuracy], feed_dict={x: np.random.rand(5, 8), y: np.random.randint(2, size=5)}) print("epoch:", i, _loss, _accuracy)
参考文章
https://www.cnblogs.com/jhc888007/p/10390282.html
