def maincount(arr, cols, m):
#取数据
arr1 = arr - 0
x = arr1[:, 0:cols - 1]
y = arr1[:, cols - 1:cols]
scaler = MinMaxScaler()
#X = scaler.fit_transform(x)
X = x
#print(X.shape,X)
Y = OneHotEncoder().fit_transform(y).todense() #one-hot编码
batch_size = 50 # 使用MBGD算法，设定batch_size为109
A = Y.sum(axis=0)[0, 0]
B = Y.sum(axis=0)[0, 1]
cha = abs(A - B)
if A < B:
flag = 0
else:
flag = 1
count = 0
while count == cha:
a = random.randint(0, len(y))
if y[a][flag] == 1:
X.append(X[a])
Y.append(Y[a])
count += 1
deletecol(m, X)
X, X_test, Y, Y_test = train_test_split(X,
Y,
test_size=0.25,
random_state=0)
tf.reset_default_graph()
#单纯使用简单神经网络
x = tf.placeholder(tf.float32, [None, 200]) #28*28
y = tf.placeholder(tf.float32, [None, 2])
W_fc0 = weight_variable([20 * 10, 32],
name='w0') #上一层有7*7*64个神经元，全连接层有1024个神经元
b_fc0 = bias_variable([32]) #1024个节点
h_fc0 = tf.nn.relu(tf.matmul(x, W_fc0) + b_fc0)
W_fc1 = weight_variable([32, 32], name='w1') #上一层有7*7*64个神经元，全连接层有1024个神经元
b_fc1 = bias_variable([32]) #1024个节点
h_fc1 = tf.nn.relu(tf.matmul(h_fc0, W_fc1) + b_fc1)
#keep_prob用来表示神经元的输出概率
keep_prob = tf.placeholder(tf.float32)
h_fc1_drop = tf.nn.dropout(h_fc1, keep_prob)
#初始化第个全连接层
W_fc2 = weight_variable([32, 2], name='w2')
b_fc2 = bias_variable([2])
#计算输出
prediction = tf.nn.softmax(tf.matmul(h_fc1_drop, W_fc2) + b_fc2)
#缓存初始权值
#交叉熵代价函数
ss = 0.025
cross_entropy = tf.reduce_mean(
tf.nn.softmax_cross_entropy_with_logits(labels=y, logits=prediction) +
tf.contrib.layers.l2_regularizer(ss)(W_fc0) +
tf.contrib.layers.l2_regularizer(ss)(W_fc1) +
tf.contrib.layers.l2_regularizer(ss)(W_fc2))
#使用AdamOptimizer进行优化
train_step = tf.train.AdamOptimizer(1e-4).minimize(cross_entropy)
#结果存放在一个布尔列表中
correct_prediction = tf.equal(tf.argmax(prediction, 1),
tf.argmax(y, 1)) #argmax返回一维张量中最大的值所在的位置
#求准确率
accuracy = tf.reduce_mean(tf.cast(correct_prediction, tf.float32))
# 以下为召回率、精确率、F1等的计算
predictions = tf.argmax(prediction, 1)
actuals = tf.argmax(y, 1)
ones_like_actuals = tf.ones_like(actuals)
zeros_like_actuals = tf.zeros_like(actuals)
ones_like_predictions = tf.ones_like(predictions)
zeros_like_predictions = tf.zeros_like(predictions)
tp_op = tf.reduce_sum(
tf.cast(
tf.logical_and(tf.equal(actuals, ones_like_actuals),
tf.equal(predictions, ones_like_predictions)),
"float"))
tn_op = tf.reduce_sum(
tf.cast(
tf.logical_and(tf.equal(actuals, zeros_like_actuals),
tf.equal(predictions, zeros_like_predictions)),
"float"))
fp_op = tf.reduce_sum(
tf.cast(
tf.logical_and(tf.equal(actuals, zeros_like_actuals),
tf.equal(predictions, ones_like_predictions)),
"float"))
fn_op = tf.reduce_sum(
tf.cast(
tf.logical_and(tf.equal(actuals, ones_like_actuals),
tf.equal(predictions, zeros_like_predictions)),
"float"))
n_batch = Y.shape[0] // batch_size
saver1 = tf.train.Saver([W_fc0, W_fc1, W_fc2])
with tf.Session() as sess:
list1 = []
list2 = []
list3 = []
list4 = []
list5 = []
#sess.run(tf.global_variables_initializer())
saver1.restore(sess, './seed1')
sess.run(b_fc0.initializer)
sess.run(b_fc1.initializer)
sess.run(b_fc2.initializer)
for epoch in range(2000):
for batch in range(n_batch):
batch_xs, batch_ys = generatebatch(X, Y, batch, batch_size,
Y.shape[0])
sess.run(train_step,
feed_dict={
x: batch_xs,
y: batch_ys,
keep_prob: 0.5
})
acc = sess.run(accuracy, feed_dict={x: X, y: Y, keep_prob: 1.0})
acc1 = sess.run(accuracy,
feed_dict={
x: X_test,
y: Y_test,
keep_prob: 1.0
})
list1.append(acc1)
# print ("Iter " + str(epoch) + ", Training Accuracy= " + str(acc)+", Testing Accuracy= "+ str(acc1))
#以下为召回率、精确率、F1等的计算
#print(sess.run(predictions,feed_dict={x:X,y:Y,keep_prob:1.0}))
#print(sess.run(actuals, feed_dict={x: X, y: Y, keep_prob: 1.0}))
tp, tn, fp, fn = \
sess.run(
[tp_op, tn_op, fp_op, fn_op],
#feed_dict={x:X,y:Y,keep_prob:1.0}
feed_dict = {x: X_test, y: Y_test, keep_prob: 0.7}
)
if tp == 0 or tn == 0 or fp == 0 or fn == 0:
pass
#print(str(epoch),tp, tn, fp, fn)
else:
tpr = float(tp) / (float(tp) + float(fn))
fpr = float(fp) / (float(tp) + float(fn))
accuracynum = (float(tp) + float(tn)) / (
float(tp) + float(fp) + float(fn) + float(tn))
list2.append(accuracynum)
recall = tpr
list3.append(recall)
precisionnum = float(tp) / (float(tp) + float(fp))
list4.append(precisionnum)
f1_score = (2 *
(precisionnum * recall)) / (precisionnum + recall)
list5.append(f1_score)
#print ("Iter " + str(epoch) + ", Training Accuracy= " + str(acc)+", Testing Accuracy= "+ str(acc1))
#print( str(epoch),'Precision = ', precisionnum,' Recall = ', recall,' F1 Score = ', f1_score,' Accuracy = ', accuracynum)
print(m, '预测最大值为：', max(list1), ',accracy:', max(list2), ',recall:',
max(list3), ',precision:', max(list4), ',f1:', max(list5))
