Question

我急切地启用了执行功能，但不确定如何从测试或培训数据中找到特异性值或假阳性等值

test_dataset = tf.data.TextLineDataset(fileid + "test")
test_dataset = test_dataset.map(parse_csv)
test_dataset = test_dataset.shuffle(buffer_size=1000)
test_dataset = test_dataset.batch(32)

test_accuracy = tfe.metrics.Accuracy()

for (x, y) in test_dataset:
  prediction = tf.argmax(model(x), axis=1, output_type=tf.int32)
  test_accuracy(prediction, y)

这里是代码，我已经确定了准确性，但是由于我的测试数据，它的准确率是90％，因此需要另一种分析数据的方法

Answer 1

想办法了

truepositive = 0
falsepositive = 0
truenegative = 0
falsenegative = 0

y = y.numpy()
prediction = prediction.numpy()
for index in range(0,116):

  if prediction[index] ==1 and prediction[index] ==y[index]:
      truepositive = truepositive + 1
  if prediction[index] ==0 and prediction[index] ==y[index]:
      truenegative = truenegative + 1
  if prediction[index] ==0 and prediction[index] !=y[index]:
      falsenegative = falsenegative + 1  
  if prediction[index] ==1 and prediction[index] !=y[index]:
      falsepositive = falsepositive + 1  

print("sensitivity" + str(truepositive/(truepositive+falsenegative)))
print("specificity" + str(truenegative/(truenegative+falsepositive)) )

Tensorflow急切的假阳性/阴性假率

1 个答案: