import tensorflowas tf
def tf_confusion_metrics(predict, real):
predictions = tf.argmax(predict, 1)
actuals = tf.argmax(real, 1)
ones_like_actuals = tf.ones_like(actuals)# 维度和actuals一样的全1的张量
zeros_like_actuals = tf.zeros_like(actuals)
ones_like_predictions = tf.ones_like(predictions)
zeros_like_predictions = tf.zeros_like(predictions)
tp = tf.reduce_sum(
tf.cast(
tf.logical_and(
tf.equal(actuals, ones_like_actuals), # 实际值为P
tf.equal(predictions, ones_like_predictions)# 预测值也为P
),
"float"
)
)
tn = tf.reduce_sum(
tf.cast(
tf.logical_and(
tf.equal(actuals, zeros_like_actuals), # 实际值为N
tf.equal(predictions, zeros_like_predictions)# 预测值为N
),
"float"
)
)
fp = tf.reduce_sum(
tf.cast(
tf.logical_and(
tf.equal(actuals, zeros_like_actuals), # 实际值为N
tf.equal(predictions, ones_like_predictions)# 预测值为P
),
"float"
)
)
fn = tf.reduce_sum(
tf.cast(
tf.logical_and(
tf.equal(actuals, ones_like_actuals), # 实际值为P
tf.equal(predictions, zeros_like_predictions)# 预测值为N
),
"float"
)
)
return tp, fn, fp, tn
def ACC(P, Y):
tp, fn, fp, tn = tf_confusion_metrics(P, Y)
accuracy = (float(tp) +float(tn)) / (float(tp) +float(fp) +float(fn) +float(tn))
return accuracy
def Precision(P, Y):
tp, fn, fp, tn = tf_confusion_metrics(P, Y)
precision =float(tp) / (float(tp) +float(fp))
return precision
def Recall(P, Y):
tp, fn, fp, tn = tf_confusion_metrics(P, Y)
recall =float(tp) / (float(tp) +float(fn))
return recall
def F1_score(P, Y):
tp, fn, fp, tn = tf_confusion_metrics(P, Y)
precision =float(tp) / (float(tp) +float(fp))
recall =float(tp) / (float(tp) +float(fn))
f1_score = (2 * (precision * recall)) / (precision + recall)
return f1_score
def TPR(P, Y):
tp, fn, fp, tn = tf_confusion_metrics(P, Y)
tpr =float(tp) / (float(tp) +float(fn))
return tpr
def FPR(P, Y):
tp, fn, fp, tn = tf_confusion_metrics(P, Y)
fpr =float(fp) / (float(fp) +float(tn))
return fpr
def FNR(P, Y):
tp, fn, fp, tn = tf_confusion_metrics(P, Y)
fnr =float(fn) / (float(tp) +float(fn))
return fnr