我有一个用于分类独热编码组标签的神经网络模型（组是互斥的），其结尾是（layerActivs[-1]是最后一层的激活）：
probs = sess.run(tf.nn.softmax(layerActivs[-1]),...)classes = sess.run(tf.round(probs))preds = sess.run(tf.argmax(classes))tf.round被用来将任何低概率强制设为0。如果某一观测的所有概率都低于50%，这意味着不会预测任何类。例如，如果有4个类，我们可能会有probs[0,:] = [0.2,0,0,0.4]，因此classes[0,:] = [0,0,0,0]；接着preds[0] = 0。
显然这是模糊的，因为如果我们有probs[1,:]=[.9,0,.1,0] -> classes[1,:] = [1,0,0,0] -> preds[1] = 0，结果会是一样的。这在使用Tensorflow内置的度量类时是个问题，因为这些函数无法区分没有预测和预测为类0的情况。这段代码展示了这一点：
import numpy as npimport tensorflow as tfimport pandas as pd''' prepare '''classes = 6n = 100# simulate datanp.random.seed(42)simY = np.random.randint(0,classes,n) # pretend actual datasimYhat = np.random.randint(0,classes,n) # pretend pred datatruth = np.sum(simY == simYhat)/ntabulate = pd.Series(simY).value_counts()# create placeholderslab = tf.placeholder(shape=simY.shape, dtype=tf.int32)prd = tf.placeholder(shape=simY.shape, dtype=tf.int32)AM_lab = tf.placeholder(shape=simY.shape,dtype=tf.int32)AM_prd = tf.placeholder(shape=simY.shape,dtype=tf.int32)# create one-hot encoding objectssimYOH = tf.one_hot(lab,classes)# create accuracy objectsacc = tf.metrics.accuracy(lab,prd) # real accuracy with tf.metricsaccOHAM = tf.metrics.accuracy(AM_lab,AM_prd) # OHE argmaxed to labels - expected to be correct# now setup to pretend we ran a model & generated OHE predictions all unclassedz = np.zeros(shape=(n,classes),dtype=float)testPred = tf.constant(z)''' run it all '''# setupsess = tf.Session()sess.run([tf.global_variables_initializer(),tf.local_variables_initializer()])# real accuracy with tf.metricsACC = sess.run(acc,feed_dict = {lab:simY,prd:simYhat})# OHE argmaxed to labels - expected to be correct, but is it?l,p = sess.run([simYOH,testPred],feed_dict={lab:simY})p = np.argmax(p,axis=-1)ACCOHAM = sess.run(accOHAM,feed_dict={AM_lab:simY,AM_prd:p})sess.close()''' print stuff '''print('Accuracy')print('-known truth: %0.4f'%truth)print('-on unprocessed data: %0.4f'%ACC[1])print('-on faked unclassed labels data (s.b. 0%%): %0.4f'%ACCOHAM[1])print('----------\nTrue Class Freqs:\n%r'%(tabulate.sort_index()/n))其输出为：
Accuracy-known truth: 0.1500-on unprocessed data: 0.1500-on faked unclassed labels data (s.b. 0%): 0.1100----------True Class Freqs:0 0.111 0.192 0.113 0.254 0.175 0.17dtype: float64Note freq for class 0 is same as faked accuracy...我尝试将没有预测的观测的preds值设为np.nan，但tf.metrics.accuracy抛出ValueError: cannot convert float NaN to integer；我也尝试了np.inf，但得到OverflowError: cannot convert float infinity to integer。
如何将四舍五入的概率转换为类预测，同时适当处理未预测的观测？
回答：
这个问题已经很久没有答案了，所以我在这里发布我的解决方案。我通过一个包含三个主要步骤的新函数将归属概率转换为类预测：
- 将任何NaN概率设为0
- 将任何低于
1/num_classes的概率设为0 - 使用
np.argmax()提取预测类，然后将任何未分类的观测随机设为一个类
结果得到的整数类标签向量可以传递给tf.metrics函数。以下是我的函数：
def predFromProb(classProbs): ''' Take in as input an (m x p) matrix of m observations' class probabilities in p classes and return an m-length vector of integer class labels (0...p-1). Probabilities at or below 1/p are set to 0, as are NaNs; any unclassed observations are randomly assigned to a class. ''' numClasses = classProbs.shape[1] # zero out class probs that are at or below chance, or NaN probs = classProbs.copy() probs[np.isnan(probs)] = 0 probs = probs*(probs > 1/numClasses) # find any un-classed observations unpred = ~np.any(probs,axis=1) # get the predicted classes preds = np.argmax(probs,axis=1) # randomly classify un-classed observations rnds = np.random.randint(0,numClasses,np.sum(unpred)) preds[unpred] = rnds return preds