I am pretty much a beginner in Tensorflow and simply follow a tutorial. There is no problem with my code, but I have a question regarding the output

accuracy: 0.95614034
accuracy_baseline: 0.6666666
auc: 0.97714674
auc_precision_recall: 0.97176754
average_loss: 0.23083039
global_step: 760
label/mean: 0.33333334
loss: 6.578666
prediction/mean: 0.3428335

I would like to know what does "prediction/mean" and "label/mean" represent?


All of these could be problem specific (except maybe accuracy). Most of it is documented here:

  • accuracy: Percentage of correct number of classifications
  • accuracy_baseline: Accuracy baseline based on labels mean. This is the best the model could do by always predicting one class. (source)
  • AUC or Area Under the (ROC) Curve is quite complicated, but tells you something about the true/false positive rate. In short: the AUC is equal to the probability that a classifier will rank a randomly chosen positive instance higher than a randomly chosen negative one.
  • auc_precision_recall: Is the percentage of relevant intstances, among the retrieved instances, that have been retrieved over the total amount of relevant instances.
  • average_loss: You're usually minimizing some function, and this is likely the average value of that function given the current batches.
  • loss: The current value of the loss (as above). Either the sum of the losses, or the loss of the last batch.
  • global_step: Number of iterations.
  • label/mean and prediction/mean: Not really sure, but I suspect that if you have two classes then the label/mean is the mean of the value labels, whilst prediction/mean could be the value of the corresponding predictions. (two classes could give you a value between 0 and 1)

Not the answer you're looking for? Browse other questions tagged or ask your own question.