how to read tensorflow confusion matrix rows and columns

Question

For my 2 classes ( 1 = [0, 1] and 0 = [1, 0] ) CNN model I use tf.confusion_matrix to finding a confusion matrix for the model. one of my results is like below for validation set:

[ [1800  17] 
  [283  600] ]

after doing some search I see more than one type of reading, some of them say [[TN FP][FN TP]] , but some others read it in this way [[TP FP][FN TN]] , I am confused which one is right for my case? please give me an answer that depends on scientific research if you can.

Answer 1

The truth is behind the code ;) https://github.com/tensorflow/tensorflow/blob/master/tensorflow/python/ops/confusion_matrix.py

Class labels are expected to start at 0. For example, if num_classes is 3, then the possible labels would be [0, 1, 2] . Note that the possible labels are assumed to be [0, 1, 2, 3, 4] , resulting in a 5x5 confusion matrix.

So better don't pass one hot tensors to the function ;) (tf.argmax might be a good friend here)

This means that the first element (row 0 col 0) corresponds with the number of elements that have been properly classified for class 0.

Row 0 col 1 will correspond with the missclassified elements of the class 0 and so on.