I am using the "tensorflow" keras. ie. i did:
from tensorflow import keras
from tensorflow.keras import layers
Not sure if this is different from Keras with TF as backend. I am on TF 1.14.0 and running this on google colab.
The problem is that each time I re-created a model (or recompile), _N will be appended to the metrics. You can see this in the printout during training, and also as keys to history.history.
32206/32206 [==============================] - 4s 138us/sample - loss: 0.8918 - precision_4: 0.6396 - recall_4: 0.4613 - val_loss: 5.5533 - val_precision_4: 0.0323 - val_recall_4: 0.0492
I am not sure if this is important for Keras to work properly but these names are an inconvenience when i tried to access them in the history. I could write more code to parse them but I like to know if i can just enforce the names in the first place. Usually, when i re-instantiated the model (or recreate from functional API), i don't intend to keep around the old version (i just overwrite it into "model"). So I am just not sure if those "_N" stuff has any importance than just the names. Would Keras somehow make use of them internally? such that I may just be better off living with those names and just parse them out properly when I need to access them later.