Open GonyRosenman opened 3 years ago
i managed to go back to the previous status of simply not being able to load the models. i think that's due to the fact my new build_model function is slightly different than how i wrote it in the first place.
the error i'm getting is the following:
WARNING:tensorflow:Inconsistent references when loading the checkpoint into this object graph. Either the Trackable object references in the Python program have changed in an incompatible way, or the checkpoint was generated in an incompatible program.
Two checkpoint references resolved to different objects (<tensorflow.python.keras.layers.core.Dense object at 0x7fdaa08d7a58> and <tensorflow.python.keras.layers.core.Dropout object at 0x7fdaa08d78d0>).
WARNING:tensorflow:Inconsistent references when loading the checkpoint into this object graph. Either the Trackable object references in the Python program have changed in an incompatible way, or the checkpoint was generated in an incompatible program.
Two checkpoint references resolved to different objects (<tensorflow.python.keras.layers.core.Dense object at 0x7fdaa08d7978> and <tensorflow.python.keras.layers.core.Dense object at 0x7fdaa08d7a58>).
was doing HP optimization with keras tuner, then my notebook erroneously got deleted.
now trying to reverse the process and i have the folder with the oracle and all the trials in my drive.
the tuner itself is recognized succesfully.
when running :
i got an error. something that i assume got to do with my new build_model function being not exactly identical to how i originally wrote it (had to extract it from my dis-functioning memory).
i tried to change more things (slight changes to the architecture of the network. to find the original one that won't cause problems) and i suspect i made some irreversible change to the oracle because it didn't load the tuner, instead started a serach and then ran into an error.
now when try to load the tuner i get the following output:
and when trying to run search:
do you recognize this situation? is it possible to reverse? most importantly, how do you load a model without the original build_model function?