You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
As I try to define a autoencoder by using my custom data loaded in to HDF5, I see that we must define top: "label" in any way so it forces you to define a dummy set of label values for instances. Although, it does not cause any problem in optimization, it causes very disturbing set of console outputs as
Train net output #1: label=0
Train net output #2: label=0
Train net output #3: label=0
Train net output #4: label=0
... (goes to number of your data dimension)
I solved this by only inserting this to data_layers.hpp line 184
I don't know whether it is acceptable solution but in practice it is pretty fine. If it is not a convenient solution please make HDF5DataLayer able to accept no label configuration.
The text was updated successfully, but these errors were encountered:
As I try to define a autoencoder by using my custom data loaded in to HDF5, I see that we must define top: "label" in any way so it forces you to define a dummy set of label values for instances. Although, it does not cause any problem in optimization, it causes very disturbing set of console outputs as
Train net output #1: label=0
Train net output #2: label=0
Train net output #3: label=0
Train net output #4: label=0
... (goes to number of your data dimension)
I solved this by only inserting this to data_layers.hpp line 184
virtual inline bool AutoTopBlobs() const { return true; }
I don't know whether it is acceptable solution but in practice it is pretty fine. If it is not a convenient solution please make HDF5DataLayer able to accept no label configuration.
The text was updated successfully, but these errors were encountered: