mbarbetti / pidgan

:package: GAN-based models to flash-simulate the LHCb PID detectors
GNU General Public License v3.0
3 stars 1 forks source link

Tutorials in Colab are outdated #7

Open mbarbetti opened 5 days ago

mbarbetti commented 5 days ago

Starting from PIDGAN v0.2.0 most of the classes and corresponding methods provided by the algorithms and players modules have been revisited to be compatible with Keras 3. The major difference with respect to the previous PIDGAN versions is the "role" of the build() method for the GAN algorithms: the order with which one calls the custom model for the first time and compiles it becomes important. In particular, if one build the model before having compiled it results with making the model to ignore the metrics eventually passed during compilation procedure. To avoid the problem, it is crucial to move the first model calling after the compile() method.

Such difference should be highlighted in the tutorials available at mbarbetti/pidgan-notebooks and properly commented within the notebooks prepared in replacement of a dedicated package documentation.