We need now to imagine how the model is saved after the training phase. We can not store the d1/d2 layer too (if precomputed=True the pickle file will explode!). Something need to be moved. I'm thinking at this piece of code:
self.known_cols = known_cols or d1_dataset.known_cols
self.unknown_cols = unknown_cols or d1_dataset.unknown_cols
self.group_cols = d1_dataset.group_cols
self.target_cols = d1_dataset.target_cols
self.feature_cols = d1_dataset.feature_colsmaybe all these information should be stored in the metadata and, during the saving/loading procedure, we will recover the important information (also the categorical and numerical scalers) for the inference phase
We need now to imagine how the model is saved after the training phase. We can not store the d1/d2 layer too (if precomputed=True the pickle file will explode!). Something need to be moved. I'm thinking at this piece of code:
maybe all these information should be stored in the metadata and, during the saving/loading procedure, we will recover the important information (also the categorical and numerical scalers) for the inference phase