Open Lulu20220 opened 10 months ago
Did you try saving the dataset as pickle? I think it should work:
import pickle
pkl_ext = '.pkl'
dataset_filename = 'lightfm_dataset_abc' + pkl_ext
# Save the dataset to the file using pickle
with open(dataset_filename, 'wb') as dataset_file:
pickle.dump(dataset, dataset_file)
And when you want to load dataset:
# Load the dataset from the specified file path using pickle
with open(dataset_filename, 'rb') as dataset_file:
dataset = pickle.load(dataset_file)
# Fit more data to the loaded dataset
# For example, you can add more user-item interactions:
# dataset.fit_partial(users=new_user_ids, items=new_item_ids)
I have a trained model saved as a pickle file. My model has both user and item features. When I load the model, I can make predictions for seen users by calling
pickled_model.predict(user_ids = [0], item_ids =[0])
But, I want to adjust it to a cold start problem, making predictions for new users. In this case, it requires to have
dataset.fit_partial()
, where thedataset
should be fitted in the old dataset. It there a way to save and retrieve the dataset? Or we need to refit it every time before callingfit_partial()