search

recommenders-team / recommenders

Best Practices on Recommendation Systems
https://recommenders-team.github.io/recommenders/intro.html
MIT License
18.82k stars 3.07k forks source link

SAR store / load trained model #207

Closed loomlike closed 5 years ago

loomlike commented 5 years ago

Would be nice to have this feature. I have to re-train the model every time I restart the machine.

yueguoguo commented 5 years ago

@loomlike The model object, in Python, can be serialized into a pickle for reuse. In Spark, you can use mleap. Maybe something worth having in the O16N notebooks.

loomlike commented 5 years ago

@yueguoguo I was thinking more like Tensorflow's: tf_model.save_weights(), tf_model.load_weights() or spark's: spark_als.save(), ALSModel.load() types of commands.

What will be the limitation of using pickle and mleap, what can we get by having our own save, load function?

miguelgfierro commented 5 years ago

closing this for now, please feel free to reopen if needed

@loomlike closing for now, please feel free to reopen in case you think we should work on this