Saving trained sklearn model

claZrumbats20 report abuse

I have trained a classification model in sklearn. Now I want to save this model in order to use it inside my Flask web application. I have found that this can be done using Joblib package. But the file with the model becomes extremely large, and I am afraid about how will it work in the web application. Are there any other methods of saving the machine learning models?


ElkelK report abuse

What is the problem with having a large file with the model inside web application? Do you have limited disk space on the webserver?

claZrumbats20 report abuse

Actually I have not deployed my web app yet, so I just asking about is it ok or not. I know that in web development everybody tries to minimize the size of data, files, etc.

ElkelK report abuse

This is true if you need to pass data over the network. For example, if you need to send it from back-end to front-end, or to some 3rd party API. But if the model will lay statically on the server, it doesn't matter.

Radek report abuse

However, in order to save disk space, you can think about how to minimize the size of the file. Try using the pickle library, for example. This is a standard way for serializing Python objects.

ElkelK report abuse

As far as I know, joblib is better for saving trained sklearn models (but I am not sure). Did you try to use compression? Joblib allows models compression. You can even set the extent of compression. But the more compressed files can require more time to load them before inference, and this can become a problem for the web application, where the speed is what really important.

Kiskris report abuse

Thanks, everybody. I have tried joblib with compression=3, and this works very well! The speed of loading is decent, and the size of the file becomes significantly smaller.

Add Answer

Need support?

Just drop us an email to ... Show more