Is there a better way to save the model to the disk?

[fMercatili]

Hello everyone. I am a novice in this field and I really would like to thank you all for this wonderful library that is making my work way easier than I expected!
For a project, I have to do a few-shot classification with a relatively large amount of samples (27 labels with roughly 500 entries each). This process takes about 15 minutes on my computer using the sentence-transformers embeddings. I wanted to save the resulting model to the disk in order to be used again in the future. I tried the instructions of the spacy documentation (https://spacy.io/usage/saving-loading), but I observe a strange behavior: if I use the "to_disk" and "from_disk" methods the model is saved on the disk, but when it's time to load it, it seems to start the loading from zero, taking 15 minutes again to load. I also tried with pickle, but I get the following error at loading time: "AttributeError: [E047] Can't assign a value to unregistered extension attribute 'cats'. Did you forget to call the set_extension method?".
Does anybody know a way to save the model (or the spacy object) to the disk and retrieve it rapidly afterwards?

[davidberenstein1957]

#4

[davidberenstein1957]

There are ways to save and load these kind of models, but I have not gotten around to properly implementing this yet, since it will take up significant time, and I didn't have too much spare time the past few months.

[davidberenstein1957]

As a work-around, you can use the stand alone usage and simply safe the sklearn classifier as a separate component.

[davidberenstein1957]

Let me know if this makes sense and works for you. If so, it might be an easy hot-fix to include the feature of passing such a classifier within the spacy pipeline along with the embedding model.

[fMercatili]

I will give it a try in a couple of hours and let you know. Thanks a lot for your reply for now 😊

[fMercatili]

Hi! Thank you again for your help. I finally solved the problem following your advice of using the standalone usage. I saved the entire classifier object with pickle.dump() and retrieved afterward with pickle as well. It only takes a couple of seconds to load the model.
This library is pure gold btw 😍

[davidberenstein1957]

Thanks for the shoutout✌️. Awesome that iw worked! Could you share your code? Then I can formalize it and potentially use it as a base for the spacy save update.

…

On 17 Jun 2022, at 17:01, fMercatili ***@***.***> wrote:  Hi! Thank you again for your help. I finally solved the problem following your advice of using the standalone usage. I saved the entire classifier object with pickle.dump() and retrieved afterward with pickle as well. It only takes a couple of seconds to load the model. This library is pure gold btw 😍 — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.

[fMercatili]

Literally four lines of code 😁

data = {
    "furniture": ["This text is about chairs.",
               "Couches, benches and televisions.",
               "I really need to get a new sofa."],
    "kitchen": ["There also exist things like fridges.",
                "I hope to be getting a new stove today.",
                "Do you also have some ovens."]
}
classifier = classyClassifier(data=data)

with open("./classifier.pkl", "wb") as f:
    pickle.dump(classifier, f)

and then to load the model and use it:

f = open("./classifier.pkl", "rb")
classifier = pickle.load(f)
print(classifier("I am looking for kitchen appliances."))