Can not open and score a Word2vec model generated in v3.8.0 in v4.2.0

[Wats0ns]

Problem description

Hello,

I'm trying to score a sentence in a script using gensim v4.2.0, with a model trained in v3.8.0. However, I meet the error AttributeError: 'Word2Vec' object has no attribute 'syn1'

Steps/code/corpus to reproduce

Link to the model file: https://filetransfer.io/data-package/DIMOegMO#link

from gensim.models import Word2vec
from gensim.utils import SaveLoad

sv = SaveLoad()
model = sv.load('test.mdl')
model.score(['test'], total_sentences=1)

Versions

>>> import platform; print(platform.platform())
Linux-5.4.0-135-generic-x86_64-with-Ubuntu-18.04-bionic
>>> import sys; print("Python", sys.version)
Python 3.6.9 (default, Nov 25 2022, 14:10:45) 
[GCC 8.4.0]
>>> import struct; print("Bits", 8 * struct.calcsize("P"))
Bits 64
>>> import numpy; print("NumPy", numpy.__version__)
NumPy 1.18.3
>>> import scipy; print("SciPy", scipy.__version__)
SciPy 1.4.1
>>> import gensim; print("gensim", gensim.__version__)
gensim 4.2.0
>>> from gensim.models import word2vec;print("FAST_VERSION", word2vec.FAST_VERSION)
FAST_VERSION 1

Thanks a lot !

[piskvorky]

The way to load (and save) Gensim models is using this pattern: https://radimrehurek.com/gensim/models/word2vec.html#usage-examples

It looks like you're doing something else – could that be the source of the error?

[Wats0ns]

@piskvorky
I'm currently saving the models as said in the documentation. Loading them instead with

model = Word2Vec.load("word2vec.model")

didn't change anything
The model is build with the following parameters:

model = Word2Vec(
            window=6,
            workers=6,
            sg=0,
            hs=1,
            seed=21,
            vector_size=300,
            min_count=4,
            batch_words=1000
        )
model.build_vocab(cleaned_content, progress_per=10000)
logging.root.level = logging.ERROR  # Temporarily reduce logging verbosity
model.train(
            cleaned_content,
            total_examples=model.corpus_count,
            epochs=60,
            report_delay=1
        )

[piskvorky]

I see. In that case, it should work. I'll try to check later, thanks for reporting.

[Wats0ns]

@piskvorky
Thanks a lot, in order to make it easier to debug, here is a fully working example script to reproduce the bug:

docker run -v $PWD:/app -w /app python:3.8-slim bash -c "pip install gensim==3.8.0; python -c \"from gensim.models import Word2Vec; model = Word2Vec(window=6,workers=6,sg=0,hs=1,seed=21,size=300,min_count=1,batch_words=1000); cleaned_content=['test']; model.build_vocab(cleaned_content, progress_per=10000); model.train(cleaned_content,total_examples=model.corpus_count,epochs=60,report_delay=1); model.save('test.mdl')\""

docker run -v $PWD:/app -w /app python:3.8-slim bash -c "pip install gensim==4.2.0; python -c \"from gensim.models import Word2Vec; model = Word2Vec.load('test.mdl'); model.score(['test'])\""

Please let me know if I can help you further

[piskvorky]

I found this dodgy code in Gensim that's probably the culprit:

https://github.com/RaRe-Technologies/gensim/blob/45d35eee1e7d0c9eb69ae76a99b2ed7cc35a1c0b/gensim/models/word2vec.py#L1989-L1991

This is a part of the code that loads (converts) old models. I'll have to check the commit history to see what this is about (CC @gojomo ), but could you try removing that block and see if that helps?

[piskvorky]

I traced the bug down to https://github.com/RaRe-Technologies/gensim/pull/2698/files

Looks like a botched migration; @Wats0ns can you verify removing the 3 lines above fixes the issue, and then submit a PR?

We're about to do a new release this week so we could still squeeze that in. Thanks.

[piskvorky]

@Wats0ns please confirm the fix.

[Wats0ns]

@piskvorky I confirm that it works now with your fix, thanks a lot for your fast responses and fix, very impressed !