Trouble re-loading Doc2Vec Model Generated In earlier Version

[fsfeng]

I trained a Doc2Vec model in an earlier version of gensim and now can not load it. I tried both Doc2Vec.load(fname) and Doc2Vec.load(fname, mmap = 'r'). Both produce the same errors:

AttributeError: Can't get attribute 'DocvecsArray' on <module 'gensim.models.doc2vec' from 'DIR/python/miniconda3/lib/python3.6/site-packages/gensim/models/doc2vec.py'>

AttributeError: 'Doc2Vec' object has no attribute 'running_training_loss'

Any help would be much appreciated. Thanks!

[menshikh-iv]

Hello, can you share your model (or install current develop branch and check again), probably this already resolved in #1974 (but I'm not sure)

[fsfeng]

Hi I tried with the develop branch install but am getting the exact same problem.

Not sure what you mean by sharing my model, but here's the code I used to create it.

from gensim import models
from gensim.models.doc2vec import TaggedDocument

class DocIterator(object):
    def __init__(self, doc_list, labels_list):
        self.labels_list = labels_list
        self.doc_list = doc_list

    def __iter__(self):
        for idx, doc in enumerate(self.doc_list):
            # print 'creating tagged document...%d' % idx
            yield TaggedDocument(words=doc.split(), tags=[self.labels_list[idx]])


iterator = DocIterator(TEXT, LABELS)

%time model = gensim.models.Doc2Vec(size=100, window=10, min_count=50, workers=4, alpha=0.025, min_alpha=0.025)
%time model.build_vocab(iterator)

print('done building vocabulary')
print('start training the model')

for epoch in range(10):
    starttime = datetime.datetime.now()
    print("start", starttime)
    print('epoch %d..' % (epoch+1))
    model.train(iterator)
    model.alpha -= 0.002
    model.min_alpha = model.alpha
    model.train(iterator)
    endtime = datetime.datetime.now()
    print("end", endtime, "took", endtime-starttime)

model.save("DIR/doc2vec_1030.model")

This was done a few months ago and now when I try to re-load the model I'm getting the error reported above.

[menshikh-iv]

@fsfeng about "sharing model" I mean load it doc2vec_1030.model to GitHub/google drive/etc and publish a link here (this needed for investigation & fixing).

What's a gensim version you are used?

CC: @manneshiva probably this is a bug, please fix it ASAP.

[marlohaering]

I encountered the same error. I trained my model with gensim version 1.0.1 and I'm trying to reload it with gensim version 3.4.0. Unfortunately, I cannot share my model.

[menshikh-iv]

@marlohaering that's expected behavior, we don't have so long backward compatibility for models.

[menshikh-iv]

@fsfeng I'm still waiting for a model from you (this needed for reproducing an error), also, what's gensim version you used for training?

[fsfeng]

Hi, I fixed this just by re-installing Gensim version 2.3.

…

On Thu, Mar 22, 2018 at 8:41 AM, Menshikh Ivan ***@***.***> wrote: @fsfeng <https://github.com/fsfeng> I'm still waiting for a model from you (this needed for reproducing an error), also, what's gensim version you used for training? — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#1977 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AKKbOITGhvpo6uG0edHhR0oRPcrcIganks5tg5vhgaJpZM4Soqlr> .

[fsfeng]

I also cannot share the model unfortunately, all is stored on a secure university virtual server.

…

On Thu, Mar 22, 2018 at 1:15 PM, Fiona Sijie Feng ***@***.***> wrote: Hi, I fixed this just by re-installing Gensim version 2.3. On Thu, Mar 22, 2018 at 8:41 AM, Menshikh Ivan ***@***.***> wrote: > @fsfeng <https://github.com/fsfeng> I'm still waiting for a model from > you (this needed for reproducing an error), also, what's gensim version you > used for training? > > — > You are receiving this because you were mentioned. > Reply to this email directly, view it on GitHub > <#1977 (comment)>, > or mute the thread > <https://github.com/notifications/unsubscribe-auth/AKKbOITGhvpo6uG0edHhR0oRPcrcIganks5tg5vhgaJpZM4Soqlr> > . >

[menshikh-iv]

@fsfeng thanks for the information, now I understand what happens.

[menshikh-iv]

fixed in #2012