Artm 0.10.0: transform with predict_class_ids returns empty df

[bt2901]

No description provided.

[mathemare]

Hello,

Seems like the last major version changed a lot of core functionality (for instance the way class_ids are used, I don't remember to have seen it in the index of Phi matrix in previous versions, maybe this regression is an impact)

Just to add an easily reproducible procedure to this issue:

> import os, artm, time
> os.system('echo "DOC1 |@default_class tok1:1 tok2:5 |@lab lab1\nDOC2 |@default_class tok1:2 tok3:1 |@lab lab2\nDOC3 |@default_class tok2:2 |@lab lab1" > \~/test')
> batch_vectorizer = artm.BatchVectorizer(data_path="\~/test", data_format='vowpal_wabbit', target_folder='kos'+str(time.time()))
> model_artm = artm.ARTM(num_topics=2, dictionary=batch_vectorizer.dictionary, class_ids={'@default_class': 1.0, '@lab': 2.0})
> model_artm.fit_offline(batch_vectorizer=batch_vectorizer)
> model_artm.transform(batch_vectorizer, predict_class_id="@lab")
> print(model_artm.transform(batch_vectorizer, predict_class_id="@lab"))
Empty DataFrame
Columns: [0, 1, 2]
Index: []

Thank you for this incredible project !
Matthieu