MemoryError: unable to allocate array with shape and data type float32 while using word2vec in python

Question

I am trying to train the word2vec model from Wikipedia text data, for that I am using following code.

import logging
import os.path
import sys
import multiprocessing

from gensim.corpora import  WikiCorpus
from gensim.models import Word2Vec
from gensim.models.word2vec import LineSentence


if __name__ == '__main__':
    program = os.path.basename(sys.argv[0])
    logger = logging.getLogger(program)

    logging.basicConfig(format='%(asctime)s : %(levelname)s : %(message)s')
    logging.root.setLevel(level=logging.INFO)
    logger.info("running %s" % ' '.join(sys.argv))

    # check and process input arguments

    if len(sys.argv) < 3:
        print (globals()['__doc__'])
        sys.exit(1)
    inp, outp = sys.argv[1:3]

    model = Word2Vec(LineSentence(inp), size=400, window=5, min_count=5, workers=multiprocessing.cpu_count())

    # trim unneeded model memory = use (much) less RAM
    model.init_sims(replace=True)

    model.save(outp)

But after 20 minutes of program running, I am getting following error

Error message

Answer 1

Ideally, you should paste the text of your error into your question, rather than a screenshot. However, I see the two key lines:

<TIMESTAMP> : INFO : estimated required memory for 2372206 words and 400 dimensions: 8777162200 bytes
...
MemoryError: unable to allocate array with shape (2372206, 400) and data type float32

After making one pass over your corpus, the model has learned how many unique words will survive, which reports how large of a model must be allocated: one taking about 8777162200 bytes (about 8.8GB). But, when trying to allocate the required vector array, you're getting a MemoryError, which indicates not enough computer addressable-memory (RAM) is available.

You can either:

1. run where there's more memory, perhaps by adding RAM to your existing system; or
2. reduce the amount of memory required, chiefly by reducing either the number of unique word-vectors you'd like to train, or their dimensional size.

You could reduce the number of words by increasing the default min_count=5 parameter to something like min_count=10 or min_count=20 or min_count=50. (You probably don't need over 2 million word-vectors – many interesting results are possible with just a vocabulary of a few tens-of-thousands of words.)

You could also set a max_final_vocab value, to specify an exact number of unique words to keep. For example, max_final_vocab=500000 would keep just the 500000 most-frequent words, ignoring the rest.

Reducing the size will also save memory. A setting of size=300 is popular for word-vectors, and would reduce the memory requirements by a quarter.

Together, using size=300, max_final_vocab=500000 should trim the required memory to under 2GB.