Inferences' issue

[quannm3110]

Hi,

Is there anyone knowing how to run prediction in batch? I got run-out-of-memory problem since my data is way too big (5 million users x 5.000 items).

Thanks in advance.

[quannm3110]

I found the way to work around with it.

from scipy.sparse import csr_matrix     

# Prediction in batch
batch_size = 1000
start_batch = 0
while start_batch < n_users:

    # min() ensures that the batch bounds doesn't go past the end of the index
    end_batch = min(start_batch + batch_size, n_users)
    prediction_batched = cf_model.predict(
          csr_matrix(users_features)[start_batch:end_batch], 
          items_features)
    start_batch = end_batch

If you want to predict a single user towards a single item, you can use the following script.

 cf_model.predict(
      csr_matrix(user_features)[325403], 
      csr_matrix(item_features)[838]
 )