Closed theodore3131 closed 4 years ago
In the README tutorial, the suggested setting for the data split is
Split the data.* files into train., dev_., and test_.* (usually 80-10-10%). While in the split_in_train_dev_test.py file, the percentage is set as TRAINING_PERCENTAGE = 90 TEST_PERCENTAGE = 0 DEV_PERCENTAGE = 10 This will cause ZeroDivisonError when evaluating the test dataset using nmt to compute its perplexity and the bleu score if a new user just follows the README instructions.
Split the data.* files into train., dev_., and test_.* (usually 80-10-10%). While in the split_in_train_dev_test.py file, the percentage is set as
TRAINING_PERCENTAGE = 90 TEST_PERCENTAGE = 0 DEV_PERCENTAGE = 10
This will cause ZeroDivisonError when evaluating the test dataset using nmt to compute its perplexity and the bleu score if a new user just follows the README instructions.
In the README tutorial, the suggested setting for the data split is