Seemingly a bug of treating the cache. (This seems to occur if whole training data is not stored in the memory.)
Current workaround: If the memory is sufficient to store whole training data, it is good to specify -m 1000000 (sufficiently large value for the option -m) to assure using the memory as possible.
Seemingly a bug of treating the cache. (This seems to occur if whole training data is not stored in the memory.)
Current workaround: If the memory is sufficient to store whole training data, it is good to specify
-m 1000000(sufficiently large value for the option-m) to assure using the memory as possible.