Enigma - 2019-09-30

My setting for multi-core training

$CFG_NPART = 8;
$CFG_QUEUE_TYPE = "Queue::POSIX";

Training Scheme
feature extraction --> LDA-MLLT --> CI training --> CD training

CASE LDA-MLLT CI training CD training
case: 1 1 1 1
case: 2 8 8 8
case: 3 1 8 8

--- single-core (1) / multi-core (8) ---

From all 3 case of training:
- Word acc case 3 > case 1 >case 2 (although the results are slightly different but it's different)
- LDA-MLLT - Word acc in single-core training is better than multi-core training
- CI and CD - Word acc in multi-core training is better than single-core training

Why are the training results different? During a single-core training and multi-core training.

 

Last edit: Enigma 2019-09-30