Cntk Speed Tests
On the following plots we report the times required to train/evaluate a 3-layer network (10^6 experience tuples) with different minibatch sizes. Clearly, batching tuples results in much faster training/evaluation:
- Training
On the following plots we report the times required to train/evaluate a 3-layer network (10^6 experience tuples) with different minibatch sizes. Clearly, batching tuples results in much faster training/evaluation:
