### Training Modes with respect to dataset ###
1. batch = use a fullset of data in each iteration for training.
2. minibatch = use a subset of data in each iteration for training.
3. on-line = it is regarded as theoretical method.
### terms ###
- Iteration = number of weight update, a single iteration equals to a single minibatch.
- epoch = number of training a fullset of data.
### Example ###
# if the number of a fullset of data is 200 and minibatch size is 10,
# then minibatch will be 20.
