We need to batch data.

[taeguk]

Using all train data or test data to run model can be heavy.
It can occurs resource error because of memory capability.
So, We need to use BATCH SIZE.

For example, the number of train data is 100000.
We can batch 100 times with BATCH_SIZE = 1000.

See https://github.com/taeguk/tensorflow-study/blob/master/MNIST/cnn.py#L159,L171

[taeguk]

If BATCH SIZE is big, resource error can be occured.
If BATCH SIZE is small, train and test are so slow.

So, we have to determine proper BATCH SIZE.
I think there is no clear way.

def find_proper_batch_size():
    lh = 1
    rh = DATA_LEN
    batch_size = 0
    while lh <= rh:
        mid = (lh + rh) // 2
        have a try to make model for checking mid is available batch size.
        if available:
            lh = mid + 1
            batch_size = mid
        else:
            rh = mid - 1
    batch_size = min(batch_size * 9 // 10 - 1, 0)    # for stable batch_size
    if batch_size < 5:
        raise BatchSizeException
    return batch_size

But, I think we can consider this way.

Hmm....
But, it can be burdensome job for us that are busy.

[ovekyc]

Its true we have to condiser about that
I ran the code with mnist 10000 rows without batch, and my mac was really stressful about that.
So we have to introduce batch size
But i think not now, we need to concentrate building whole project first

[taeguk]

Okay, I don't think it is emergency now, too.
Let's think about this issue later