How does batch size affects number of data splitted in folds?

Question

am working on an audio classification problem. am using the urbansound8k data set that contains 8732 audio.
I know that kfold is equally splitting data into k groups. each group will be used for testing and the rest will be used for training.

so if k=4, each group will contain 2,183 data. however, this result is far away from my own result

batch_size = 1
num_folds =4
no_epochs = 10

kfold = KFold(n_splits=num_folds, shuffle=False)

for train, test in kfold.split(features, labels):

  
  model = Sequential()
  model.add(Dense(1000, activation='relu'))
  model.add(Dense(no_classes, activation='softmax'))
  



  model.compile(loss=loss_function,
                 optimizer=opt,
                 metrics=['accuracy'])

  history = model.fit(features[train], labels[train],
              batch_size=batch_size,
              epochs=no_epochs,
              verbose=verbosity,
              validation_split=validation_split,shuffle=False)

this code has these results with k=4 :
-5239 per fold when using batch size = 1
-1048 per fold, batch size = 5
-524 per fold, batch size = 10

am not understanding whats the relation between these two parameters: batch size and number of data in a fold.

am ready to share my whole code if required.

Does this answer your question? [What does KFold in python exactly do?](https://stackoverflow.com/questions/36063014/what-does-kfold-in-python-exactly-do) — Chris, Jul 14 '20 at 17:44

score 0 · Accepted Answer · answered Jul 14 '20 at 17:40

0

Well, if you are interested in the relation, they are, ignoring integer rounding, inversely proportional, i.e.

batch_size * number_of_data_in_fold = some_constant

answered Jul 14 '20 at 17:40

Captain Trojan

2,800
1
11
28

how this relation is applied to my results? 1 * 5239 = 5239 so where is the 8732? – Faref Jul 14 '20 at 17:47
If you run model.fit on N items with batch size equal to B, the epoch will go through ceil(N/B) steps. – Captain Trojan Jul 14 '20 at 17:49
Thanks a lot, I understood now. So the number which epoch iterates on is how much the weights will be updated right? – Faref Jul 17 '20 at 07:47
The number of *steps* inside an epoch tells you how *often*, not much, will the weights be updated. – Captain Trojan Jul 18 '20 at 17:35

How does batch size affects number of data splitted in folds?

1 Answers1