Timeline for How to include batch size in pytorch basic example?
Current License: CC BY-SA 4.0
9 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Oct 14, 2019 at 8:15 | comment | added | MBT |
@FäridAlijani So this is a bad example for epochs. The description of t in this context would be the number of iterations / steps. So the number of times you update the weights. An epoch is usually referred making a complete iteration over your training data. As this is the case here as well t also stands for epochs since x is all the training data we have. But you are right, normally when training in epochs you would have an iterator for the epochs and an iterator that iterates over the training data. But in this minimal example this second iterator has only one iteration.
|
|
| Oct 14, 2019 at 7:57 | comment | added | doplano |
is t the number of epochs in this sample example? I believe there needs to be another for loop through training_num right after for t in range(500):
|
|
| Dec 2, 2018 at 10:30 | comment | added | MBT | @oezguensi Yes, in reality you wouldn't do that. But I guess it is just for illustration purposes how dimensions work out within a batch while keeping it simple. | |
| Nov 30, 2018 at 0:07 | comment | added | oezguensi | That seems quite useless. Why should we iterate over the same training examples again and again. Is it just the example that is wrong, because of simplicity or am I missing something? | |
| Nov 29, 2018 at 8:40 | comment | added | MBT |
@oezguensi It is N too - there is only one batch here, with batch size 64. This example iterates just 500 times over the same batch: number_of_training_examples = num_batches * batch_size, thus 1 * 64 = 64
|
|
| Nov 29, 2018 at 3:48 | comment | added | oezguensi |
If N is the batch size, where is the number of training examples defined?
|
|
| Aug 7, 2018 at 20:50 | vote | accept | DeLorean88 | ||
| Aug 7, 2018 at 20:49 | history | edited | MBT | CC BY-SA 4.0 |
added 445 characters in body
|
| Aug 7, 2018 at 20:44 | history | answered | MBT | CC BY-SA 4.0 |