The Data i collect contains of 3 features/signals (columns) over some period of time. to make this easy lets say 100 time = 100 seconds. for example like this.
| timestep | Speed | accelaration | position |
|---|---|---|---|
| 00:02:00 | 12.5 | 2 | 1 |
| 00:02:01 | 13.5 | 2.2 | 1 |
| 00:02:02 | 13.7 | 2.1 | 1 |
| . | . | . | |
| 00:03:40 | 13.7 | 2.1 | 1 |
Based on these Signals i want to classify the action my robot does in an lstm model. For the beginning I just want to know if the robot fell in the data i collected. To train my model i have 5 sequences in the shape of my Data above in which i know he fell. So i have 5x 100 seconds. In each sequence he fell so i label these as "fall". ( I know this is maybe an odd implementation but later on i want many different type of movements clasified)
So now lets get to my question: I have 5 sequences, each 100 seconds. Using Pytorch datasets and Dataloader, what should my batch size be? I thought it is supposed to be 100. So it would need 5 iterations to complete one batch. Is this correct? I am confused as it seems that generally you can choose and optimize your batch size but if i for example choose a batch size of
