I'm using Keras to classify time series of 1000 timesteps containing one feature. In keras the input shape of an LSTM is defined as (batch_size, timesteps, features). So my input is f.e. (32,1000,1), with a batch size of 32. To construct the training X I take 32 time series of different classes and put them together in one batch. After training I want to ask the LSTM to classify ONE time series. But because I’ve specified a batch size of 25 I now always need to show it 32 time series?!
In my understanding I showed the LSTM multiple samples at once during training, but shouldn't I be able to now let it classify only one sample?