I got and LSTM that gives me output (4,32,32) i pass it to the Linear Layer(hidden size of LSTM, num_classes=1) and it gives me an output shape (4,32,1). I am trying to solve a wake word model for my AI assistant. I have 2 classes i want to predict from. 0 is not wake up and 1 is the wake up AI. My batch size is 32. But the output is (4,32,1). Isnt it should be 32,1 or something like that so i will know that there is one prediction for 1 audio mfcc?
1 Answers
Not quite. You need to reshape your data to (32, 1) or (1, 32) in order for your linear layer to work. You can achieve this by adding a dimension with torch.unsqueeze() or even directly with torch.view(). If you use the unsqueeze function, the new shape should be (32, 1). If you use the view function, the new shape should be (1, 32).
Related