Implements the classical DTW algorithm in PyTorch, enabling multi-pattern matching and GPU acceleration.
To run the DTW module on a GPU, you can use the code below:
from pytorch_dtw import DTW
if torch.cuda.is_available():
device = torch.device("cuda")
else:
device = torch.device("cpu")
dtw = DTW(device=device).to(device)The DTW module currently supports matching one sequence against many other sequences. The inputs to the forward() method must have shape (1, n) and (k, m), where n is the length of the sequence to perform matching for, k is the number of patterns to match against, and m is the length of each pattern being matched against (n and m can be different).
For example:
key = np.random.rand(1, 50)
patterns = np.random.rand(10, 75)
key_tensor = torch.tensor(key).to(device)
patterns_tensor = torch.tensor(patterns).to(device)
costs = dtw(key, patterns)