Charles Truong

closing for now.

Great suggestion!

We'll look into that. Thanks.

Hello, Thanks for your message. That would be a great addition to ruptures, but there are several possible implementations. For instance, do you specify beforehand which cost function is applied...

That sounds great!

Hi, indeed using Dynp(model="l2") results in a O(log(T^3)) complexity where T is the number of samples. Using precomputed cumsum vectors will definitely make it faster. The fastest way to detect...

Hi, You can try the following cost function which implements CostL2 with pre-computed cumsum signals. Hopefully, it will show the expected/optimal log-linear time complexity. The following code defines the cost...

We will not change the current implementation of CostL2 because, for small signals (below 1000 samples approx.), the naive implementation is faster. Hope this helps

Closing for now, feel free to reopen.

> new CostL2CumSum class does not compute the same segmentation as the l2 cost. Hum, I fear that for such long signals, the cumsum vectors quickly overflow but that is...