stumpy
stumpy copied to clipboard
TDAmeritrade
STUMPY is a powerful and scalable Python library for modern time series analysis
See #1006 This PR addresses part1, 2, and 3 as discussed here: https://github.com/TDAmeritrade/stumpy/issues/1006#issuecomment-2229551395
Running the following script ``` import numpy as np import stumpy import time def test_extract_several_consensus(): Ts = [np.random.rand(n) for n in [64, 128]] Ts_ref = [T.copy() for T in Ts]...
At the SciPy 2024 conference, I learned that free GPU runners are available via the [Quansight's/MetroStar's "Open GPU Server"](https://github.com/Quansight/open-gpu-server). We may consider using this in the future
For longer time series, multi-dimensional matrix profile computation seems slow. We should explore ways to improve the algorithm. For example, if the user wants ALL dimensions (instead of a subset-`k`...
Keogh's publicly announced his [final matrix profile paper](https://www.dropbox.com/scl/fi/mt8vp7mdirng04v6llx6y/MOMP_DeskTop.pdf) [PhD Dissertation](https://escholarship.org/content/qt0pn6c98v/qt0pn6c98v.pdf) This might be interesting to explore and reproduce
In many examples of numerical instability (e.g., overflow, underflow, catastrophic cancellation, etc), we encounter issues with maintaining high precision. One way to "know" what the "right" answer is would be...
Several years ago, we considered (see #266) adding a variant of GPU-STUMP that utilized cooperative groups and that would allow us to push the multiple kernel launches onto the device....
There is a new paper [here](https://arxiv.org/pdf/2409.09298)
I think we can replace https://github.com/TDAmeritrade/stumpy/blob/423c679fdfcdd547744d2b9a2aed7565124fd6e1/stumpy/mpdist.py#L100-L106 with ``` P_ABBA = np.empty(2 * j, dtype=np.float64) if k is None: percentage = np.clip(percentage, 0.0, 1.0) k = math.ceil(percentage * (2 * Q.shape[0]))...
Metadata
Owner
Metadata
STUMPY is a powerful and scalable Python library for modern time series analysis