Paper: RoughPy #904
Paper: RoughPy #904
Conversation
|
Curvenote Preview
|
There was a problem hiding this comment.
Firstly I would like to apologise to the authors as the subject of this paper is outwith my wheelhouse, and as such have provided minor comments where I believe this will help the manuscript.
Review:
This manuscript describes RoughPy, a Python package for viewing streams of data through the lens of rough paths and using the tools and techniques of signatures. The manuscript is well-written and clear, with good accompanying documentation at https://roughpy.org/.
The codebase at https://github.com/datasig-ac-uk/RoughPy seems complete with well-written documentation, testing (including automated testing through Github Actions), and regular releases. There are currently no outstanding issues (https://github.com/datasig-ac-uk/RoughPy/issues) that cause concern, however, it would be nice to merge in datasig-ac-uk/RoughPy#91 or an equivalent. The inclusion of an issue template is noted and appreciated.
I am personally unsure of the reasoning behind the tests/build (windows-2019) failure in the CI:
FAILED tests/streams/test_lie_increment_path.py::test_from_array_width_3_2_increments - AssertionError: { 1() 3/3(1) 7(2) 1(3) } != { 1() 3(1) 7(2) 1(3) }
while ubuntu-latest/macos-11 pass, but would encourage the authors to investigate a fix to this issue.
| Rough path theory gives us tools to work with sequential, ordered data in a | ||
| mathematically rigorous way, which should provide a means to overcome some of | ||
| the inherent complexity of the data. | ||
| In this paper, we introduce a new package *RoughPy* for working with sequential |
There was a problem hiding this comment.
While later in the manuscript you refer to DataSig:
DataSig website (https://datasig.ac.uk/examples).
I think you may be missing a link to the RoughPy Github page or website in this manuscript
| Traditionally, these data are tricky to work with because of the exponential | ||
| complexity and different scales of the underlying process. | ||
| Until recently, with the development of transformers and large language models, | ||
| it has been difficult to capture the long-term pattern whist also capturing the |
| ### Example | ||
| In this section we show a very simple example of how to use RoughPy to construct | ||
| a stream and compute a signature. | ||
| This example is similar to the first few steps of the tutorial found in the | ||
| RoughPy documentation.[^roughpydocs] | ||
| We refer the reader to this documentation for much more detail. | ||
| We will construct a stream in $\mathbb{R}^{26}$ by taking each letter in a word, | ||
| "scipy" in this example, as the increments of a path: | ||
| ```python | ||
| import numpy as np | ||
|
|
||
| text = "scipy" | ||
| increments = np.zeros((5, 26), dtype="int8") | ||
| for i, c in enumerate(text): | ||
| increments[i, ord(c) - 97] = 1 | ||
| ``` |
There was a problem hiding this comment.
It could be useful for someone glancing through the manuscript to know that it is pip installable!
There was a problem hiding this comment.
Thanks so much for your comments. I completely forgot to merge the code of conduct, thanks for reminding me.
The failed CI on Windows is a very frustrating bug in the Windows compilation of the GNU Multiprecision library, and it is a royal pain that comes and goes seemingly at random. I don't yet have a reason why it fails.
I shall fix the issues that you have highlighted.
If you are creating this PR in order to submit a draft of your paper, please name your PR with
Paper: <title>. An editor will then add apaperlabel and GitHub Actions will be run to check and build your paper.See the project readme for more information.
Editor: Charles Lindsey @cdlindsey
Reviewers: