atrcheema / seqmetrics Goto Github PK
View Code? Open in Web Editor NEWVarious errors for tabular/structured/time-series data
Home Page: https://seqmetrics.readthedocs.io
License: GNU General Public License v3.0
Various errors for tabular/structured/time-series data
Home Page: https://seqmetrics.readthedocs.io
License: GNU General Public License v3.0
Related to: openjournals/joss-reviews#6450
This is a longer review with explicit comments to the paper and the software. I have subdivided into different categories related to different aspects of the paper + software. Bullet points that I have marked with (Important) is currently blocking issues for me to recommend this paper being accepted at JOSS.
Overall the documentation is clear and concise. The API is clearly laid out and there are examples included. Some minor suggestions for improvements in the near future:
pip install SeqMetrics[all]
. Please add these additional install instructions.The paper mentions multiple frameworks for comparison and clearly lay out that SegMetrics core value is having more metrics than any other package. I would have preferred if the authors had a more nuanced view of the other frameworks e.g. there are features that metrics package from keras or torchmetrics have that SegMetrics does not. I do not consider this a huge problem, but worth considering.
Additionally, in the README.md
of the project there are multiple related projects mentioned at the bottom that are not included in the paper ( forecasting_metrics, hydroeval etc). I would like to ask the authors why these are not reference in the paper.
On the other hand, all the frameworks mentioned in the paper are not listed in the related section on the README.md. Again, minor stuff.
Overall adding such application really makes SegMetrics easy to use even for non-programmers. Really nice. I would like the authors to consider the following problems:
The app should be better documented, especially for instructions for typing/pasting values. From the code I can see that a comma separated list is expected, but this is not clear from the instructions. A simple numpy array does not work for example. Including fig2 and fig3 to the documentation and README file would definitely help.
Since it is a simple streamlit app that users can deploy themself without too much hassle, I really think the authors should consider adding instructions on how the app can be deployed by users locally (lets say that I do not trust streamlit servers with my data but still want the nice interface). This probably requires a bit of refactoring of the repository to include the app in the src
directory and the addition of pip install SegMetrics[app]
option for installing. Additionally, the paper should be updated to reflect that the webinterface can be self hosted.
The paper uses the word robust or robustness 7 times. I therefore went into this thinking that the kind of unittesting that the authors have implemented would be excellet. However, I am a bit underwelmed by the authors unittesting. As someone that is working in the exact same space, I cannot stress how important testing that a given metric is computing the correct result is. If it is not computing the correct result the core value of such package vanishes.
As an example of the problems I have:
Lines 231 to 239 in f1b8858
r2_score
metric:Lines 11 to 12 in f1b8858
When the unittests are improve then a general improvement to the CI would be needed:
README.md
numpy-1.21.6
, which is around 2 years old at this point. I see this as a overall consequence that the authors have not included some upper/lower bounds on supported numpy/scipy versions in the requirements file.I want to stress that the reason I am nitpicking on this is because the authors are claiming that the framework is very robust in the paper which is a claim not fully supported by the code.
Hi,
I believe the definition of MARE is not correct.
This should change from (ae) to (are), it means the absolute of relative errors should be taken into consideration.
Bests
A declarative, efficient, and flexible JavaScript library for building user interfaces.
๐ Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
An Open Source Machine Learning Framework for Everyone
The Web framework for perfectionists with deadlines.
A PHP framework for web artisans
Bring data to life with SVG, Canvas and HTML. ๐๐๐
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
Some thing interesting about web. New door for the world.
A server is a program made to process requests and deliver data to clients.
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
Some thing interesting about visualization, use data art
Some thing interesting about game, make everyone happy.
We are working to build community through open source technology. NB: members must have two-factor auth.
Open source projects and samples from Microsoft.
Google โค๏ธ Open Source for everyone.
Alibaba Open Source for everyone
Data-Driven Documents codes.
China tencent open source team.