sktime - A Unified Framework for ML with Time Series

[mloning]

Project Lead: @mloning Mentor: @martinagvilas Project repository: https://github.com/alan-turing-institute/sktime

Welcome to OLS-2! This issue will be used to track your project and progress during the program. Please use this checklist over the next few weeks as you start Open Life Science program :tada:.

---

Week 1 (31 August - 4 September 2020): Meet your mentor!

• [x] Create an account on GitHub
• [x] Check if you have access to the HackMD notes set up for your meetings with your mentor
• [x] Prepare to meet your mentor(s) by completing a short homework provided in the HackMD notes
• [x] Complete your own copy of the open leadership self-assessment and share it to your mentor If you're a group, each teammate should complete this assessment individually. This is here to help you set your own personal goals during the program. No need to share your results, but be ready to share your thoughts with your mentor.
• [x] Make sure you know when and how you'll be meeting with your mentor.

Before Week 2 (7 - 11 September 2020): Cohort Call (Welcome to Open Life Science!)

• [x] Create an issue on the OLS-2 GitHub repository for your OLS work and share the link to your mentor.
• [x] Draft a brief vision statement using your goals: This lesson from the Open Leadership Training Series (OLTS) might be helpful
• [x] Leave a comment on this issue with your draft vision statement & be ready to share this on the call
• [x] Check the Syllabus for notes and connection info for all the cohort calls.

Before Week 3 (14 - 18 September 2020): Meet your mentor!

• [x] Look up two other projects and comment on their issues with feedback on their vision statement
• [ ] Complete this compare and contrast assignment about current and desired community interactions and value exchanges
• [x] Complete your Open Canvas (instructions, canvas)
• [x] Share a link to your Open Canvas in your GitHub issue
• [ ] Start your Roadmap
• [ ] Comment on your issue with your draft Roadmap
• [x] Suggest a cohort name at the bottom of the shared notes and vote on your favorite with a +1

Before Week 4 (21 - 25 September 2020): Cohort Call (Tooling and roadmapping for Open projects)

• [x] Look up two other projects and comment on their issues with feedback on their open canvas.

Week 5 and later

• [x] Create a GitHub repository for your project
• [x] Add the link to your repository in your issue
• [x] Use your canvas to start writing a README.md file, or landing page, for your project
• [x] Link to your README in a comment on this issue
• [x] Add an open license to your repository as a file called LICENSE.md
• [x] Add a Code of Conduct to your repository as a file called CODE_OF_CONDUCT.md
• [x] Invite new contributors to into your work!

This issue is here to help you keep track of work as you start Open Life Science program. Please refer to the OLS-2 Syllabus for more detailed weekly notes and assignments past week 4.

[mloning]

Vision statement

We want to enable understandable and accessible machine learning with time series by providing instructive documentation and by building a friendly, collaborative and inclusive community. The aim is to unify the time series analysis field by providing a common framework for multiple learning tasks, bringing together contributors from academia and the wider data science community into a joint framework and embedding best practices into the time series analysis field

[mloning]

Open canvas

https://docs.google.com/presentation/d/1Lxd9lU_7gKWdIHicLDlN1rn6ciagAqLi5AD0aR7Xg8g/edit?usp=sharing

[koudyk]

Hi @mloning, this looks like an exciting project! It would be very useful to have a package like scikit-learn for time series. After reading your vision statement, I feel like it could a few clarifications:

• What will you personally work on for OLS? After looking at the github page for sktime, I can see that it has many contributors. Are you focusing on a certain problem for OLS? On the other hand, this will be clarified in your Roadmap, so maybe it isn't necessary here.
• The idea of unifying the ecosystem might be a bit vague for those with less programming experience. Would it be worthwhile to say what's the problem with a non-unified framework, and describe how you're going to unify it?

[mloning]

Thanks @koudyk, yes we need to state the problems of the current ecosystem first. Some key problems in our view are:

• many specialised smaller packages for specific modelling approaches or learning tasks which often do not integrate with more foundational libraries like scikit-learn and are often incompatible with each other
• methodology experts who develop algorithms are often not well connected to domain experts who work with time series data

What do you think?

During OLS, I wanted to focus on three things (subject to change):

• user guide (high level documentation of sktime and intro to ML with time series)
• governance document outlining community structure and decision making processes (already written and I received feedback from my mentor)
• working on ideas how to engage and integrate other package developers to work towards a more unified ecosystem (specifying common APIs, building up best practices, common ways of doing things)

[CooperSmout]

Hi @mloning, sounds cool. By 'time series' do you mean 'time series neuroimaging data', or something else? Might want to clarify this for the reader, otherwise people will bring in their own biases (my background is cog neuro so this makes me instantly think of neuro data).

We want to enable understandable and accessible machine learning with time series by providing instructive documentation and by building a friendly, collaborative and inclusive community.

I also think your sentence structure might be better if you flipped it around, stating what you're building first and then what is the outcome, e.g.,

--> We want to build ... <community/documentation etc> ... to enable... <understanding/accessibility etc>

Just a thought :)

[baileythegreen]

@mloning Do you know of the PythonistaCafé? It's an online community of Python programmers with some very active members who do different things, including package developers. It is not free, but might be a good investment for your project.

It's probably going to be more helpful from the package development side, but I had a quick look, and there do seem to be some discussions about time series analysis, though I can't say for certain that those would be directly-relevant for you.

[mloning]

Hi @baileythegreen, thanks for the suggestion! I didn't know about it. Are you part of it? I'm not sure what the exact benefit would be over public discussions on stackoverflow and so on?

[baileythegreen]

@mloning I am a member (or I wouldn't be able to search it). Stackoverflow can be a useful resource (and has the advantage of being an open one), and I certainly use it a lot (though not only for Python stuff). The PythonistaCafé community has a membership system to help ensure exchanges are respectful and constructive, and to foster actual relationships between members. It may also mean that members are more active because they do pay something for it. There are a handful of people (really knowledgeable ones) that I know are posting and responding to stuff there like every day.

It isn't overly-expensive – $49 per quarter, so you could give it a try to see if it's a good fit.

[koudyk]

Thanks @koudyk, yes we need to state the problems of the current ecosystem first. Some key problems in our view are:

* many specialised smaller packages for specific modelling approaches or learning tasks which often do not integrate with more foundational libraries like scikit-learn and are often incompatible with each other

* methodology experts who develop algorithms are often not well connected to domain experts who work with time series data

What do you think?

During OLS, I wanted to focus on three things (subject to change):

* user guide (high level documentation of sktime and intro to ML with time series)

* governance document outlining community structure and decision making processes (already written and I received feedback from my mentor)

* working on ideas how to engage and integrate other package developers to work towards a more unified ecosystem (specifying common APIs, building up best practices, common ways of doing things)

It's more clear to me now! Sounds like you're going about unifying the package in a good way. Again, I'm not sure whether our vision statement is supposed to be the vision statement for the larger project, or for the portion that we're working on in OLS. So I'm not sure whether you should include your specific OLS goals in the vision statement. @malvikasharan can you clarify?

[malvikasharan]

@all-contributors please add @mloning for idea, content and review on other issues.

[allcontributors[bot]]

@malvikasharan

I've put up a pull request to add @mloning! :tada:

[malvikasharan]

@koudyk sorry for not seeing this earlier.

In principle, your vision statement should be for your overall project. However, it's ok to have a short-term vision if you only want to focus on what you are doing at OLS.

[mloning]

I've opened a draft PR for the roadmap and software management plan https://github.com/alan-turing-institute/sktime/pull/467