Next steps todolist

[NicolasGensollen]

Just a todo list to keep track of things we decided to do during our nilearn meeting yesterday:

General todo list:

• Send out email to registrants on the 26th of April
• Send out a poll (on Twitter a week before the hack). (@thomasbazeille will prepare the poll)
• Compile list of current sprint registrants from email s.t. we can contact them more directly with questions and also with reminders (@emdupre)
• Follow up with Chris about nibabel elements of hack + project board (@NicolasGensollen)
• Contact Matthew Brett
• Contact Oscar Esteban
• Update website to switch timing of reports, surface modelling sessions (@NicolasGensollen)
• Fix Logo on website (@NicolasGensollen)
• Add easy first issues (@NicolasGensollen and anybody willing to add some more)

Prepare slides (till Friday 30th):

• BIDS (@tsalo and maybe @effigies)
• Contribution crash course (@GaelVaroquaux)
• Community feedback on nilearn website (@GaelVaroquaux)
• HTML Reports (@NicolasGensollen and @emdupre)
• Whats new and introduction to the sprint (@NicolasGensollen and @bthirion)
• Surface Modeling (@NicolasGensollen and @jeromedockes)
• Barriers to adoption (@thomasbazeille)

[NicolasGensollen]

I am working on the materials for the whats new / intro session. Since the goal is to have something interactive and code oriented, I think the Notebook format seems like a good choice for the support.

I'm planning to do one notebook per following subject:

• GLM module
• Decoder object
• FREM object
• New plotting functions and mosaic mode
• DiFuMo atlas

For now, I've put them in a public repo under my name (but we can move them later to a common repo for all materials?) here:

https://github.com/NicolasGensollen/nilearn-dev-days-2021-introduction

It's still not finished and pretty drafty, but feel free to have a look if you have some time.
Comments and suggestions are welcome! 😃

[jeromedockes]

thanks @NicolasGensollen ! sorry for the late reply.
Indeed I think it would be best to create a folder in this repo to hold all the materials

in the meanwhile I have put a draft of what could be discussed in the session about surfaces in this gist, please let me know what you think and I can send an invitation to edit. I think the opening presentation should be very short, briefly highlighting the points that need work (surface data representation, serialization, masking, plotting), and then see what participants want to discuss.

[jeromedockes]

re. the introduction notebooks, I think they look great! I just wonder if it is not a bit too much material to cover in a relatively short session?

[NicolasGensollen]

Thanks @jeromedockes ! 👍

I think it would be best to create a folder in this repo to hold all the materials

I agree, I created a dedicated folder to hold them: https://github.com/nilearn/dev-days-2021/tree/main/Notebooks
I also added the notebooks I made so far such that everything is in the same place now. 🎉

I just wonder if it is not a bit too much material to cover in a relatively short session?

Yes, most likely. This session will be approx 40 minutes for Nilearn and 20 minutes for Nibabel so we won't have much time indeed. I was thinking about dropping detailed explanations for DecoderRegressor and FREM objects and just introduce them quickly and explain that they can be used like the Decoder object. 🤔
But even like this, we might have to cut some stuffs...

in the meanwhile I have put a draft of what could be discussed in the session about surfaces in this gist

Awesome! I had a quick look at it, and I think it looks great! 👍 I really like the possible API examples that you give, it makes you want to start coding right away 😃
Could you commit what you did to this repo such that we can edit if needed?

[jeromedockes]

thanks! I'll move it to the directory you created, but it seems I don't have write access to this repository?

[jeromedockes]

I was thinking about dropping detailed explanations for DecoderRegressor and FREM objects and just introduce them quickly and explain that they can be used like the Decoder object.

that's a good idea, they are used in the same way so saying that they exist should be enough

[NicolasGensollen]

it seems I don't have write access to this repository?

I added you with write access, I thought you had these rights by default.

[bthirion]

Thx @NicolasGensollen !
A few comments:
[decoder notebook]

• can be simplified by considering only a LeaveOneGroupOut cv, not the 5-fold.
• I'd skip the Why is the decoder object useful? section: either people realize the benefits it brings
• Provide custom estimators to the Decoder -> do you need help on that ?
• Decoder for regression: I'd skip that and point to the example.

[Difumo]

• looks OK to me. We can skip the report of the NiftiMapsMasker object

[glm]

• I don't like the first level example (I think we should basically get rid of it)

[plotting]

• on the carpet plot with tissue types, what are the grey stripes: missing data ?
• add something for plot_markers.

[reporting]

• put less images to display in NiftiMapsMasker
• GLMReporter: if possible, change the example
• the report displayed in cell 6 is not super clear. Whe should work on improving it.

[thomasbazeille]

Hello Nicolas ! Sorry to be a little late for the answer. I think having materials ready is great and always reusable (for example maybe we can post them on twitter sometimes to communicate on new features). However I feel there is way too much to cover here for the time you planned.

1. Some time should be dedicated to the introduction :
   This is also the kick-off event for the sprint so I think it's very important presenters take time to greet people and tell them how the sprint is going to work (at least 5-10 minutes) to explain them broadly the tools / aims / hours. I think you should also answer questions on the sprint itself before going to the what's new ! All in all I guess you have roughly 30 mins dedicated to this material

2. I never managed to show a notebook in less than 10 minutes if you want to cover it so I guess there are too many of them here. Here's how I would prioritize :

• Decoder example should be made way simpler, I guess just to compare how we did decoding without it and how it's simpler with Decoder Object (I can do a pass on it if it's helpful)
• Plotting is ok IMHO
• Reporting demo is very valuable, especially mentionning the feedback session and what we expect from people. Just it's currently way too long. Fewer examples would make the functionnalities clearer I guess.
• I think Difumo can just be mentionned, there's nothing much to demonstrate here so this is not priority IMHO
• I'm not sure demonstrating GLM should be a priority either. People that needed it already knew nistats and mostly need to know all those functions are now available in Nilearn

Sorry for the long answer

[NicolasGensollen]

Thanks for the answer @thomasbazeille !
Yes, I totally agree, the notebooks are way too many and way too long... 😅
We can discuss this during the meeting, but the plan for the intro/whatsnew session would be something like:

• Kickoff / intro - 10-15min - either Bertrand, Gael, or Elizabeth?
• Nilearn whatsnew - 30-35min - Nicolas and Bertrand
  • Plotting notebook (start with that since it's nice and simple) - Nicolas
  • GLM notebook (to be simplified) - Bertrand
  • Decoder notebook (to be simplified) - Nicolas
• Nibabel whatsnew - 15-20min - Chris

I think keeping the notebooks there doesn't hurt. If people are interested and want to read them after the session, they will have the opportunity to do so.

Also, the reporting notebook will be for the HTML reports session, not for the intro/whatsnew session.