# :loudspeaker::mag::zap: April Data Study Group Catch-up Presentations :zap::mag::loudspeaker:

---

Please leave your constructive thoughts, comments and questions for each group presentation below.

Add your name to your feedback if you feel comfortable so the teams can reach out to you and discuss further.

Feedback could include:

What you found interesting :grey_question::question::grey_question:

What might they have not thought about/suggestions :grey_question::question::grey_question:

---

**Running order:**

:microscope: Odin Vision :microscope:

:books: Entale :books:

:nut_and_bolt: AMRC :nut_and_bolt:

:city_sunrise: CityMaaS :city_sunrise:

:pound: DWP :pound:

---

:microscope: **Odin Vision Questions & Comments**:microscope:
*Leave your feedback and comments below:*

* Thank you for the great presentation!
* Was there any significance in choosing 1 and 7 as the handwritten digits? — Not really, we also experimented with different pairs and got similar results
*
*

---

**:books: Entale Questions & Comments** :books:
*Leave your feedback and comments below:*

* Really interesting project - you talk a bit about user history and I was wondering the role that user demographics/characteristics play in that. Similar to what you say about "brave" versus "safe" users. There might be room for defining the users based on a number of characteristics, e.g. I think YouTube does that with things popular among young users etc. (Antonia)
* Sorry, second comment: Isn't a rabbit hole per definition a longer journey over time? i.e. a user not clicking on one podcast but continuing to click on similar podcasts over time (Antonia) *podcasts, my apologies - This is true and also something we have thought about is having some user 'content vector' that could be updated over time to try and emulate their interests changing. But also including an episode by episode recommendation which disregards the users previous listens and bases its recommendations on the previous episode, which is probably more like a YouTube rabbit hole definition.
*
* One of the things YouTube uses as a really strong signal is engagement - have you considered joining a secondary dataset (like a Twitter named entity search) to get some of that? RP - This is a good question, I'll put this in our slack group and see if anyone has a coherent answer to this!

---

:nut_and_bolt: **AMRC Questions & Comments** :nut_and_bolt:
*Leave your feedback and comments below:*

* What does your synthetic model look like? Do you have constraints on the outputs after noising? RP
*
*
*

----
:city_sunrise: **CityMaaS Questions & Comments** :city_sunrise:
*Leave your feedback and comments below:*

* Do you think the 73% yes is from people only filling in the field if they are accessible? Or why do you think the data is skewed? AD - Just to follow up on the answer as well, that's why we want to try undersampling the 'yes's rather than e.g. also marking any POIs with no data as 'no's.
*
* Interesting heatmap - knowing London, it seems like there's a definite economic element there with higher-income areas being less accessible. Have you considered adding that as a feature?
* https://pypi.org/project/imbalanced-learn/ Thank you for this link!

---
:pound: **DWP Questions & Comments** :pound:
*Leave your feedback and comments below:*

* Couldn't we ask all the questions we want to ask of the data, ask of the model instead? I mean the model that would be used to generate the synthetic data. Just a thought. DF
*
*
*

---