3 core principles to accelerate education — a powerful experiment
One day experiment having these principles applied to technology
On the last Saturday we organised together with Tech at Católica and CTIE the second edition of a remote event called HackAtHome (pun intended) about the Covid-19 dataset.
Thorly’s responsibility was the learning experience and technology that powers the day. For that we implemented some of the technology and ideas we have been developing to increase the quality of education. We based the entire event in three main principles that we believe are transversal in education and are the core of our teaching plan.
- Intuition must come first
- Documentation must be built with different options on how students consume information.
- Evaluation must be fair, transparent and purposeful.
Our approach and learning hypothesis
The premise of a student who has never seen Python nor any Data Science related topic, being able, at the end of the day, to submit predictions of the confirmed cases, over the next few days, seems absurd. “Absurd” it’s a fun place to try new ideas 🤘.
The day started with one class and, for those who know a bit about the topic, may seem obvious that one or two hours is not enough to teach the basics for this. Here’s a question we made when preparing this day.
How much do we have to teach so that people can select a column of a pandas DataFrame to be able to do some analysis?
Consider the code below as the necessary code to fulfil the task.
We have been doing research on how to evaluate what student’s are writing in their code and we have developed a tool to create an abstract syntax tree of a python script, using the AST python module (we will publish more on this soon). Here’s what we have to teach if we go the traditional way.
In a nutshell:
- Modules
- Assignment
- Expressions
- Statements
- Data structures
- Functions
It becomes clear that the traditional way won’t get us there and takes too long for people to be able to enjoy the day. We end up questioning: Do we actually need to teach all of this?
The answer is: not yet. We don’t need to teach all of this for someone who is trying to enjoy the day and make some predictions. We will not argue that we are teaching Data Science in one day, but we believe we can open the conversation on how we can start people in the field and show them the fun before all the struggle they will have to face — after the yet.
Here’s some solutions we came up with and implemented.
Intuition must come first
One can argue that programming is hard but intuition usually boil down to simple concepts — there’s even a big push to teach programming to kids. For a Data Science challenge, this should remains true. Sure, it can become highly complex and full of intricacies where experts are required but there’s also a place for generating value without this deep technical knowledge. And you can access that by providing the right intuitions on the tools you are challenging students to use throughout the day.
Teaching dynamics
We went through the dataset together with students while explaining the tools we were using for each goal. This yields a simple yet powerful result in student’s head: They are able to map a specific problem to the intuition and then to the necessary code that answers the business question.
Just loading the dataset is already an opportunity to start explaining that you will be using pandas. We start explaining that they will see a typical pattern throughout the day — .<something>
You can get people excited with important topics of their daily lives and show them that these simple tools might be enough to generate value. We used a hot topic which are fake news. We gave our students news and let them come up with the classification “Fake” or “Not fake”.
They loved it!
One can ask:
- Did you cover all the concepts behind pandas? No.
- Did you teach Python fundamentals to our students? No.
- Did you show people some tools and how simple it can be to generate some value from a dataset? Yes! And that’s the point of this experiment. Getting people excited about learning more.
Difficult topics made simple
Explaining students how the behaviour of the confirmed cases is similar to an exponential is not hard — it’s all over the news and people are in general familiar with the concept. Explaining them how to fit an exponential to the data, that’s an entirely new conversation. In general, people are familiar with different topics but they tend to struggle on the implementation part.
We used the same approach. Start with intuition. Do you know what people understand very well? That we eat more ice creams on Summer. 🍦
This is great to start teaching what students will need to fit a curve to the confirmed cases distribution. You can easily show that it’s just a matter of figuring out the m and b on y=mx + b that has the lowest possible error.
One way of learning that usually works is through experimentation and visual materials. For this, widgets are a great solution. We presented how the different models look like by changing their parameters. It is spectacular to see that while changing the parameters, student will say “stop stop stop! That’s perfect!” — it converged.
Good documentation with increasing level of detail helps students to be on track during the day
Given the decision we took on the first topic — to focus on intuitions first — we spent most of our in-class time telling students to focus on that and not on the code. That’s great but comes with a downside: we now have to provide proper documentation to allow students to learn the technical bits throughout the day.
We have been working for a while in creating technology that allows a teacher to easily create one class and let the student decide what is the preferred way to see it. We believe that by providing the freedom to choose different ways to consume information, we are providing a better experience for each student’s specific learning style.
For this HackatHome we decided to experiment a methodology of notebooks + annotation (inspired on AllenNLP). With this, we use the notebook to tell the story and share the overall mindset, make some jokes and add some gifs and we used annotation to get into the details of a specific block of code, line by line.
Fairness, transparency and purposeful evaluation make people work harder on evaluation moments
The main problem with typical evaluation is that usually it’s not fair. And fair does not mean lack of transparency or bias rather because it lacks diversity which leads to unfairness. Single type of evaluation will never be fair as it doesn’t have into account different student’s reactions to that specific type.
We try to replicate evaluation to capture the factors that matter on a day to day life of a Data Scientist. We did this by splitting it into three different parts:
- Technical evaluation on predictions — all teams submitted their predictions and the result were present on a leaderboard throughout all day, bringing healthy competition and motivation to the different teams.
- Business question — all teams had 15min to deliver answer to different business questions, making sure that they knew how to use the technical tools to derive conclusions with business value.
- Presentation — presenting their work to a technical and non-technical audience to avoid technical bias. In here, the teams had to explain not only what they did but why they were doing it so.
All of the three factor entered in a final equation to calculate the winning team of the day. All factors are present on a final leaderboard where can check their performance.
Here’s the video of one of the HackAtHome
Conclusion
A learning experience has a multitude of dimensions way beyond passing knowledge to students.
As a company, we defined 5 different areas that we have been working and we will keep developing over the next years:
- Transferring Knowledge
- Cloud-based Classroom
- Evaluation
- Engagement
- Content Management
This is why we believe HackAtHome is such a powerful experiment. We were able, in one day, to test innovative solutions in these different areas. With that, comes a great feedback from the students and the fact that we can accelerate education.
This event was a joint initiative from three different organisations and we must say that we could not be happier with everyone’s contribution and the outcome of that.
We hope to inspire you to create new approaches to education and that we can start a conversation on how we can advance education together, as a community. We need to start small but there’s no foreseeable limit to what technology can bring to education and the new experience that we will be building in the near future. For this event we had stretching, mug challenge and peer-to-peer learning, what should we try next?
Get in touch
Our mission is to rethink technological education and accelerate the transition to a cloud classroom. We are an EdTech startup developing technology with focus on adaptive learning, fair evaluation and testing, gamification and innovative learning frameworks. If you are into EdTech or just have interest in sharing some ideas or business opportunities, feel free to get in touch via ricardo@thorly.education and we can have a call!