Onboarding [Before Session #1]

Before session #1, the idea is create a solid onboarding for the upcoming folks. The intuition here is that in the first 3 weeks of the group, you’d like a Core group skimmed out & onboarded with project essentials. That way, most group time can be used for creative exploration.

We’ve figured out a 3 part to help with that.

Please add the following to the landing page of your Working group, below the heading ‘Onboarding’.

Part 1

Creating a ‘solution architecture’ for your group. That is a fancy way of saying that you have to break down the intended project to its component parts.

Find below a sample.

We recommed using draw.io for this. And don’t worry if you’re not comprehensive the first time, you can always re-visit to tweak.

Note: Each box is essentially a ‘small task’ in the project. How big or small said ‘task’ is ultimately a judgement call. The one rule of thumb is ‘keep it small enough that it can be picked up & put in a project with a different dataset’.

The text near the shapes are techniques the group would like to explore.

Part 2

Create Recipes

A ‘Recipe’ is a list of resources that - when followed along - have you ‘do something’.

Put another way, you should be able to say for any Recipe.

Minimum Goal

By the end, a reviewer will have, with understanding …

Started from: [Default: Scratch]

Scratch

Succeeded at:

Building their first Neural network!

To re-emphasize, not ‘teach you to be able to do’ … but ‘have you do’.

Samples [Of the 4 common types of Recipes]:

A Recipe on obtaining domain knowledge: Satellite Image Interpretation. (Minimal domain knowledge critical for a good data scientist)

A Recipe on doing a non-coding task: Data Acquisition.

A Recipe on doing a typical coding task: Building landcover classification model.

A Recipe on learning a tool: Google Colab Essentials. (Agg Intellect groups are expected to use Colab for projects to ensure easy sharing unless a strong reason otherwise.)

For each of the shapes in your solution architecture diagram, take time to create at least 1 onboarding Recipe.

On how to create a Recipe, refer to the starter video below.

PS. Write-up assets can be made directly using the Chrome Extension now.

Part 3

Once you have created a Recipe, the next step is ensuring it in fact does what it says it does.

Essentially, whatever you ‘wrote someone can get done’ with the Recipe - we’ll have an independent reviewer do it. If there are gaps, they will coordinate to help fill.

For that reason, assign ‘Aggregate Intellect’ as the reviewer in your Recipe [as opposed to ‘auto assign reviewer’]. He will double check your work to ensure it runs robustly.

Be sure to send in all your Recipes for review at least 5 days session #1 of your group, and be active on the AISC Slack after. Your assigned reviewer/s will reach out to help fill in gaps with feedback.

Part 4

If all that seems a bit overwhelming, then don’t worry. To help save time, we’ve creating a growing ‘menu of validated recipes’ you can use as a starting point. Think building blocks for onboarding you can mix & match for your use case.

How to use?

Take this ‘Training your first Random Forest’ Recipe from our menu. A reviewer has gone through it in detail to ensure it can be trusted.

In practice, applying Random Forests to your use case might need some additional pre-processing or tweaking. And chance are, you know the additional resources / can put together a small writeup that talk about said ‘additional steps’.

Now instead of building a Recipe from scratch, that first teaches the foundations of Random Forests, then the specific kind of random forests you need, you can simply create a Recipe in Recipe.

Sample [Random Forests applied to Landcover data]

Refer to this tutorial on adding ‘recipes to recipes’.

Our Menu of Recipes

Essential Tools

Google Colab Essentials

Common Models

Training your first Random Forest

Model Optimization related

Vanilla Model Optimization & Robustness techniques