dschonholtz/arc

The Abstraction and Reasoning Corpus

arc's Issues

Make positional connections a group

In other words. Blocks that are next to each other should be grouped to create a larger whole

Groups need human readable names

Groups need to have human readable names the way groups are generated procedurally should cause a text field in the class to be populated that would make sense to a person reading it

Build a simple python flask app

In order to be able to process the data we are going to want to do the work of solving in python.
Not because the proposed plan in the solver project couldn't be done in JavaScript but because it will likely need to be optimized with machine learning algorithms to be more efficient when the scope of the data increases.

Therefore, this ticket is going to represent serving a flask application that serves the testing interface.

make repeated shapes a group

Have the flask app fetch a random or a given task and load it

Define and document the structure of the flask app for the grouping milestone

I want to make sure I spend the proper amount of time structuring this application.
I will likely need to serve a separate page that is similar to the existing testing_interface page that shows the groupings.
I will also need models
1 that fetches the data from the github api.
2 that does the actual grouping by feature

What this ticket is, is writing down exactly what that structure is and how the grouping will be done in the wiki.

Groups need to be able to compare their subgroups

Groups will often be composed of several subcomponents. Groups need an effective way of comparing these as features.

Create generic one hot encoded group object

Each frame shown will have a series of groups layered over it that the computer will then care about.
We need to make a generic way of creating these.

These masks are the same thing as our groups

Allow user to select which groups are displayed

The user should be able to see what groups were generated and then to display them over the existing frames.
The user should be able to click on checkboxes to select which groups are being selected.

Individual block subgroup needs to be added

The properties found in several of the other groups will still need to be applicable to individual blocks.
Individual blocks should be instantiated as a special instance of AbstractGroup to make sure they are still properly comparable.

Create color grouping leveraging the generic masking class

This will be the easiest grouping to build and represent and also will likely be the most powerful.

This should be a sub class of the generic mask/grouping that is discussed in the issue #6

Make jquery call that gets groups from server

This will just be the color grouper for right now

Make symmetry based group