Use a class-based/OOP approach for pixel and cell clustering process

[alex-l-kong]

Is your feature request related to a problem? Please describe.

NOTE: this is a long-term project that should only be completed once we've fleshed out the vast majority of the pixel and cell clustering issues and requests.

Currently, the pipeline is completely functional in its approach. There are a few problems with this:

• Excessive parameter passing: we often have to pass the names of intermediate files, and it gets cumbersome having to keep track of all of these in one notebook.
• Resource management: because we have to save intermediate files for many variables, it becomes difficult to keep track of all the main and intermediate data we create. We would like to keep some of this info stored in one shared resource, which a class allows us to do. Additionally, this may help remove the need to save a few intermediate files.
• Experiment tracking: once we start addressing this, it will be easier to use an OOP approach for computing metrics and storing previous parameters. It may play better with model-saving libraries such as joblib.

Describe the solution you'd like

We should create two classes: PixelCluster and CellCluster, that serve as hubs for each experiment. Each class should include their respective functions.

Ideally, we would piggyback off of sklearn.BaseEstimator, which includes functions such as fit, transform, and fit_transform. However, since the training is done in R and it doesn't follow a traditional Pythonic approach, we may need to think of a different approach.

[ngreenwald]

This was mostly addressed in the recent refactor.