Readme

The library is organized around three core modules:

• qdisc.dataset: Handles the loading and management of quantum datasets, including their associated types and tuning parameters used throughout the pipeline.
• qdisc.vae: Implements the Variational Auto-Encoder (VAE) model and a Trainer wrapper that learns a low-dimensional representation of the quantum phase space.
• qdisc.sr: Provides Symbolic Regression (SR) methods to derive compact, interpretable analytical descriptors for the clusters identified in the learned representation.

The library also includes two supporting submodules:

• qdisc.nn: Contains neural network architectures used as building blocks within the pipeline.
• qdisc.clustering: Implements a clustering algorithm that can be applied to the learned representation to identify and label distinct phases.

The interplay between these modules across the full pipeline is illustrated below:

The qdisc.vae module combines data from qdisc.dataset with neural network architectures from qdisc.nn to learn a low-dimensional representation of the phase space. The qdisc.sr module then operates on this representation — and, for some approaches, on the VAE itself — to produce compact, interpretable analytical expressions.