Fix: Prevent crash on cross-sectional data in get_correlation_matrix and update np.nan

[divye-joshi]

Description

This PR fixes two crashes that occur when a user attempts to compute a correlation matrix or plot thrivelines using cross-sectional data (such as the fcon1000 dataset used in the tutorials).

Motivation and Context

Currently, if a user runs model.compute_correlation_matrix(train) on purely cross-sectional data, the algorithm attempts to merge subjects across different ages. Since there are no overlapping subject_ids, it creates an empty array. This eventually gets passed to scikit-learn's LinearRegression.fit(), which triggers a cryptic crash: ValueError: Found array with 0 sample(s).

Additionally, the code used np.NaN, which was permanently removed in NumPy 2.0, causing an immediate AttributeError on newer Python environments.

Changes Made

• Added a boundary check: Inserted a check at the beginning of get_correlation_matrix in pcntoolkit/math_functions/thrive.py. If the data lacks longitudinal measurements (no duplicated subject_ids), it now raises a clear, descriptive ValueError explaining that longitudinal data is required.
• NumPy 2.0 Compatibility: Replaced the outdated np.NaN with np.nan in thrive.py.

[contsili]

thanks @divye-joshi !

@likeajumprope is also working on a new updated version of how thrivelines are computed. So I would suggest to not spent more time investigating the current implementation as it has some bugs/wrong implementations

[divye-joshi]

@contsili No problem! Thanks for letting me know about this.

I am trying to understand normative velocity modeling, so I experimented with longitudinal data and encountered some bugs in thrivelines computation and plotting too. It's reassuring that it's been worked on. Do let me know if i can be of any help.