Fix: numpy scalars break JSON export of results

[alexisdubs]

main_SPA() sanitizes fitting_result before writing the JSON results file, but the loop at Code-SPA/SPA.py:565-570 only converts np.ndarray values. Numpy scalars pass through untouched, so any model that stores an integer hyperparameter as np.int64 crashes when the results are saved. PLS and SPLS both store n_components this way.

Traceback (most recent call last):
  File "SPA.py", line 581, in main_SPA
    json.dump(fr2, f, indent = 4)
  ...
TypeError: Object of type int64 is not JSON serializable

The model itself trains correctly, and the pickle is written before the JSON, so results are not lost — but the run ends in a traceback and no .json file is produced.

Reproduces on NumPy 2.x. Seen on Python 3.11.9 / NumPy 2.4.6 / pandas 3.0.5 / scikit-learn 1.9.0, Windows:

import SPA
SPA.main_SPA('train.csv', test_data='test.csv', model_name=['PLS'], cv_method='KFold', K_fold=3)

Fix

Pass default= to json.dump so numpy scalars are converted via .item(). np.generic is the base class for every numpy scalar type, so this covers int64, float64, bool_ and the rest in one place rather than enumerating them.

The handler only ever sees objects the encoder was already going to reject, so output for models that currently serialize successfully is byte-for-byte unchanged. Verified by running Examples/Concrete_data.csv (80/20 split) before and after the change:

Model	test R² before	test R² after
OLS	0.5951	0.5951
LCEN	0.8841	0.8841
PLS	crash	0.5934
SPLS	crash	0.5934

🤖 Generated with Claude Code