Performance Optimization: Reduce Redundant Computations in Polynomial Regression & Eye-Tracking Clustering

[Soumyals]

Issues Identified:
⚠️ Repeated instantiation of models (StandardScaler, PolynomialFeatures, LinearRegression) within loops.
⚠️ Suboptimal filtering & grouping of data using .groupby().apply().
⚠️ Multiple redundant DataFrame operations slowing down performance.
⚠️ Lack of efficient lookup for True vs. Predicted values causing slowdowns in dictionary operations.

Optimizations suggested:
✅ ReusedStandardScaler instance instead of reinitializing multiple times.
✅ Create fit_polynomial_regression() function to avoid redundant model instantiations.
✅ Optimize groupby.apply() using NumPy for precision and accuracy calculations.
✅ Used .query() for efficient DataFrame filtering instead of chained conditions.
✅ Optimize dictionary creation with .set_index() for fast lookups.
✅ Set n_init='auto' in KMeans for better stability and efficiency.

Expected Behavior:
The optimized implementation should produce the same output as the original.
Performance should improve significantly by reducing redundant computations.
The clustering and prediction results should remain stable across multiple runs.

[Soumyals]

Greetings,
I would like to discuss about this issue and how i have optimized the gaze tracker code.
Could you please look into it and assign this issue to me if it seems suitable.
Thank you

Regards,
Soumya

[Soumyals]

Hey @marcgc21 Could you assign this issue to me please, I would like to raise a PR.
Thank you,
Soumya LS

[marcgc21]

Hi @Soumyals, assigned to you!