feat: Update representation of Predictions

[mhovd]

Also removes ndarrayfrom our dependencies

[Copilot]

Pull Request Overview

This PR migrates the codebase from the ndarray library to the faer linear algebra library for matrix operations. The change introduces a custom PredictionMatrix struct to handle 2D prediction data, replacing ndarray::Array2 usage throughout the codebase.

• Replaced ndarray dependency with faer for matrix operations
• Introduced PredictionMatrix as a custom 2D container for predictions
• Updated function signatures to use faer::Mat<f64> and Vec<f64> instead of ndarray types

Reviewed Changes

Copilot reviewed 8 out of 8 changed files in this pull request and generated 2 comments.

Show a summary per file

File	Description
Cargo.toml	Removed ndarray dependency, added faer
src/simulator/likelihood/mod.rs	Introduced PredictionMatrix, updated psi() to use faer::Mat, added serialization support
src/simulator/equation/sde/mod.rs	Updated Predictions trait implementation to use PredictionMatrix
src/optimize/spp.rs	Changed from Array1<f64> to Vec<f64> and faer::Mat
src/lib.rs	Updated public exports to expose PredictionMatrix instead of PopulationPredictions
src/error/mod.rs	Removed NdarrayShapeError variant
examples/likelihood.rs	Migrated from ndarray::Array1 to native Vec operations
examples/gendata.rs	Updated to use Option-returning row() method

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

The append_column method doesn't handle the empty matrix edge case correctly. When nrows == 0, the validation at line 188 will pass (since column.len() == 0), but the loop at line 195-197 won't execute (since self.data is empty), yet self.ncols will still be incremented at line 198. This creates an inconsistent state where ncols > 0 but data is empty. Add a guard: if self.nrows == 0 { return Ok(()); }

[Copilot]

Potential panic if column is empty. This can occur when self.nrows() == 0. Add a guard to check if the column is empty before calling first().unwrap(), or return early if self.nrows() == 0 in the get_predictions method.

[github-actions]

Bencher Report

Branch	faer
Testbed	rust-moan

Click to view all benchmark results

Benchmark	Latency	Benchmark Result nanoseconds (ns) (Result Δ%)	Upper Boundary nanoseconds (ns) (Limit %)
Conditional dose modification	📈 view plot 🚷 view threshold	1,518.80 ns (+0.89%) Baseline: 1,505.46 ns	1,628.13 ns (93.28%)
Create large dataset (100 subjects)	📈 view plot 🚷 view threshold	69,492.00 ns (+0.39%) Baseline: 69,224.71 ns	75,152.87 ns (92.47%)
Data expand complex (1h intervals)	📈 view plot 🚷 view threshold	39,853.00 ns (+0.66%) Baseline: 39,591.86 ns	40,679.21 ns (97.97%)
Data expand simple (1h intervals)	📈 view plot 🚷 view threshold	695.60 ns (-0.27%) Baseline: 697.50 ns	713.32 ns (97.52%)
Data expand with additional time	📈 view plot 🚷 view threshold	53,363.00 ns (-2.11%) Baseline: 54,515.86 ns	56,850.20 ns (93.87%)
Filter exclude subjects	📈 view plot 🚷 view threshold	35,988.00 ns (-3.94%) Baseline: 37,466.00 ns	39,488.17 ns (91.14%)
Filter include subjects	📈 view plot 🚷 view threshold	10,260.00 ns (-2.21%) Baseline: 10,492.14 ns	11,203.85 ns (91.58%)
Modify all bolus doses	📈 view plot 🚷 view threshold	1,453.90 ns (+2.41%) Baseline: 1,419.71 ns	1,512.02 ns (96.16%)
Modify all infusion doses	📈 view plot 🚷 view threshold	1,510.10 ns (+0.42%) Baseline: 1,503.83 ns	1,625.69 ns (92.89%)
SubjectBuilder multi-occasion	📈 view plot 🚷 view threshold	345.05 ns (+0.97%) Baseline: 341.75 ns	371.40 ns (92.90%)
SubjectBuilder simple	📈 view plot 🚷 view threshold	96.46 ns (+2.21%) Baseline: 94.38 ns	118.50 ns (81.40%)
SubjectBuilder with covariates	📈 view plot 🚷 view threshold	433.06 ns (+1.14%) Baseline: 428.16 ns	445.34 ns (97.24%)
one_compartment	📈 view plot 🚷 view threshold	47,651.00 ns (+0.34%) Baseline: 47,491.71 ns	47,965.16 ns (99.35%)
one_compartment_covariates	📈 view plot 🚷 view threshold	53,426.00 ns (+0.81%) Baseline: 52,998.14 ns	53,648.96 ns (99.58%)
readme 20	📈 view plot 🚷 view threshold	727,570.00 ns (+0.61%) Baseline: 723,168.57 ns	729,306.41 ns (99.76%)
two_compartment	📈 view plot 🚷 view threshold	48,887.00 ns (+0.97%) Baseline: 48,416.29 ns	49,333.04 ns (99.10%)

🐰 View full continuous benchmarking report in Bencher