Betti figures: Torus — trad_link

Results exist for 36 out of 36 experiments (20 missing checkpoints out of 6588 total).

Generated from analysis/betti.py. Dataset folder: torus. Homology folder: trad_link.

Asymmetric std across runs (dim × width)

Torus: β₀ vs epoch, mean across runs, faceted by dimension and width

Open this figure alone

Torus: β₁ vs epoch, mean across runs, faceted by dimension and width

Open this figure alone

Torus: β₂ vs epoch, mean across runs, faceted by dimension and width

Open this figure alone

Torus: β₃ vs epoch, mean across runs, faceted by dimension and width

Open this figure alone

Asymmetric std across runs (facet by width and depth)

Torus: β₀ vs epoch(dimension), faceted by width and depth

Open this figure alone

Torus: β₁ vs epoch(dimension), faceted by width and depth

Open this figure alone

Torus: β₂ vs epoch(dimension), faceted by width and depth

Open this figure alone

Torus: β₃ vs epoch(dimension), faceted by width and depth

Open this figure alone

Topology dynamics and early signals

Torus: mean |Δβ| per epoch (β₀)

Open this figure alone

Torus: mean |Δβ| per epoch (β₀, aggregated over width and depth)

Open this figure alone

Torus: mean |Δβ| per epoch (β₁)

Open this figure alone

Torus: mean |Δβ| per epoch (β₁, aggregated over width and depth)

Open this figure alone

Torus: mean |Δβ| per epoch (β₂)

Open this figure alone

Torus: mean |Δβ| per epoch (β₂, aggregated over width and depth)

Open this figure alone

Torus: mean |Δβ| per epoch (β₃)

Open this figure alone

Torus: mean |Δβ| per epoch (β₃, aggregated over width and depth)

Open this figure alone

Torus: mean |Δβ| per epoch (combined β indices)

Open this figure alone

Torus: mean |Δβ| per epoch (combined β indices, aggregated over width and depth)

Open this figure alone

Torus: |Δβ| vs test accuracy (β₀)

Open this figure alone

Torus: |Δβ| vs test accuracy (β₀, aggregated over width and depth)

Open this figure alone

Torus: |Δβ| vs test accuracy (β₁)

Open this figure alone

Torus: |Δβ| vs test accuracy (β₁, aggregated over width and depth)

Open this figure alone

Torus: |Δβ| vs test accuracy (β₂)

Open this figure alone

Torus: |Δβ| vs test accuracy (β₂, aggregated over width and depth)

Open this figure alone

Torus: |Δβ| vs test accuracy (β₃)

Open this figure alone

Torus: |Δβ| vs test accuracy (β₃, aggregated over width and depth)

Open this figure alone

Torus: |Δβ| vs test accuracy (combined β indices)

Open this figure alone

Torus: |Δβ| vs test accuracy (combined β indices, aggregated over width and depth)

Open this figure alone

Torus: topology stabilization vs accuracy convergence

Open this figure alone

Individual trajectories vs epoch

Torus: β₀ vs epoch, faceted by dimension and width

Open this figure alone

Torus: β₁ vs epoch, faceted by dimension and width

Open this figure alone

Torus: β₂ vs epoch, faceted by dimension and width

Open this figure alone

Torus: β₃ vs epoch, faceted by dimension and width

Open this figure alone

3D: epoch × z-rank × Betti

Torus: β₀ 3D (epoch × z-rank × Betti), by dimension

Open this figure alone

Torus: β₁ 3D (epoch × z-rank × Betti), by dimension

Open this figure alone

Torus: β₂ 3D (epoch × z-rank × Betti), by dimension

Open this figure alone

Torus: β₃ 3D (epoch × z-rank × Betti), by dimension

Open this figure alone

Topology vs performance

Torus: β₁ vs test accuracy (final epoch)

Open this figure alone

Torus: region count vs test accuracy

Open this figure alone