Finding

garrett-2023-visual-behavior.yes-no-change-detection

Garrett ME et al. Stimulus novelty uncovers coding diversity in visual cortical circuits. bioRxiv, 2023. · Allen Visual Behavior change detection · hit rate by condition

Observed curve and fits

Fit diagnostics

Caveats on these fits (2 terms)

Each tag below is attached to one or more candidate fits for this finding. The chip column in the table flags which row carries which tag.

yes/no SDT
Signal-detection fit uses a go/no-go yes/no variant with one d-prime and one criterion, not a 2AFC choice rule.
condition-rate null
One-parameter Bernoulli baseline over condition levels; useful for categorical or ordinal hit-rate curves without a sensory evidence axis.
Variant AIC Δ AIC RMSE Caveats
sdt-yes-no winner 282.9 0.0 3.65e-8
  • yes/no SDT
bernoulli-condition-rate 341.4 58.5 0.4212
  • condition-rate null
Full diagnostics (logL · n · params · max abs error)
Variant logL n Free params Max |error| Predicted points
sdt-yes-no -139.5 270 2 5.12e-8 2
bernoulli-condition-rate -169.7 270 1 0.5895 2

Residuals

Fit parameters and provenance

sdt-yes-no.garrett-2023-visual-behavior.yes-no-change-detection

dirty fit

criterion=1.352 · d_prime=2.067

Method
scipy.optimize.minimize
Commit
3ec74b4
Predicted points
2
BIC
290.1

Fitted with scipy.optimize.minimize via behavtaskatlas.model_fits.sdt-yes-no; success=True; CONVERGENCE: NORM OF PROJECTED GRADIENT <= PGTOL

bernoulli-condition-rate.garrett-2023-visual-behavior.yes-no-change-detection

dirty fit

response_rate=0.678

Method
manual
Commit
3ec74b4
Predicted points
2
BIC
345.0

Fitted with manual via behavtaskatlas.model_fits.bernoulli-condition-rate; success=True; closed-form binomial MLE

Observed points

Signal state p_go n
0.000 0.0882 34
1.000 0.7627 236

Take it with you

Cover sheet

Self-contained Markdown for citation, slides, or notebooks

The cover sheet pins the finding to the atlas commit and includes the observed points, fit ranking, caveats, and provenance — everything needed to drop into a paper or notebook without losing the trail back to the deploy.