ntnsndr/protocol-bicorder
1
0
Fork 0
Code Issues 2 Pull Requests Actions Packages Projects Releases Wiki Activity

Fix inverted sign in formal/informal LDA analysis gradient

Browse Source 
The model (analysis/bicorder_model.json) maps a positive LDA score to
cluster 2 = Institutional/Bureaucratic = "formal", but ldaScoreToScale
added the score (5 + score*4/3), sending formal/institutional protocols
toward 9 (informal) and vice versa. bicorder.json defines this gradient
as 1 = formal, 9 = informal, so the score must be subtracted.

- Flip the sign: value = 5 - (ldaScore * 4/3); correct the doc comment to
  state the model's actual sign convention
- Rename calculateBureaucratic -> calculateFormalInformal and update the
  stale analysisOrder comment, matching bicorder.json's formal/informal terms

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
This commit is contained in:
Nathan Schneider
2026-06-30 16:36:01 -06:00
parent f07708f296
commit fd556d967f
1 changed files with 22 additions and 17 deletions
Download Patch File Download Diff File Expand all files Collapse all files
bicorder-app/src/App.svelte
+22 -17
View File
@@ -60,8 +60,8 @@
});
// Analysis screens (shown in both shortform and longform)
// Show useful/not-useful gradient first (index 3), then the others
const analysisOrder = [3, 0, 1, 2]; // useful/not-useful, hardness, polarization, bureaucratic
// Show the useful gradient first (index 3), then the others
const analysisOrder = [3, 0, 1, 2]; // useful, hardness, polarization, formal/informal
analysisOrder.forEach((index) => {
screens.push({ type: 'analysis', index, gradient: data.analysis[index] });
});
@@ -303,25 +303,30 @@
function ldaScoreToScale(ldaScore: number | null): number | null {
/**
* Convert LDA score to 1-9 scale.
* LDA scores typically range from -4 to +4 (8 range)
* Target scale is 1 to 9 (8 range)
* Convert LDA score to the analysis[2] "formal vs informal" 1-9 scale.
* LDA scores typically range from -4 to +4 (8 range); target is 1-9.
*
* Formula: value = 5 + (ldaScore * 4/3)
* - LDA -3 or less → 1 (bureaucratic)
* - LDA 0 → 5 (boundary)
* - LDA +3 or more → 9 (relational)
* The model's sign convention (see analysis/bicorder_model.json):
* positive LDA → cluster 2 = Institutional/Bureaucratic = "formal"
* negative LDA → cluster 1 = Relational/Cultural = "informal"
* bicorder.json defines this gradient as 1 = formal, 9 = informal, so a
* positive LDA score must map toward 1 (formal). The score is therefore
* subtracted, not added.
*
* Formula: value = 5 - (ldaScore * 4/3)
* - LDA +3 or more → 1 (formal / institutional / bureaucratic)
* - LDA 0 → 5 (boundary, characteristics of both families)
* - LDA -3 or less → 9 (informal / relational / cultural)
*/
if (ldaScore === null) return null;
// Scale: value = 5 + (ldaScore * 1.33)
const value = 5 + (ldaScore * 4.0 / 3.0);
const value = 5 - (ldaScore * 4.0 / 3.0);
// Clamp to 1-9 range and round
return Math.round(Math.max(1, Math.min(9, value)));
}
function calculateBureaucratic(): number | null {
function calculateFormalInformal(): number | null {
// Collect all diagnostic gradients with their set and gradient info
const ratings: Record<string, number> = {};
@@ -329,7 +334,7 @@
const setName = diagnosticSet.set_name;
diagnosticSet.gradients.forEach((gradient) => {
if (gradient.value !== null) {
// Create dimension name in format: SetName_left_vs_right
// Dimension name must match the model's keys: SetName_left_vs_right
const dimensionName = `${setName}_${gradient.term_left}_vs_${gradient.term_right}`;
ratings[dimensionName] = gradient.value;
}
@@ -343,10 +348,10 @@
// Get prediction from classifier (need detailed: true to get ldaScore)
const result = classifier.predict(ratings, { detailed: true });
// Convert LDA score to 1-9 scale
// Convert LDA score to the 1-9 formal/informal scale
return ldaScoreToScale(result.ldaScore);
} catch (error) {
console.error('Error calculating bureaucratic/relational score:', error);
console.error('Error calculating formal/informal score:', error);
return null;
}
}
@@ -362,8 +367,8 @@
// Polarized/Centrist
data.analysis[1].value = calculatePolarization();
} else if (index === 2) {
// Bureaucratic/Relational (LDA classifier)
data.analysis[2].value = calculateBureaucratic();
// Formal/Informal (LDA classifier)
data.analysis[2].value = calculateFormalInformal();
}
}
});