• PCA Component 1 (X-axis in the plot) → The most important direction of variance in the data.
  
• PCA Component 2 (Y-axis in the plot) → The second most important direction of variance (uncorrelated with Component 1).
  

• Similar folios (in terms of text content) are closer together.
  
• Different folios are farther apart.
  

• X (PCA Component 1): Main variation in folio text.
  
• Y (PCA Component 2): Second major variation in folio text.
  
• Z (Depth): Position of the folio within the document (line number).
  

(07-02-2025, 06:53 AM)MarcoP Wrote: You are not allowed to view links. Register or Login to view.I like the 3d visualization!

What are the colors and shapes in the 2d plot? EDIT: I now understand that the colors are the clusters and the shapes the Currier classification as A (circle), B (triangle), C/undecided (square). If so, it's clear that PCA1 is basically enough to separate A from B. I don't remember if we were ever able to identify an intuitive meaning for PCA2.



At the bottom of this comment, a similar result (same V-shape) is shown for selected bigrams, rather than the entire set of unique words.



You are not allowed to view links. Register or Login to view.



Colors detailed here: You are not allowed to view links. Register or Login to view.



It's interesting that bigrams show Pharma (yellow) as a partially separated cluster.

1. High Positive Loadings (Strong Contribution to One Side)
   • Words like "shedy" (0.3327), "chedy" (0.3215), "qokeedy" (0.2618), "qokain" (0.2556), and "qokedy" (0.2362) are dominant.
     
   • Many of these words share prefixes "qok-" and "shedy".
     
   • This suggests that folios containing these words are clustered together, likely forming a distinct linguistic or stylistic group.
     
   • The presence of "ol" (0.1917) and "qol" (0.1595) could indicate a second pattern within this cluster.
     
2. Negative Loadings (Opposite Side of the Spectrum)
   • Words like "daiin" (-0.1378), "chor" (-0.1231), and "chol" (-0.1229) push in the opposite direction.
     
   • This indicates that folios dominated by "daiin" and "chor" are distinctly different from those dominated by "shedy" and "qokeedy".
     
   • "cthy" (-0.0998) and "chy" (-0.0733) are also found in the negative range, reinforcing this separation.
     
3. Patterns Noticed
   • The "qok-" prefix is heavily weighted in PCA 1, suggesting a major stylistic or lexical variation in some folios.
     
   • Words ending in "-edy" (shedy, chedy, qokeedy) seem to be characteristic of one category.
     
   • Words with "daiin" and "chor" are defining features of another category.
     

1. High Positive Loadings (One Group of Texts)
   • "ar" (0.2653), "aiin" (0.2365), "al" (0.1773), and "or" (0.1129) dominate this component.
     
   • These words tend to begin with vowels, especially A and O.
     
   • This suggests a separate folio style where vowel-initial words are more common.
     
2. Negative Loadings (Opposing Text Style)
   • "daiin" (-0.2322), "shedy" (-0.1937), "qokain" (-0.1881), "qokeedy" (-0.1519), "chor" (-0.1392) are dominant on this side.
     
   • These words are more consonant-heavy, with strong "qok-" and "shedy" patterns.
     
   • This confirms the trend that "daiin" is in a separate category from vowel-initial words.
     
3. Patterns Noticed
   • PCA 2 separates texts that favor vowel-heavy words ("aiin", "ar", "al") from those using consonant-heavy words ("qokain", "shedy", "chor").
     
   • The negative side is similar to PCA 1's positive side, reinforcing that the "qok-" and "shedy" words are a distinct cluster.
     

• PCA 1 is mostly driven by whether a folio contains "qok-" and "shedy" words or "daiin" and "chor" words.
  
• PCA 2 shows a vowel vs. consonant-heavy variation, suggesting another structural difference in the text.