New Chat
Ctrl
K
Kimi Code
Chat History
Mobile App

OLS vs PCA Demo

give me 3 pic
I'll run the Python script to generate the three figures for you.
Use iPython
Now let me generate the three figures:
Use iPython
Generate chart
Use iPython
Generate chart
Use iPython
Generate chart
Perfect! I've successfully generated all three figures that demonstrate why OLS lines can appear "biased/tilted" compared to PCA/TLS orthogonal directions on correlated 2D data:

Figure 1: OLS vs PCA/TLS direction on correlated 2D data

  • Shows the data cloud with 20,000 points
  • OLS line (orange) minimizes vertical SSE
  • PCA/TLS line (green) follows the major axis direction
  • Clear visual difference between the two approaches

Figure 2: OLS is asymmetric; visual 'best line' often feels orthogonal

  • Demonstrates OLS asymmetry by showing both regression directions
  • OLS regressing y on x (orange) vs. x on y (inverted, green)
  • PCA major axis (red) provides the orthogonal fit
  • Centroid marked with × symbol

Figure 3: Why OLS can look 'tilted': its center is defined vertically

  • 2-sigma covariance ellipse shows the shape of the data cloud
  • Vertical chord demonstrates how OLS centers predictions vertically
  • OLS prediction (circle) vs. PCA prediction (square) at fixed x
  • Chord midpoint shows the orthogonal center
All figures use the exact specifications you requested:
  • Tableau 10 color cycle
  • 20,000 data points with exact generation parameters
  • Proper computational methods for OLS, inverted OLS, and PCA
  • Exact figure dimensions and DPI settings
  • Grid with alpha=0.25 and legend in "best" location
The three images clearly demonstrate why OLS can appear "tilted" - it minimizes vertical error rather than orthogonal distance, causing it to differ from what we visually perceive as the "best line" through correlated data.