Changing linear regression back to log based

analysis/base_loss/centroid_convergence_plotter.py

@@ -43,7 +43,7 @@ def replicate_base_loss(theta_array, desired_theta, base_key):
 # ---------------------------------------------------------------------
 # Helper: safe regression function (handles log transforms, filters invalid data)
 # ---------------------------------------------------------------------
-def safe_compute_best_fit(x, y, log_x=False, log_y=False):
+def safe_compute_best_fit(x, y, log_x=False, log_y=True):
     """
     Computes a best-fit line using linear regression, optionally on log(x) and/or log(y).
     Returns:
@@ -215,7 +215,7 @@ def process_condition(condition_name):
     ax_top.set_ylabel(f"Final Loss @ epoch {final_epoch}")
     ax_top.set_title(f"{condition_name}: Final Loss vs. Exponent")
     # best-fit line
-    xs, y_fit, slope, intercept, R2 = safe_compute_best_fit(df["Exponent"], df["final_loss"], log_x=False, log_y=False)
+    xs, y_fit, slope, intercept, R2 = safe_compute_best_fit(df["Exponent"], df["final_loss"], log_x=False, log_y=True)
     if xs is not None:
         ax_top.plot(xs, y_fit, "k--", label=f"slope={slope:.3f}, int={intercept:.3f}, R²={R2:.3f}")
     ax_top.legend(fontsize=8)
@@ -236,7 +236,7 @@ def process_condition(condition_name):
     ax_bot.set_ylabel("Epochs to Convergence")
     ax_bot.set_title(f"{condition_name}: Convergence vs. Exponent")
     # best-fit line
-    xs, y_fit, slope, intercept, R2 = safe_compute_best_fit(df["Exponent"], df["epochs_to_convergence"], log_x=False, log_y=False)
+    xs, y_fit, slope, intercept, R2 = safe_compute_best_fit(df["Exponent"], df["epochs_to_convergence"], log_x=False, log_y=True)
     if xs is not None:
         ax_bot.plot(xs, y_fit, "k--", label=f"slope={slope:.3f}, int={intercept:.3f}, R²={R2:.3f}")
     ax_bot.legend(fontsize=8)

analysis/base_loss_learning_rate_sweep/centroid_convergence_plotter.py

@@ -43,7 +43,7 @@ def replicate_base_loss(theta_array, desired_theta, base_key):
 # ---------------------------------------------------------------------
 # Helper: safe regression function (handles log transforms, filters invalid data)
 # ---------------------------------------------------------------------
-def safe_compute_best_fit(x, y, log_x=False, log_y=False):
+def safe_compute_best_fit(x, y, log_x=False, log_y=True):
     """
     Computes a best-fit line using linear regression, optionally on log(x) and/or log(y).
     Returns:
@@ -215,7 +215,7 @@ def process_condition(condition_name):
     ax_top.set_ylabel(f"Final Loss @ epoch {final_epoch}")
     ax_top.set_title(f"{condition_name}: Final Loss vs. Exponent")
     # best-fit line
-    xs, y_fit, slope, intercept, R2 = safe_compute_best_fit(df["Exponent"], df["final_loss"], log_x=False, log_y=False)
+    xs, y_fit, slope, intercept, R2 = safe_compute_best_fit(df["Exponent"], df["final_loss"], log_x=False, log_y=True)
     if xs is not None:
         ax_top.plot(xs, y_fit, "k--", label=f"slope={slope:.3f}, int={intercept:.3f}, R²={R2:.3f}")
     ax_top.legend(fontsize=8)
@@ -236,7 +236,7 @@ def process_condition(condition_name):
     ax_bot.set_ylabel("Epochs to Convergence")
     ax_bot.set_title(f"{condition_name}: Convergence vs. Exponent")
     # best-fit line
-    xs, y_fit, slope, intercept, R2 = safe_compute_best_fit(df["Exponent"], df["epochs_to_convergence"], log_x=False, log_y=False)
+    xs, y_fit, slope, intercept, R2 = safe_compute_best_fit(df["Exponent"], df["epochs_to_convergence"], log_x=False, log_y=True)
     if xs is not None:
         ax_bot.plot(xs, y_fit, "k--", label=f"slope={slope:.3f}, int={intercept:.3f}, R²={R2:.3f}")
     ax_bot.legend(fontsize=8)

analysis/time_weighting/centroid_convergence_plotter.py

@@ -127,7 +127,7 @@ def find_convergence_epoch(data_dict, desired_theta, threshold):
 # ---------------------------------------------------------------------
 # Helper: Compute best-fit line (with R^2) for scatter plot data.
 # ---------------------------------------------------------------------
-def safe_compute_best_fit(x, y, log_x=False, log_y=False):
+def safe_compute_best_fit(x, y, log_x=False, log_y=True):
     """
     Computes the best-fit line using linear regression.
     Filters out NaN and non-positive values if log transforms are used.
@@ -256,7 +256,7 @@ for cond_name in all_subdirs:
     axes[0].set_ylabel(f"Loss at Epoch {final_epoch}")
     axes[0].set_yscale("log")
     axes[0].set_title(f"{cond_name}: Loss vs. $t_{{median}}$")
-    xs, y_fit, slope, intercept, R2 = safe_compute_best_fit(df["t_median"], df["final_loss"], log_x=False, log_y=False)
+    xs, y_fit, slope, intercept, R2 = safe_compute_best_fit(df["t_median"], df["final_loss"], log_x=False, log_y=True)
     if xs is not None:
         axes[0].plot(xs, y_fit, "k--", label=f"Fit: slope={slope:.3f}, int={intercept:.3f}\n$R^2$={R2:.3f}")
     axes[0].legend(fontsize=8)
@@ -273,7 +273,7 @@ for cond_name in all_subdirs:
     axes[1].set_ylabel("Epochs to Convergence")
     axes[1].set_yscale("log")
     axes[1].set_title(f"{cond_name}: Convergence vs. $t_{{median}}$")
-    xs, y_fit, slope, intercept, R2 = safe_compute_best_fit(df["t_median"], df["epochs_to_convergence"], log_x=False, log_y=False)
+    xs, y_fit, slope, intercept, R2 = safe_compute_best_fit(df["t_median"], df["epochs_to_convergence"], log_x=False, log_y=True)
     if xs is not None:
         axes[1].plot(xs, y_fit, "k--", label=f"Fit: slope={slope:.3f}, int={intercept:.3f}\n$R^2$={R2:.3f}")
     axes[1].legend(fontsize=8)

analysis/time_weighting_learning_rate_sweep/centroid_convergence_plotter.py

@@ -127,7 +127,7 @@ def find_convergence_epoch(data_dict, desired_theta, threshold):
 # ---------------------------------------------------------------------
 # Helper: Compute best-fit line (with R^2) for scatter plot data.
 # ---------------------------------------------------------------------
-def safe_compute_best_fit(x, y, log_x=False, log_y=False):
+def safe_compute_best_fit(x, y, log_x=False, log_y=True):
     """
     Computes the best-fit line using linear regression.
     Filters out NaN and non-positive values if log transforms are used.
@@ -256,7 +256,7 @@ for cond_name in all_subdirs:
     axes[0].set_ylabel(f"Loss at Epoch {final_epoch}")
     axes[0].set_yscale("log")
     axes[0].set_title(f"{cond_name}: Loss vs. $t_{{median}}$")
-    xs, y_fit, slope, intercept, R2 = safe_compute_best_fit(df["t_median"], df["final_loss"], log_x=False, log_y=False)
+    xs, y_fit, slope, intercept, R2 = safe_compute_best_fit(df["t_median"], df["final_loss"], log_x=False, log_y=True)
     if xs is not None:
         axes[0].plot(xs, y_fit, "k--", label=f"Fit: slope={slope:.3f}, int={intercept:.3f}\n$R^2$={R2:.3f}")
     axes[0].legend(fontsize=8)
@@ -273,7 +273,7 @@ for cond_name in all_subdirs:
     axes[1].set_ylabel("Epochs to Convergence")
     axes[1].set_yscale("log")
     axes[1].set_title(f"{cond_name}: Convergence vs. $t_{{median}}$")
-    xs, y_fit, slope, intercept, R2 = safe_compute_best_fit(df["t_median"], df["epochs_to_convergence"], log_x=False, log_y=False)
+    xs, y_fit, slope, intercept, R2 = safe_compute_best_fit(df["t_median"], df["epochs_to_convergence"], log_x=False, log_y=True)
     if xs is not None:
         axes[1].plot(xs, y_fit, "k--", label=f"Fit: slope={slope:.3f}, int={intercept:.3f}\n$R^2$={R2:.3f}")
     axes[1].legend(fontsize=8)