First one-shot pass

src/engine/learning_rate.rs

@@ -0,0 +1,59 @@
+#[allow(dead_code)]
+pub fn polynomial_regression(times: &[f64]) -> Option<f64> {
+    if times.len() < 3 {
+        return None;
+    }
+
+    let n = times.len();
+    let xs: Vec<f64> = (0..n).map(|i| i as f64).collect();
+
+    let x_mean: f64 = xs.iter().sum::<f64>() / n as f64;
+    let y_mean: f64 = times.iter().sum::<f64>() / n as f64;
+
+    let mut ss_xy = 0.0;
+    let mut ss_xx = 0.0;
+    let mut ss_yy = 0.0;
+
+    for i in 0..n {
+        let dx = xs[i] - x_mean;
+        let dy = times[i] - y_mean;
+        ss_xy += dx * dy;
+        ss_xx += dx * dx;
+        ss_yy += dy * dy;
+    }
+
+    if ss_xx < 1e-10 || ss_yy < 1e-10 {
+        return None;
+    }
+
+    let slope = ss_xy / ss_xx;
+    let r_squared = (ss_xy * ss_xy) / (ss_xx * ss_yy);
+
+    if r_squared < 0.5 {
+        return None;
+    }
+
+    let predicted_next = y_mean + slope * (n as f64 - x_mean);
+    Some(predicted_next.max(0.0))
+}
+
+#[allow(dead_code)]
+pub fn learning_rate_description(times: &[f64]) -> &'static str {
+    match polynomial_regression(times) {
+        Some(predicted) => {
+            if times.is_empty() {
+                return "No data";
+            }
+            let current = times.last().unwrap();
+            let improvement = (current - predicted) / current * 100.0;
+            if improvement > 5.0 {
+                "Improving"
+            } else if improvement < -5.0 {
+                "Slowing down"
+            } else {
+                "Steady"
+            }
+        }
+        None => "Not enough data",
+    }
+}