Normalize rss using mean weights in gamlss update

simon-hirsch · Jul 17, 2024 · 7c11956 · 7c11956
1 parent c41bff1
commit 7c11956
Show file tree

Hide file tree

Showing 3 changed files with 29 additions and 26 deletions.
diff --git a/.gitignore b/.gitignore
@@ -14,4 +14,5 @@ development
 literature/*
 presentation/*
 .cache
-.devcontainer/*
+.devcontainer/*
+build.txt
diff --git a/example.ipynb b/example.ipynb
@@ -114,16 +114,16 @@
       "LASSO Coefficients \n",
       "\n",
       "[[151.969   3.973  25.   ]\n",
-      " [ -0.     -0.     -0.   ]\n",
-      " [ -9.815  -0.     -0.   ]\n",
-      " [ 24.232   0.044  -0.   ]\n",
-      " [ 13.786   0.      0.   ]\n",
-      " [ -4.886  -0.     -0.   ]\n",
+      " [ -0.     -0.      0.   ]\n",
+      " [ -9.815  -0.      0.   ]\n",
+      " [ 24.232   0.044   0.   ]\n",
+      " [ 13.786   0.     -0.   ]\n",
+      " [ -4.886  -0.      0.   ]\n",
       " [ -0.      0.      0.   ]\n",
-      " [-10.781  -0.026   0.   ]\n",
-      " [  0.      0.      0.   ]\n",
+      " [-10.781  -0.026  -0.   ]\n",
+      " [  0.      0.     -0.   ]\n",
       " [ 24.851   0.      0.   ]\n",
-      " [  2.192   0.      0.   ]]\n"
+      " [  2.192   0.     -0.   ]]\n"
      ]
     }
    ],
@@ -210,17 +210,17 @@
       "\n",
       "Coefficients after update call \n",
       "\n",
-      "[[152.144   3.933   3.475]\n",
-      " [ -0.71   -0.045   0.511]\n",
-      " [-12.521  -0.098   0.618]\n",
-      " [ 24.582   0.036  -0.223]\n",
-      " [ 15.341   0.067  -0.232]\n",
-      " [-33.336  -0.412   2.878]\n",
-      " [ 19.242   0.388  -2.365]\n",
-      " [  2.281  -0.002  -0.51 ]\n",
-      " [  7.311  -0.089   0.267]\n",
-      " [ 35.07    0.19   -1.478]\n",
-      " [  2.397   0.044  -0.235]]\n"
+      "[[152.023   3.917   2.737]\n",
+      " [ -0.     -0.      0.   ]\n",
+      " [-10.774  -0.      0.   ]\n",
+      " [ 24.566   0.035  -0.277]\n",
+      " [ 14.204   0.     -0.   ]\n",
+      " [ -5.537   0.      0.   ]\n",
+      " [ -0.     -0.      0.   ]\n",
+      " [-10.838  -0.      0.   ]\n",
+      " [  0.      0.     -0.   ]\n",
+      " [ 25.509   0.     -0.   ]\n",
+      " [  1.748   0.     -0.143]]\n"
      ]
     }
    ],
@@ -239,9 +239,9 @@
  ],
  "metadata": {
   "kernelspec": {
-   "display_name": "Python (ROLCH Test)",
+   "display_name": "Python 3",
    "language": "python",
-   "name": "myenv"
+   "name": "python3"
   },
   "language_info": {
    "codemirror_mode": {
@@ -253,7 +253,7 @@
    "name": "python",
    "nbconvert_exporter": "python",
    "pygments_lexer": "ipython3",
-   "version": "3.11.9"
+   "version": "3.10.12"
   }
  },
  "nbformat": 4,

diff --git a/src/rolch/online_gamlss.py b/src/rolch/online_gamlss.py
@@ -240,7 +240,7 @@ def update_beta(
             rss = (
                 (residuals**2).flatten() * w
                 + (1 - self.forget) * (self.rss[param] * self.sum_of_weights[param])
-            ) / (self.sum_of_weights[param] * (1 - self.forget) + w)
+            ) / (self.mean_of_weights[param] * (1 - self.forget) + w)
 
         elif (self.method == "lasso") & self.intercept_only[param]:
             lambda_max = None
@@ -264,7 +264,7 @@ def update_beta(
             rss = (
                 (residuals**2).flatten() * w
                 + (1 - self.forget) * (self.rss[param] * self.sum_of_weights[param])
-            ) / (self.sum_of_weights[param] * (1 - self.forget) + w)
+            ) / (self.mean_of_weights[param] * (1 - self.forget) + w)
 
         elif self.method == "lasso":
             intercept = (
@@ -293,7 +293,7 @@ def update_beta(
             rss = (
                 (residuals**2).flatten() * w
                 + (1 - self.forget) * (self.rss[param] * self.sum_of_weights[param])
-            ) / (self.sum_of_weights[param] * (1 - self.forget) + w)
+            ) / (self.mean_of_weights[param] * (1 - self.forget) + w)
 
             model_params_n = np.sum(np.isclose(beta_path, 0), axis=1)
             best_ic = select_best_model_by_information_criterion(
@@ -405,6 +405,7 @@ def fit(
         # distribution parameter for
         # model selection online
         self.sum_of_weights = {}
+        self.mean_of_weights = {}
 
         (
             self.betas,
@@ -779,6 +780,7 @@ def _inner_fit(
 
             ## Sum of weights
             self.sum_of_weights[param] = np.sum(w * wt)
+            self.mean_of_weights[param] = np.mean(w * wt)
 
             self.beta_iterations_inner[param][iteration_outer][iteration_inner] = beta
             self.beta_path_iterations_inner[param][iteration_outer][