Polish quickstart adapter validation and propensity caveat

Author: Yurashku

examples/quickstart_own_data_ru.ipynb

@@ -26,10 +26,10 @@
    "id": "a7dc85d4",
    "metadata": {
     "execution": {
-     "iopub.execute_input": "2026-04-15T11:48:46.950173Z",
-     "iopub.status.busy": "2026-04-15T11:48:46.949455Z",
-     "iopub.status.idle": "2026-04-15T11:48:51.464433Z",
-     "shell.execute_reply": "2026-04-15T11:48:51.461888Z"
+     "iopub.execute_input": "2026-04-15T12:51:31.970528Z",
+     "iopub.status.busy": "2026-04-15T12:51:31.970334Z",
+     "iopub.status.idle": "2026-04-15T12:51:33.760421Z",
+     "shell.execute_reply": "2026-04-15T12:51:33.759220Z"
     }
    },
    "outputs": [],
@@ -77,10 +77,10 @@
    "id": "34eb1d72",
    "metadata": {
     "execution": {
-     "iopub.execute_input": "2026-04-15T11:48:51.473375Z",
-     "iopub.status.busy": "2026-04-15T11:48:51.472289Z",
-     "iopub.status.idle": "2026-04-15T11:48:51.528923Z",
-     "shell.execute_reply": "2026-04-15T11:48:51.526616Z"
+     "iopub.execute_input": "2026-04-15T12:51:33.763410Z",
+     "iopub.status.busy": "2026-04-15T12:51:33.763012Z",
+     "iopub.status.idle": "2026-04-15T12:51:33.787338Z",
+     "shell.execute_reply": "2026-04-15T12:51:33.786535Z"
     }
    },
    "outputs": [
@@ -184,6 +184,10 @@
     "df['accept'] = np.random.binomial(1, p)\n",
     "\n",
     "# Если у вас есть logged propensity, храните её в отдельной колонке\n",
+    "# Важно: в реальных данных propensity_A должна логироваться исторической policy A,\n",
+    "# а не придумываться вручную как в этом synthetic-демо.\n",
+    "# Если истинная logged propensity недоступна, используйте propensity_source='estimated'\n",
+    "# (или auto fallback) и интерпретируйте результат осторожнее.\n",
     "df['propensity_A'] = np.random.uniform(0.05, 0.95, size=n)\n",
     "\n",
     "# Шаг 1: формализуем контракт данных\n",
@@ -222,10 +226,10 @@
    "id": "2b792b2f",
    "metadata": {
     "execution": {
-     "iopub.execute_input": "2026-04-15T11:48:51.534425Z",
-     "iopub.status.busy": "2026-04-15T11:48:51.533667Z",
-     "iopub.status.idle": "2026-04-15T11:49:06.388899Z",
-     "shell.execute_reply": "2026-04-15T11:49:06.386732Z"
+     "iopub.execute_input": "2026-04-15T12:51:33.789745Z",
+     "iopub.status.busy": "2026-04-15T12:51:33.789542Z",
+     "iopub.status.idle": "2026-04-15T12:51:38.519607Z",
+     "shell.execute_reply": "2026-04-15T12:51:38.518761Z"
     }
    },
    "outputs": [
@@ -254,11 +258,15 @@
     "\n",
     "    def action_probs(self, x: pd.DataFrame) -> np.ndarray:\n",
     "        a = x[self.action_col].to_numpy()\n",
+    "        unknown = sorted({act for act in np.unique(a) if act not in self._idx})\n",
+    "        if unknown:\n",
+    "            raise ValueError(\n",
+    "                f\"Unknown actions in column '{self.action_col}': {unknown}. \"\n",
+    "                f\"Known action_space: {self.action_space}\"\n",
+    "            )\n",
     "        probs = np.zeros((len(x), len(self.action_space)), dtype=float)\n",
     "        for i, act in enumerate(a):\n",
-    "            j = self._idx.get(act)\n",
-    "            if j is not None:\n",
-    "                probs[i, j] = 1.0\n",
+    "            probs[i, self._idx[act]] = 1.0\n",
     "        return probs\n",
     "\n",
     "# Мини-демо: эмулируем колонку action_B и сравнение через compare_policies\n",
@@ -313,10 +321,10 @@
    "id": "540f2f9b",
    "metadata": {
     "execution": {
-     "iopub.execute_input": "2026-04-15T11:49:06.400372Z",
-     "iopub.status.busy": "2026-04-15T11:49:06.399109Z",
-     "iopub.status.idle": "2026-04-15T11:49:06.412147Z",
-     "shell.execute_reply": "2026-04-15T11:49:06.409880Z"
+     "iopub.execute_input": "2026-04-15T12:51:38.526806Z",
+     "iopub.status.busy": "2026-04-15T12:51:38.526452Z",
+     "iopub.status.idle": "2026-04-15T12:51:38.534354Z",
+     "shell.execute_reply": "2026-04-15T12:51:38.532784Z"
     }
    },
    "outputs": [],
@@ -351,10 +359,10 @@
    "id": "7c240d61",
    "metadata": {
     "execution": {
-     "iopub.execute_input": "2026-04-15T11:49:06.417844Z",
-     "iopub.status.busy": "2026-04-15T11:49:06.417305Z",
-     "iopub.status.idle": "2026-04-15T11:49:36.897842Z",
-     "shell.execute_reply": "2026-04-15T11:49:36.893050Z"
+     "iopub.execute_input": "2026-04-15T12:51:38.536240Z",
+     "iopub.status.busy": "2026-04-15T12:51:38.536072Z",
+     "iopub.status.idle": "2026-04-15T12:51:48.413851Z",
+     "shell.execute_reply": "2026-04-15T12:51:48.411129Z"
     }
    },
    "outputs": [
@@ -421,10 +429,10 @@
    "id": "1fd6d1d4",
    "metadata": {
     "execution": {
-     "iopub.execute_input": "2026-04-15T11:49:36.904965Z",
-     "iopub.status.busy": "2026-04-15T11:49:36.904343Z",
-     "iopub.status.idle": "2026-04-15T11:50:20.869170Z",
-     "shell.execute_reply": "2026-04-15T11:50:20.865727Z"
+     "iopub.execute_input": "2026-04-15T12:51:48.416283Z",
+     "iopub.status.busy": "2026-04-15T12:51:48.415590Z",
+     "iopub.status.idle": "2026-04-15T12:52:02.311901Z",
+     "shell.execute_reply": "2026-04-15T12:52:02.309828Z"
     }
    },
    "outputs": [