Set small learning rate to avoid early overfitting (fix #2100)

Author: Evgueni-Petrov-aka-espetrov

model_analysis/object_importance_tutorial.ipynb

@@ -65,7 +65,7 @@
  "cell_type": "markdown",
  "metadata": {},
  "source": [
- "#### Let's train CatBoost on clear data and take a look at the quality:"
+ "#### Let's train CatBoost on clear data and take a look at the quality. We set a small learning rate to avoid overfitting when we start removing noisy objects."
  ]
  },
  {
@@ -77,12 +77,12 @@
  "name": "stdout",
  "output_type": "stream",
  "text": [
- "0.2157984851490331\n"
+ "0.22947301323494568\n"
  ]
  }
  ],
  "source": [
- "cb = CatBoost({'iterations': 100, 'verbose': False, 'random_seed': 42})\n",
+ "cb = CatBoost({'iterations': 100, 'verbose': False, 'random_seed': 42, 'learning_rate': 0.001})\n",
  "cb.fit(train_pool);\n",
  "print(cb.eval_metrics(validation_pool, ['RMSE'])['RMSE'][-1])"
  ]
@@ -124,7 +124,7 @@
  "name": "stdout",
  "output_type": "stream",
  "text": [
- "0.25915746122622113\n"
+ "0.24770929523786442\n"
  ]
  }
  ],
@@ -168,22 +168,22 @@
  },
  {
  "cell_type": "code",
- "execution_count": 8,
+ "execution_count": 7,
  "metadata": {},
  "outputs": [
  {
  "name": "stdout",
  "output_type": "stream",
  "text": [
- "RMSE on validation datset when 0 harmful objects from train are dropped: 0.25915746122622113\n",
- "RMSE on validation datset when 250 harmful objects from train are dropped: 0.25601149050939825\n",
- "RMSE on validation datset when 500 harmful objects from train are dropped: 0.25158044983631966\n",
- "RMSE on validation datset when 750 harmful objects from train are dropped: 0.24570533776587475\n",
- "RMSE on validation datset when 1000 harmful objects from train are dropped: 0.24171376432589384\n",
- "RMSE on validation datset when 1250 harmful objects from train are dropped: 0.23716221792112202\n",
- "RMSE on validation datset when 1500 harmful objects from train are dropped: 0.23352830055657348\n",
- "RMSE on validation datset when 1750 harmful objects from train are dropped: 0.23035731488436903\n",
- "RMSE on validation datset when 2000 harmful objects from train are dropped: 0.2275943109556251\n"
+ "RMSE on validation datset when 0 harmful objects from train are dropped: 0.24770929523786442\n",
+ "RMSE on validation datset when 250 harmful objects from train are dropped: 0.2447175042288005\n",
+ "RMSE on validation datset when 500 harmful objects from train are dropped: 0.24225895802476696\n",
+ "RMSE on validation datset when 750 harmful objects from train are dropped: 0.23953255257505965\n",
+ "RMSE on validation datset when 1000 harmful objects from train are dropped: 0.23730021406692955\n",
+ "RMSE on validation datset when 1250 harmful objects from train are dropped: 0.23571326583727906\n",
+ "RMSE on validation datset when 1500 harmful objects from train are dropped: 0.23414087500696676\n",
+ "RMSE on validation datset when 1750 harmful objects from train are dropped: 0.23269021852578387\n",
+ "RMSE on validation datset when 2000 harmful objects from train are dropped: 0.231598588484771\n"
  ]
  }
  ],
@@ -210,9 +210,9 @@
  " \n",
  "||RMSE on the validation dataset|\n",
  "|-|-|\n",
- "|Clear train dataset: | 0.215798485149|\n",
- "|Noisy train dataset: | 0.259157461226|\n",
- "|Purified train dataset: | 0.227594310956|"
+ "|Clear train dataset: | 0.22947301323494568|\n",
+ "|Noisy train dataset: | 0.24770929523786442|\n",
+ "|Purified train dataset: | 0.231598588484771|"
  ]
  },
  {
@@ -244,4 +244,4 @@
  },
  "nbformat": 4,
  "nbformat_minor": 1
-}
+}