Skip to content

Commit

Permalink
FIX typos in imbalance learn
Browse files Browse the repository at this point in the history
  • Loading branch information
@tomMoral
tomMoral committed Nov 13, 2020
1 parent d66713e commit cfeefc0
Showing 1 changed file with 38 additions and 21 deletions.
59 changes: 38 additions & 21 deletions 05_imbalance_learning/01_imbalanced_learning.ipynb
View file
Original file line number Diff line number Diff line change
Expand Up @@ -165,6 +165,7 @@
"outputs": [],
"source": [
"from sklearn.dummy import DummyClassifier\n",
"\n",
"dummy_clf = DummyClassifier(strategy='most_frequent')\n",
"dummy_clf.fit(data, target)"
]
Expand Down Expand Up @@ -308,8 +309,10 @@
" index=dummy_clf.classes_\n",
")\n",
"sns.set(font_scale=1.4) # for label size\n",
"sns.heatmap(cm_df, annot=True, annot_kws={\"size\": 16},\n",
" cmap='Oranges',)\n",
"sns.heatmap(\n",
" cm_df, annot=True, annot_kws={\"size\": 16},\n",
" cmap='Oranges'\n",
")\n",
"\n",
"plt.xlim(0, 2)\n",
"plt.ylim(0, 2)"
Expand Down Expand Up @@ -340,9 +343,11 @@
"from sklearn.metrics import recall_score\n",
"\n",
"precision = precision_score(\n",
" target, y_pred, pos_label=' >50K')\n",
" target, y_pred, pos_label=' >50K'\n",
")\n",
"recall = recall_score(\n",
" target, y_pred, pos_label=' >50K')\n",
" target, y_pred, pos_label=' >50K'\n",
")\n",
"\n",
"print(f\"The recall of the dummy model is \"\n",
" f\"{recall}\")\n",
Expand Down Expand Up @@ -410,7 +415,8 @@
" display(HTML(\n",
" f\"<h5>{name}</h5>\"\n",
" f\"Test Accuracy: {score:7.2%} <br/>\"\n",
" f\"Balanced accuracy: {balanced_score:7.2%} <br/>\"))"
" f\"Balanced accuracy: {balanced_score:7.2%} <br/>\"\n",
" ))"
]
},
{
Expand Down Expand Up @@ -438,11 +444,11 @@
"cell_type": "markdown",
"metadata": {},
"source": [
"### 2.2 Make use of the `class_weight` parameter\n",
"### 2.2 Make use of the `class_weight` parameter to modify the training cost\n",
"\n",
"A first class of methods rely on sample weights to correct the imbalance. The core idea here is to weight prediction mistakes on the minority class higher than mistakes on the most common class.\n",
"\n",
"#### In linear model\n",
"#### In linear model - modifying the cost function\n",
"\n",
"In `scikit-learn`, some estimators have a `class_weight` parameter that permits to do this. The idea is that the ERM is changed such that\n",
"$$\n",
Expand All @@ -459,11 +465,14 @@
"outputs": [],
"source": [
"binary_encoding_columns = ['sex']\n",
"one_hot_encoding_columns = ['workclass', 'education', 'marital-status',\n",
" 'occupation', 'relationship',\n",
" 'race', 'native-country']\n",
"scaling_columns = ['age', 'education-num', 'hours-per-week',\n",
" 'capital-gain', 'capital-loss']"
"one_hot_encoding_columns = [\n",
" 'workclass', 'education', 'marital-status',\n",
" 'occupation', 'relationship', 'race', 'native-country'\n",
"]\n",
"scaling_columns = [\n",
" 'age', 'education-num', 'hours-per-week',\n",
" 'capital-gain', 'capital-loss'\n",
"]"
]
},
{
Expand Down Expand Up @@ -525,7 +534,7 @@
"cell_type": "markdown",
"metadata": {},
"source": [
"#### In tree-based model\n",
"#### In tree-based model - modifying the purity criterion\n",
"\n",
"In tree based models, the `class_weight` option is used to chose on the splits. Indeed, the purity criterion (which is minimize for to chose the split) is computed using these weights. In the leaf, the weights are used to compute the class to output."
]
Expand Down Expand Up @@ -553,7 +562,7 @@
"from sklearn.ensemble import RandomForestClassifier\n",
"\n",
"preprocessor_rf = ColumnTransformer([\n",
" ('binary-encoder', OrdinalEncoder(), ordinal_encoding_columns),\n",
" ('ordinal-encoder', OrdinalEncoder(), ordinal_encoding_columns),\n",
" ('standard-scaler', FunctionTransformer(validate=False), scaling_columns)\n",
"])\n",
"model_rf = make_pipeline(\n",
Expand All @@ -580,7 +589,8 @@
"source": [
"model_rf_balanced = clone(model_rf)\n",
"model_rf_balanced.set_params(\n",
" randomforestclassifier__class_weight='balanced')\n",
" randomforestclassifier__class_weight='balanced'\n",
")\n",
"model_rf_balanced.name = \"Balanced Random Forest\"\n",
"\n",
"evaluate_classifier(model_rf_balanced)"
Expand All @@ -601,7 +611,8 @@
"source": [
"model_rf_subbalanced = clone(model_rf)\n",
"model_rf_subbalanced.set_params(\n",
" randomforestclassifier__class_weight='balanced_subsample')\n",
" randomforestclassifier__class_weight='balanced_subsample'\n",
")\n",
"model_rf_subbalanced.name = \"Balanced Subsample Random Forest\"\n",
"\n",
"evaluate_classifier(model_rf_subbalanced)"
Expand Down Expand Up @@ -768,7 +779,9 @@
"\n",
"model_balanced_rf = make_pipeline(\n",
" preprocessor_rf,\n",
" BalancedRandomForestClassifier(n_estimators=100, random_state=42)\n",
" BalancedRandomForestClassifier(\n",
" n_estimators=100, random_state=42\n",
" )\n",
")\n",
"model_balanced_rf.name = \"Balanced Random Forest\""
]
Expand Down Expand Up @@ -810,8 +823,10 @@
"source": [
"model_bagging = make_pipeline(\n",
" preprocessor_rf,\n",
" BaggingClassifier(base_estimator=HistGradientBoostingClassifier(),\n",
" n_estimators=10, random_state=42)\n",
" BaggingClassifier(\n",
" base_estimator=HistGradientBoostingClassifier(),\n",
" n_estimators=10, random_state=42\n",
" )\n",
")\n",
"model_bagging.name = \"Bagging Model\""
]
Expand Down Expand Up @@ -841,8 +856,10 @@
"from imblearn.ensemble import BalancedBaggingClassifier\n",
"model_balanced_bagging = make_pipeline(\n",
" preprocessor_rf,\n",
" BalancedBaggingClassifier(base_estimator=HistGradientBoostingClassifier(),\n",
" n_estimators=10, random_state=42)\n",
" BalancedBaggingClassifier(\n",
" base_estimator=HistGradientBoostingClassifier(),\n",
" n_estimators=10, random_state=42\n",
" )\n",
")\n",
"model_balanced_bagging.name = \"Balanced Bagging Model\""
]
Expand Down

0 comments on commit cfeefc0

Please sign in to comment.