Pushing the docs to dev/ for branch: master, commit 7b346e784b1391815bd17f64fd239d3fbd78dd5a

Author: sklearn-ci

dev/_downloads/auto_examples_jupyter.zip

@@ -15,7 +15,7 @@
       "cell_type": "markdown",
       "metadata": {},
       "source": [
-        "\n# Balance model complexity and cross-validated score\n\n\nThis example balances model complexity and cross-validated score by\nfinding a decent accuracy within 1 standard deviation of the best accuracy\nscore while minimising the number of PCA components [1].\n\nThe figure shows the trade-off between cross-validated score and the number\nof PCA components. The balanced case is when n_components=6 and accuracy=0.80,\nwhich falls into the range within 1 standard deviation of the best accuracy\nscore.\n\n[1] Hastie, T., Tibshirani, R.,, Friedman, J. (2001). Model Assessment and\nSelection. The Elements of Statistical Learning (pp. 219-260). New York,\nNY, USA: Springer New York Inc..\n\n"
+        "\n# Balance model complexity and cross-validated score\n\n\nThis example balances model complexity and cross-validated score by\nfinding a decent accuracy within 1 standard deviation of the best accuracy\nscore while minimising the number of PCA components [1].\n\nThe figure shows the trade-off between cross-validated score and the number\nof PCA components. The balanced case is when n_components=10 and accuracy=0.88,\nwhich falls into the range within 1 standard deviation of the best accuracy\nscore.\n\n[1] Hastie, T., Tibshirani, R.,, Friedman, J. (2001). Model Assessment and\nSelection. The Elements of Statistical Learning (pp. 219-260). New York,\nNY, USA: Springer New York Inc..\n\n"
       ]
     },
     {
@@ -26,7 +26,7 @@
       },
       "outputs": [],
       "source": [
-        "# Author: Wenhao Zhang <[email protected]>\n\nprint(__doc__)\n\nimport numpy as np\nimport matplotlib.pyplot as plt\n\nfrom sklearn.datasets import load_digits\nfrom sklearn.decomposition import PCA\nfrom sklearn.model_selection import GridSearchCV\nfrom sklearn.pipeline import Pipeline\nfrom sklearn.svm import LinearSVC\n\n\ndef lower_bound(cv_results):\n    \"\"\"\n    Calculate the lower bound within 1 standard deviation\n    of the best `mean_test_scores`.\n\n    Parameters\n    ----------\n    cv_results : dict of numpy(masked) ndarrays\n        See attribute cv_results_ of `GridSearchCV`\n\n    Returns\n    -------\n    float\n        Lower bound within 1 standard deviation of the\n        best `mean_test_score`.\n    \"\"\"\n    best_score_idx = np.argmax(cv_results['mean_test_score'])\n\n    return (cv_results['mean_test_score'][best_score_idx]\n            - cv_results['std_test_score'][best_score_idx])\n\n\ndef best_low_complexity(cv_results):\n    \"\"\"\n    Balance model complexity with cross-validated score.\n\n    Parameters\n    ----------\n    cv_results : dict of numpy(masked) ndarrays\n        See attribute cv_results_ of `GridSearchCV`.\n\n    Return\n    ------\n    int\n        Index of a model that has the fewest PCA components\n        while has its test score within 1 standard deviation of the best\n        `mean_test_score`.\n    \"\"\"\n    threshold = lower_bound(cv_results)\n    candidate_idx = np.flatnonzero(cv_results['mean_test_score'] >= threshold)\n    best_idx = candidate_idx[cv_results['param_reduce_dim__n_components']\n                             [candidate_idx].argmin()]\n    return best_idx\n\n\npipe = Pipeline([\n        ('reduce_dim', PCA(random_state=42)),\n        ('classify', LinearSVC(random_state=42)),\n])\n\nparam_grid = {\n    'reduce_dim__n_components': [2, 4, 6, 8]\n}\n\ngrid = GridSearchCV(pipe, cv=10, n_jobs=1, param_grid=param_grid,\n                    scoring='accuracy', refit=best_low_complexity)\nX, y = load_digits(return_X_y=True)\ngrid.fit(X, y)\n\nn_components = grid.cv_results_['param_reduce_dim__n_components']\ntest_scores = grid.cv_results_['mean_test_score']\n\nplt.figure()\nplt.bar(n_components, test_scores, width=1.3, color='b')\n\nlower = lower_bound(grid.cv_results_)\nplt.axhline(np.max(test_scores), linestyle='--', color='y',\n            label='Best score')\nplt.axhline(lower, linestyle='--', color='.5', label='Best score - 1 std')\n\nplt.title(\"Balance model complexity and cross-validated score\")\nplt.xlabel('Number of PCA components used')\nplt.ylabel('Digit classification accuracy')\nplt.xticks(n_components.tolist())\nplt.ylim((0, 1.0))\nplt.legend(loc='upper left')\n\nbest_index_ = grid.best_index_\n\nprint(\"The best_index_ is %d\" % best_index_)\nprint(\"The n_components selected is %d\" % n_components[best_index_])\nprint(\"The corresponding accuracy score is %.2f\"\n      % grid.cv_results_['mean_test_score'][best_index_])\nplt.show()"
+        "# Author: Wenhao Zhang <[email protected]>\n\nprint(__doc__)\n\nimport numpy as np\nimport matplotlib.pyplot as plt\n\nfrom sklearn.datasets import load_digits\nfrom sklearn.decomposition import PCA\nfrom sklearn.model_selection import GridSearchCV\nfrom sklearn.pipeline import Pipeline\nfrom sklearn.svm import LinearSVC\n\n\ndef lower_bound(cv_results):\n    \"\"\"\n    Calculate the lower bound within 1 standard deviation\n    of the best `mean_test_scores`.\n\n    Parameters\n    ----------\n    cv_results : dict of numpy(masked) ndarrays\n        See attribute cv_results_ of `GridSearchCV`\n\n    Returns\n    -------\n    float\n        Lower bound within 1 standard deviation of the\n        best `mean_test_score`.\n    \"\"\"\n    best_score_idx = np.argmax(cv_results['mean_test_score'])\n\n    return (cv_results['mean_test_score'][best_score_idx]\n            - cv_results['std_test_score'][best_score_idx])\n\n\ndef best_low_complexity(cv_results):\n    \"\"\"\n    Balance model complexity with cross-validated score.\n\n    Parameters\n    ----------\n    cv_results : dict of numpy(masked) ndarrays\n        See attribute cv_results_ of `GridSearchCV`.\n\n    Return\n    ------\n    int\n        Index of a model that has the fewest PCA components\n        while has its test score within 1 standard deviation of the best\n        `mean_test_score`.\n    \"\"\"\n    threshold = lower_bound(cv_results)\n    candidate_idx = np.flatnonzero(cv_results['mean_test_score'] >= threshold)\n    best_idx = candidate_idx[cv_results['param_reduce_dim__n_components']\n                             [candidate_idx].argmin()]\n    return best_idx\n\n\npipe = Pipeline([\n        ('reduce_dim', PCA(random_state=42)),\n        ('classify', LinearSVC(random_state=42, C=0.01)),\n])\n\nparam_grid = {\n    'reduce_dim__n_components': [6, 8, 10, 12, 14]\n}\n\ngrid = GridSearchCV(pipe, cv=10, n_jobs=1, param_grid=param_grid,\n                    scoring='accuracy', refit=best_low_complexity)\nX, y = load_digits(return_X_y=True)\ngrid.fit(X, y)\n\nn_components = grid.cv_results_['param_reduce_dim__n_components']\ntest_scores = grid.cv_results_['mean_test_score']\n\nplt.figure()\nplt.bar(n_components, test_scores, width=1.3, color='b')\n\nlower = lower_bound(grid.cv_results_)\nplt.axhline(np.max(test_scores), linestyle='--', color='y',\n            label='Best score')\nplt.axhline(lower, linestyle='--', color='.5', label='Best score - 1 std')\n\nplt.title(\"Balance model complexity and cross-validated score\")\nplt.xlabel('Number of PCA components used')\nplt.ylabel('Digit classification accuracy')\nplt.xticks(n_components.tolist())\nplt.ylim((0, 1.0))\nplt.legend(loc='upper left')\n\nbest_index_ = grid.best_index_\n\nprint(\"The best_index_ is %d\" % best_index_)\nprint(\"The n_components selected is %d\" % n_components[best_index_])\nprint(\"The corresponding accuracy score is %.2f\"\n      % grid.cv_results_['mean_test_score'][best_index_])\nplt.show()"
       ]
     }
   ],

dev/_downloads/auto_examples_python.zip

@@ -8,7 +8,7 @@
 score while minimising the number of PCA components [1].
 
 The figure shows the trade-off between cross-validated score and the number
-of PCA components. The balanced case is when n_components=6 and accuracy=0.80,
+of PCA components. The balanced case is when n_components=10 and accuracy=0.88,
 which falls into the range within 1 standard deviation of the best accuracy
 score.
 
@@ -77,11 +77,11 @@ def best_low_complexity(cv_results):
 
 pipe = Pipeline([
         ('reduce_dim', PCA(random_state=42)),
-        ('classify', LinearSVC(random_state=42)),
+        ('classify', LinearSVC(random_state=42, C=0.01)),
 ])
 
 param_grid = {
-    'reduce_dim__n_components': [2, 4, 6, 8]
+    'reduce_dim__n_components': [6, 8, 10, 12, 14]
 }
 
 grid = GridSearchCV(pipe, cv=10, n_jobs=1, param_grid=param_grid,