Pushing the docs to dev/ for branch: master, commit 24890352abc794ca32b9d2e6904e3be5a108f1d8

Author: sklearn-ci

dev/_downloads/auto_examples_jupyter.zip

@@ -15,7 +15,7 @@
       "cell_type": "markdown",
       "metadata": {},
       "source": [
-        "\n# Comparing anomaly detection algorithms for outlier detection on toy datasets\n\n\nThis example shows characteristics of different anomaly detection algorithms\non 2D datasets. Datasets contain one or two modes (regions of high density)\nto illustrate the ability of algorithms to cope with multimodal data.\n\nFor each dataset, 15% of samples are generated as random uniform noise. This\nproportion is the value given to the nu parameter of the OneClassSVM and the\ncontamination parameter of the other outlier detection algorithms.\nDecision boundaries between inliers and outliers are displayed in black.\n\nLocal Outlier Factor (LOF) does not show a decision boundary in black as it\nhas no predict method to be applied on new data.\n\nWhile these examples give some intuition about the algorithms, this\nintuition might not apply to very high dimensional data.\n\nFinally, note that parameters of the models have been here handpicked but\nthat in practice they need to be adjusted. In the absence of labelled data,\nthe problem is completely unsupervised so model selection can be a challenge.\n\n"
+        "\n# Comparing anomaly detection algorithms for outlier detection on toy datasets\n\n\nThis example shows characteristics of different anomaly detection algorithms\non 2D datasets. Datasets contain one or two modes (regions of high density)\nto illustrate the ability of algorithms to cope with multimodal data.\n\nFor each dataset, 15% of samples are generated as random uniform noise. This\nproportion is the value given to the nu parameter of the OneClassSVM and the\ncontamination parameter of the other outlier detection algorithms.\nDecision boundaries between inliers and outliers are displayed in black\nexcept for Local Outlier Factor (LOF) as it has no predict method to be applied\non new data when it is used for outlier detection.\n\nThe :class:`svm.OneClassSVM` is known to be sensitive to outliers and thus does\nnot perform very well for outlier detection. This estimator is best suited for\nnovelty detection when the training set is not contaminated by outliers.\nThat said, outlier detection in high-dimension, or without any assumptions on\nthe distribution of the inlying data is very challenging, and a One-class SVM\nmight give useful results in these situations depending on the value of its\nhyperparameters.\n\n:class:`covariance.EllipticEnvelope` assumes the data is Gaussian and learns\nan ellipse. It thus degrades when the data is not unimodal. Notice however\nthat this estimator is robust to outliers.\n\n:class:`ensemble.IsolationForest` and :class:`neighbors.LocalOutlierFactor`\nseem to perform reasonably well for multi-modal data sets. The advantage of\n:class:`neighbors.LocalOutlierFactor` over the other estimators is shown for\nthe third data set, where the two modes have different densities. This\nadvantage is explained by the local aspect of LOF, meaning that it only\ncompares the score of abnormality of one sample with the scores of its\nneighbors.\n\nFinally, for the last data set, it is hard to say that one sample is more\nabnormal than another sample as they are uniformly distributed in a\nhypercube. Except for the :class:`svm.OneClassSVM` which overfits a little, all\nestimators present decent solutions for this situation. In such a case, it\nwould be wise to look more closely at the scores of abnormality of the samples\nas a good estimator should assign similar scores to all the samples.\n\nWhile these examples give some intuition about the algorithms, this\nintuition might not apply to very high dimensional data.\n\nFinally, note that parameters of the models have been here handpicked but\nthat in practice they need to be adjusted. In the absence of labelled data,\nthe problem is completely unsupervised so model selection can be a challenge.\n\n"
       ]
     },
     {
@@ -26,7 +26,7 @@
       },
       "outputs": [],
       "source": [
-        "# Author: Alexandre Gramfort <[email protected]>\n#         Albert Thomas <[email protected]>\n# License: BSD 3 clause\n\nimport time\n\nimport numpy as np\nimport matplotlib\nimport matplotlib.pyplot as plt\n\nfrom sklearn import svm\nfrom sklearn.datasets import make_moons, make_blobs\nfrom sklearn.covariance import EllipticEnvelope\nfrom sklearn.ensemble import IsolationForest\nfrom sklearn.neighbors import LocalOutlierFactor\n\nprint(__doc__)\n\nmatplotlib.rcParams['contour.negative_linestyle'] = 'solid'\n\n# Example settings\nn_samples = 300\noutliers_fraction = 0.15\nn_outliers = int(outliers_fraction * n_samples)\nn_inliers = n_samples - n_outliers\n\n# define outlier/anomaly detection methods to be compared\nanomaly_algorithms = [\n    (\"Robust covariance\", EllipticEnvelope(contamination=outliers_fraction)),\n    (\"One-Class SVM\", svm.OneClassSVM(nu=outliers_fraction, kernel=\"rbf\",\n                                      gamma=0.1)),\n    (\"Isolation Forest\", IsolationForest(contamination=outliers_fraction,\n                                         random_state=42)),\n    (\"Local Outlier Factor\", LocalOutlierFactor(\n        n_neighbors=35, contamination=outliers_fraction))]\n\n# Define datasets\nblobs_params = dict(random_state=0, n_samples=n_inliers, n_features=2)\ndatasets = [\n    make_blobs(centers=[[0, 0], [0, 0]], cluster_std=0.5,\n               **blobs_params)[0],\n    make_blobs(centers=[[2, 2], [-2, -2]], cluster_std=[1.5, .3],\n               **blobs_params)[0],\n    4. * (make_moons(n_samples=n_samples, noise=.05, random_state=0)[0] -\n          np.array([0.5, 0.25])),\n    14. * (np.random.RandomState(42).rand(n_samples, 2) - 0.5)]\n\n# Compare given classifiers under given settings\nxx, yy = np.meshgrid(np.linspace(-7, 7, 150),\n                     np.linspace(-7, 7, 150))\n\nplt.figure(figsize=(len(anomaly_algorithms) * 2 + 3, 12.5))\nplt.subplots_adjust(left=.02, right=.98, bottom=.001, top=.96, wspace=.05,\n                    hspace=.01)\n\nplot_num = 1\nrng = np.random.RandomState(42)\n\nfor i_dataset, X in enumerate(datasets):\n    # Add outliers\n    X = np.concatenate([X, rng.uniform(low=-6, high=6,\n                       size=(n_outliers, 2))], axis=0)\n\n    for name, algorithm in anomaly_algorithms:\n        t0 = time.time()\n        algorithm.fit(X)\n        t1 = time.time()\n        plt.subplot(len(datasets), len(anomaly_algorithms), plot_num)\n        if i_dataset == 0:\n            plt.title(name, size=18)\n\n        # fit the data and tag outliers\n        if name == \"Local Outlier Factor\":\n            y_pred = algorithm.fit_predict(X)\n        else:\n            y_pred = algorithm.fit(X).predict(X)\n\n        # plot the levels lines and the points\n        if name != \"Local Outlier Factor\":  # LOF does not implement predict\n            Z = algorithm.predict(np.c_[xx.ravel(), yy.ravel()])\n            Z = Z.reshape(xx.shape)\n            plt.contour(xx, yy, Z, levels=[0], linewidths=2, colors='black')\n\n        colors = np.array(['#377eb8', '#ff7f00'])\n        plt.scatter(X[:, 0], X[:, 1], s=10, color=colors[(y_pred + 1) // 2])\n\n        plt.xlim(-7, 7)\n        plt.ylim(-7, 7)\n        plt.xticks(())\n        plt.yticks(())\n        plt.text(.99, .01, ('%.2fs' % (t1 - t0)).lstrip('0'),\n                 transform=plt.gca().transAxes, size=15,\n                 horizontalalignment='right')\n        plot_num += 1\n\nplt.show()"
+        "# Author: Alexandre Gramfort <[email protected]>\n#         Albert Thomas <[email protected]>\n# License: BSD 3 clause\n\nimport time\n\nimport numpy as np\nimport matplotlib\nimport matplotlib.pyplot as plt\n\nfrom sklearn import svm\nfrom sklearn.datasets import make_moons, make_blobs\nfrom sklearn.covariance import EllipticEnvelope\nfrom sklearn.ensemble import IsolationForest\nfrom sklearn.neighbors import LocalOutlierFactor\n\nprint(__doc__)\n\nmatplotlib.rcParams['contour.negative_linestyle'] = 'solid'\n\n# Example settings\nn_samples = 300\noutliers_fraction = 0.15\nn_outliers = int(outliers_fraction * n_samples)\nn_inliers = n_samples - n_outliers\n\n# define outlier/anomaly detection methods to be compared\nanomaly_algorithms = [\n    (\"Robust covariance\", EllipticEnvelope(contamination=outliers_fraction)),\n    (\"One-Class SVM\", svm.OneClassSVM(nu=outliers_fraction, kernel=\"rbf\",\n                                      gamma=0.1)),\n    (\"Isolation Forest\", IsolationForest(contamination=outliers_fraction,\n                                         random_state=42)),\n    (\"Local Outlier Factor\", LocalOutlierFactor(\n        n_neighbors=35, contamination=outliers_fraction))]\n\n# Define datasets\nblobs_params = dict(random_state=0, n_samples=n_inliers, n_features=2)\ndatasets = [\n    make_blobs(centers=[[0, 0], [0, 0]], cluster_std=0.5,\n               **blobs_params)[0],\n    make_blobs(centers=[[2, 2], [-2, -2]], cluster_std=[0.5, 0.5],\n               **blobs_params)[0],\n    make_blobs(centers=[[2, 2], [-2, -2]], cluster_std=[1.5, .3],\n               **blobs_params)[0],\n    4. * (make_moons(n_samples=n_samples, noise=.05, random_state=0)[0] -\n          np.array([0.5, 0.25])),\n    14. * (np.random.RandomState(42).rand(n_samples, 2) - 0.5)]\n\n# Compare given classifiers under given settings\nxx, yy = np.meshgrid(np.linspace(-7, 7, 150),\n                     np.linspace(-7, 7, 150))\n\nplt.figure(figsize=(len(anomaly_algorithms) * 2 + 3, 12.5))\nplt.subplots_adjust(left=.02, right=.98, bottom=.001, top=.96, wspace=.05,\n                    hspace=.01)\n\nplot_num = 1\nrng = np.random.RandomState(42)\n\nfor i_dataset, X in enumerate(datasets):\n    # Add outliers\n    X = np.concatenate([X, rng.uniform(low=-6, high=6,\n                       size=(n_outliers, 2))], axis=0)\n\n    for name, algorithm in anomaly_algorithms:\n        t0 = time.time()\n        algorithm.fit(X)\n        t1 = time.time()\n        plt.subplot(len(datasets), len(anomaly_algorithms), plot_num)\n        if i_dataset == 0:\n            plt.title(name, size=18)\n\n        # fit the data and tag outliers\n        if name == \"Local Outlier Factor\":\n            y_pred = algorithm.fit_predict(X)\n        else:\n            y_pred = algorithm.fit(X).predict(X)\n\n        # plot the levels lines and the points\n        if name != \"Local Outlier Factor\":  # LOF does not implement predict\n            Z = algorithm.predict(np.c_[xx.ravel(), yy.ravel()])\n            Z = Z.reshape(xx.shape)\n            plt.contour(xx, yy, Z, levels=[0], linewidths=2, colors='black')\n\n        colors = np.array(['#377eb8', '#ff7f00'])\n        plt.scatter(X[:, 0], X[:, 1], s=10, color=colors[(y_pred + 1) // 2])\n\n        plt.xlim(-7, 7)\n        plt.ylim(-7, 7)\n        plt.xticks(())\n        plt.yticks(())\n        plt.text(.99, .01, ('%.2fs' % (t1 - t0)).lstrip('0'),\n                 transform=plt.gca().transAxes, size=15,\n                 horizontalalignment='right')\n        plot_num += 1\n\nplt.show()"
       ]
     }
   ],

dev/_downloads/auto_examples_python.zip

@@ -10,10 +10,36 @@
 For each dataset, 15% of samples are generated as random uniform noise. This
 proportion is the value given to the nu parameter of the OneClassSVM and the
 contamination parameter of the other outlier detection algorithms.
-Decision boundaries between inliers and outliers are displayed in black.
-
-Local Outlier Factor (LOF) does not show a decision boundary in black as it
-has no predict method to be applied on new data.
+Decision boundaries between inliers and outliers are displayed in black
+except for Local Outlier Factor (LOF) as it has no predict method to be applied
+on new data when it is used for outlier detection.
+
+The :class:`svm.OneClassSVM` is known to be sensitive to outliers and thus does
+not perform very well for outlier detection. This estimator is best suited for
+novelty detection when the training set is not contaminated by outliers.
+That said, outlier detection in high-dimension, or without any assumptions on
+the distribution of the inlying data is very challenging, and a One-class SVM
+might give useful results in these situations depending on the value of its
+hyperparameters.
+
+:class:`covariance.EllipticEnvelope` assumes the data is Gaussian and learns
+an ellipse. It thus degrades when the data is not unimodal. Notice however
+that this estimator is robust to outliers.
+
+:class:`ensemble.IsolationForest` and :class:`neighbors.LocalOutlierFactor`
+seem to perform reasonably well for multi-modal data sets. The advantage of
+:class:`neighbors.LocalOutlierFactor` over the other estimators is shown for
+the third data set, where the two modes have different densities. This
+advantage is explained by the local aspect of LOF, meaning that it only
+compares the score of abnormality of one sample with the scores of its
+neighbors.
+
+Finally, for the last data set, it is hard to say that one sample is more
+abnormal than another sample as they are uniformly distributed in a
+hypercube. Except for the :class:`svm.OneClassSVM` which overfits a little, all
+estimators present decent solutions for this situation. In such a case, it
+would be wise to look more closely at the scores of abnormality of the samples
+as a good estimator should assign similar scores to all the samples.
 
 While these examples give some intuition about the algorithms, this
 intuition might not apply to very high dimensional data.
@@ -64,6 +90,8 @@
 datasets = [
     make_blobs(centers=[[0, 0], [0, 0]], cluster_std=0.5,
                **blobs_params)[0],
+    make_blobs(centers=[[2, 2], [-2, -2]], cluster_std=[0.5, 0.5],
+               **blobs_params)[0],
     make_blobs(centers=[[2, 2], [-2, -2]], cluster_std=[1.5, .3],
                **blobs_params)[0],
     4. * (make_moons(n_samples=n_samples, noise=.05, random_state=0)[0] -