Pushing the docs to dev/ for branch: main, commit 3d16a21b6c9c6940509d98e5e0c030658f7c348c

Author: sklearn-ci

dev/_downloads/07fcc19ba03226cd3d83d4e40ec44385/auto_examples_python.zip

@@ -26,7 +26,86 @@
       },
       "outputs": [],
       "source": [
-        "from sklearn.ensemble import HistGradientBoostingRegressor\nfrom sklearn.inspection import PartialDependenceDisplay\nimport numpy as np\nimport matplotlib.pyplot as plt\n\n\nrng = np.random.RandomState(0)\n\nn_samples = 5000\nf_0 = rng.rand(n_samples)\nf_1 = rng.rand(n_samples)\nX = np.c_[f_0, f_1]\nnoise = rng.normal(loc=0.0, scale=0.01, size=n_samples)\n\n# y is positively correlated with f_0, and negatively correlated with f_1\ny = 5 * f_0 + np.sin(10 * np.pi * f_0) - 5 * f_1 - np.cos(10 * np.pi * f_1) + noise\n\nfig, ax = plt.subplots()\n\n\n# Without any constraint\ngbdt = HistGradientBoostingRegressor()\ngbdt.fit(X, y)\ndisp = PartialDependenceDisplay.from_estimator(\n    gbdt,\n    X,\n    features=[0, 1],\n    feature_names=(\n        \"First feature\",\n        \"Second feature\",\n    ),\n    line_kw={\"linewidth\": 4, \"label\": \"unconstrained\", \"color\": \"tab:blue\"},\n    ax=ax,\n)\n\n# With monotonic increase (1) and a monotonic decrease (-1) constraints, respectively.\ngbdt = HistGradientBoostingRegressor(monotonic_cst=[1, -1])\ngbdt.fit(X, y)\n\nPartialDependenceDisplay.from_estimator(\n    gbdt,\n    X,\n    features=[0, 1],\n    line_kw={\"linewidth\": 4, \"label\": \"constrained\", \"color\": \"tab:orange\"},\n    ax=disp.axes_,\n)\n\nfor f_idx in (0, 1):\n    disp.axes_[0, f_idx].plot(\n        X[:, f_idx], y, \"o\", alpha=0.3, zorder=-1, color=\"tab:green\"\n    )\n    disp.axes_[0, f_idx].set_ylim(-6, 6)\n\nplt.legend()\nfig.suptitle(\"Monotonic constraints effect on partial dependences\")\n\nplt.show()"
+        "from sklearn.ensemble import HistGradientBoostingRegressor\nfrom sklearn.inspection import PartialDependenceDisplay\nimport numpy as np\nimport matplotlib.pyplot as plt\n\n\nrng = np.random.RandomState(0)\n\nn_samples = 1000\nf_0 = rng.rand(n_samples)\nf_1 = rng.rand(n_samples)\nX = np.c_[f_0, f_1]\nnoise = rng.normal(loc=0.0, scale=0.01, size=n_samples)\n\n# y is positively correlated with f_0, and negatively correlated with f_1\ny = 5 * f_0 + np.sin(10 * np.pi * f_0) - 5 * f_1 - np.cos(10 * np.pi * f_1) + noise"
+      ]
+    },
+    {
+      "cell_type": "markdown",
+      "metadata": {},
+      "source": [
+        "Fit a first model on this dataset without any constraints.\n\n"
+      ]
+    },
+    {
+      "cell_type": "code",
+      "execution_count": null,
+      "metadata": {
+        "collapsed": false
+      },
+      "outputs": [],
+      "source": [
+        "gbdt_no_cst = HistGradientBoostingRegressor()\ngbdt_no_cst.fit(X, y)"
+      ]
+    },
+    {
+      "cell_type": "markdown",
+      "metadata": {},
+      "source": [
+        "Fit a second model on this dataset with monotonic increase (1)\nand a monotonic decrease (-1) constraints, respectively.\n\n"
+      ]
+    },
+    {
+      "cell_type": "code",
+      "execution_count": null,
+      "metadata": {
+        "collapsed": false
+      },
+      "outputs": [],
+      "source": [
+        "gbdt_with_monotonic_cst = HistGradientBoostingRegressor(monotonic_cst=[1, -1])\ngbdt_with_monotonic_cst.fit(X, y)"
+      ]
+    },
+    {
+      "cell_type": "markdown",
+      "metadata": {},
+      "source": [
+        "Let's display the partial dependence of the predictions on the two features.\n\n"
+      ]
+    },
+    {
+      "cell_type": "code",
+      "execution_count": null,
+      "metadata": {
+        "collapsed": false
+      },
+      "outputs": [],
+      "source": [
+        "fig, ax = plt.subplots()\ndisp = PartialDependenceDisplay.from_estimator(\n    gbdt_no_cst,\n    X,\n    features=[0, 1],\n    feature_names=(\n        \"First feature\",\n        \"Second feature\",\n    ),\n    line_kw={\"linewidth\": 4, \"label\": \"unconstrained\", \"color\": \"tab:blue\"},\n    ax=ax,\n)\nPartialDependenceDisplay.from_estimator(\n    gbdt_with_monotonic_cst,\n    X,\n    features=[0, 1],\n    line_kw={\"linewidth\": 4, \"label\": \"constrained\", \"color\": \"tab:orange\"},\n    ax=disp.axes_,\n)\n\nfor f_idx in (0, 1):\n    disp.axes_[0, f_idx].plot(\n        X[:, f_idx], y, \"o\", alpha=0.3, zorder=-1, color=\"tab:green\"\n    )\n    disp.axes_[0, f_idx].set_ylim(-6, 6)\n\nplt.legend()\nfig.suptitle(\"Monotonic constraints effect on partial dependences\")\nplt.show()"
+      ]
+    },
+    {
+      "cell_type": "markdown",
+      "metadata": {},
+      "source": [
+        "We can see that the predictions of the unconstrained model capture the\noscillations of the data while the constrained model follows the general\ntrend and ignores the local variations.\n\n"
+      ]
+    },
+    {
+      "cell_type": "markdown",
+      "metadata": {},
+      "source": [
+        "\n## Using feature names to specify monotonic constraints\n\nNote that if the training data has feature names, it's possible to specifiy the\nmonotonic constraints by passing a dictionary:\n\n"
+      ]
+    },
+    {
+      "cell_type": "code",
+      "execution_count": null,
+      "metadata": {
+        "collapsed": false
+      },
+      "outputs": [],
+      "source": [
+        "import pandas as pd\n\nX_df = pd.DataFrame(X, columns=[\"f_0\", \"f_1\"])\n\ngbdt_with_monotonic_cst_df = HistGradientBoostingRegressor(\n    monotonic_cst={\"f_0\": 1, \"f_1\": -1}\n).fit(X_df, y)\n\nnp.allclose(\n    gbdt_with_monotonic_cst_df.predict(X_df), gbdt_with_monotonic_cst.predict(X)\n)"
       ]
     }
   ],