updated notebooks for current packages as of 2024-02

06_EvaluatingClassifiers_code.ipynb

@@ -64,7 +64,7 @@
    "source": [
     "# helpful stdlib tool for cleaning up printouts\n",
     "import textwrap\n",
-    "print(textwrap.fill(str(sorted(metrics.SCORERS.keys())), \n",
+    "print(textwrap.fill(str(sorted(metrics.get_scorer_names())), \n",
     "                    width=70))"
    ]
   },
@@ -278,7 +278,7 @@
    "outputs": [],
    "source": [
     "print(\"'Multi-label' Encoding\")\n",
-    "print(skpre.label_binarize(iris.target, [0,1,2])[checkout])"
+    "print(skpre.label_binarize(iris.target, classes=[0,1,2])[checkout])"
    ]
   },
   {
@@ -287,7 +287,7 @@
    "metadata": {},
    "outputs": [],
    "source": [
-    "iris_multi_tgt = skpre.label_binarize(iris.target, [0,1,2])\n",
+    "iris_multi_tgt = skpre.label_binarize(iris.target, classes=[0,1,2])\n",
     "\n",
     "# im --> \"iris multi\"\n",
     "(im_train_ftrs, im_test_ftrs, \n",
@@ -358,14 +358,14 @@
     "    classes   = np.unique(test_tgt)\n",
     "    n_classes = len(classes)\n",
     "\n",
-    "    indicator = skpre.label_binarize(test_tgt, classes)\n",
+    "    indicator = skpre.label_binarize(test_tgt, classes=classes)\n",
     "    avg_auc_sum = 0.0\n",
     "\n",
     "    # comparing class i and class j\n",
     "    for ij in it.combinations(classes, 2):\n",
     "        # use use sum to act like a logical or\n",
     "        ij_indicator = indicator[:,ij].sum(axis=1, \n",
-    "                                           dtype=np.bool)\n",
+    "                                           dtype=np.bool_)\n",
     "        \n",
     "        # slightly ugly, can't broadcast these as indexes\n",
     "        # use .ix_ to save the day\n",
@@ -712,7 +712,7 @@
    "outputs": [],
    "source": [
     "macro_precision = metrics.make_scorer(metrics.precision_score,\n",
-    "                                      average='macro')\n",
+    "                                      average='macro', zero_division=0)\n",
     "macro_recall    = metrics.make_scorer(metrics.recall_score,\n",
     "                                      average='macro')\n",
     "htm_scorer = metrics.make_scorer(hand_and_till_M_statistic, \n",
@@ -802,9 +802,16 @@
     "    # save as\n",
     "    df.to_csv('portugese_student_numeric_discrete.csv', index=False)"
    ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": []
   }
  ],
  "metadata": {},
  "nbformat": 4,
- "nbformat_minor": 2
+ "nbformat_minor": 4
 }

08_MoreClassificationMethods_code.ipynb

@@ -271,7 +271,7 @@
     "ftrs, tgt = datasets.make_blobs(centers=2,\n",
     "                                n_features=3,\n",
     "                                n_samples=200,\n",
-    "                                center_box = [-2.0, 2.0],\n",
+    "                                center_box = (-2.0, 2.0),\n",
     "                                random_state=1099)\n",
     "\n",
     "# note, using three features, but graphing only two dimensions\n",
@@ -522,7 +522,7 @@
     "logreg_classifiers = {'LogReg(saga)': LogReg(solver='saga', \n",
     "                                             multi_class='multinomial',\n",
     "                                             max_iter=1000),\n",
-    "                      'LogReg(SGD)' :  SGD(loss='log', max_iter=1000)}\n",
+    "                      'LogReg(SGD)' :  SGD(loss='log_loss', max_iter=1000)}\n",
     "\n",
     "fig, axes = plt.subplots(1,2,figsize=(12,4))\n",
     "axes = axes.flat\n",
@@ -558,9 +558,15 @@
     "tgt = (np.array(y) == 'red')\n",
     "\n",
     "# sm.Logit is statsmodels name for logistic regression\n",
-    "(sm.Logit(tgt, x, method='newton')\n",
-    "   .fit()\n",
-    "   .predict(x))  # training predictions"
+    "# Logit is very, very unhappy trying to deal with a perfectly \n",
+    "#       separable example.  so, there are many weird arguments.\n",
+    "#       and it still seems to fail.\n",
+    "# FIXME.  I want just a simple example that recovers the target\n",
+    "#         function.\n",
+    "#(sm.Logit(tgt, x)\n",
+    "#   .fit(method='newton', skip_hessian=True, \n",
+    "#        full_output=False, warn_convergence=False)\n",
+    "#   .predict(x))  # training predictions"
    ]
   },
   {
@@ -944,9 +950,9 @@
    "outputs": [],
    "source": [
     "classifier_parade = \\\n",
-    "    {'LogReg(1)' : linear_model.LogisticRegression(max_iter=1000),\n",
-    "     'LogReg(2)' : linear_model.SGDClassifier(loss='log',\n",
-    "                                              max_iter=1000),\n",
+    "    {'LogReg(1)' : linear_model.LogisticRegression(max_iter=10000),\n",
+    "     'LogReg(2)' : linear_model.SGDClassifier(loss='log_loss',\n",
+    "                                              max_iter=10000),\n",
     "\n",
     "     'QDA' : discriminant_analysis.QuadraticDiscriminantAnalysis(),\n",
     "     'LDA' : discriminant_analysis.LinearDiscriminantAnalysis(),\n",
@@ -986,6 +992,13 @@
     "ax.set_ylabel('Accuracy')\n",
     "ax.legend(loc='lower center', ncol=2);"
    ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": []
   }
  ],
  "metadata": {},

10_Manual_Feature_Engineering_code.ipynb

@@ -101,7 +101,11 @@
    "outputs": [],
    "source": [
     "plt.subplots(1,1,figsize=(4,3))\n",
-    "ax = sns.distplot(iris_df['sepal length'], hist=False, rug=True)\n",
+    "\n",
+    "sns.rugplot(iris_df['sepal length'])\n",
+    "ax = sns.kdeplot(iris_df['sepal length'])\n",
+    "\n",
+    "ax.set_ylim(bottom=0.0)\n",
     "ax.set_ylabel(\"Approximate %\");"
    ]
   },
@@ -113,10 +117,10 @@
    "source": [
     "# apply binary threshold to numeric with sklearn is tricky\n",
     "column = iris_df[['sepal length']] # keep 2Dness b/c sk complains\n",
-    "col_mean = column.mean().values    # and sk fails with Series/DF\n",
+    "col_mean = column.mean().values[0]    # and sk fails with Series/DF\n",
     "\n",
     "both = column.copy()\n",
-    "both['> Mean'] = skpre.binarize(column, col_mean).astype(np.bool)\n",
+    "both['> Mean'] = skpre.binarize(column, threshold=col_mean).astype(np.bool_)\n",
     "\n",
     "print('Column Mean:', col_mean)\n",
     "display(both.iloc[[0,50,100]])"
@@ -164,7 +168,7 @@
     "                                      new_iris_df.columns], \n",
     "                                     [[1, 0, 0], [0,1,2]])\n",
     "\n",
-    "new_iris_df.sort_index(axis='columns', inplace=True)\n",
+    "new_iris_df = new_iris_df.sort_index(axis='columns')\n",
     "display(new_iris_df.iloc[[0,50,100]])"
    ]
   },
@@ -468,8 +472,8 @@
     "subset = iris_df.loc[[0, 50, 100], ['sepal length', 'sepal width']]\n",
     "new_terms = pd.DataFrame(quad_inters.fit_transform(subset), \n",
     "                         index=[0, 50, 100])\n",
-    "new_terms.set_axis(['sep length', 'sep width', 'sep area'], \n",
-    "                   axis=1, inplace=True)\n",
+    "new_terms = new_terms.set_axis(['sep length', 'sep width', 'sep area'], \n",
+    "                               axis=1)\n",
     "\n",
     "# note:  creating the interaction *also* \n",
     "# includes the base terms in the interaction\n",
@@ -695,7 +699,11 @@
     "                    .fit()\n",
     "                    .predict())\n",
     "    actual = comparison[variable]\n",
-    "    sns.distplot(predicted - actual, norm_hist=True, rug=True, ax=ax)\n",
+    "\n",
+    "    sns.histplot(predicted - actual, kde=True, stat='density', ax=ax)\n",
+    "    sns.rugplot(predicted-actual)\n",
+    "\n",
+    "    ax.set_ylim(bottom=0.0)\n",
     "    ax.set_xlabel(variable)\n",
     "    ax.set_ylabel('residual')\n",
     "fig.tight_layout();"
@@ -719,8 +727,10 @@
     "                .fit()\n",
     "                .predict())\n",
     "actual = comparison['d2']\n",
-    "sns.distplot(predicted - actual, rug=True, \n",
-    "             norm_hist = True, ax=ax2)\n",
+    "#sns.distplot(predicted - actual, rug=True, \n",
+    "#             norm_hist = True, ax=ax2)\n",
+    "sns.histplot(predicted - actual, kde=True, stat='density', ax=ax)\n",
+    "sns.rugplot(predicted-actual)\n",
     "\n",
     "ax2.set_title('histogram')\n",
     "ax2.set_xlim(-3,3)\n",
@@ -770,7 +780,11 @@
     "                    .fit()\n",
     "                    .predict())\n",
     "    actual = comparison[variable]\n",
-    "    sns.distplot(predicted - actual, norm_hist=True, rug=True, ax=ax)\n",
+    "    # FIXME:\n",
+    "    # sns.distplot(predicted - actual, norm_hist=True, rug=True, ax=ax)\n",
+    "    sns.histplot(predicted - actual, kde=True, stat='density', ax=ax)\n",
+    "    sns.rugplot(predicted-actual)\n",
+    "    \n",
     "    ax.set_xlabel(variable)\n",
     "    ax.set_ylabel('residual')\n",
     "\n",
@@ -795,7 +809,8 @@
     "                .fit()\n",
     "                .predict())\n",
     "actual = magic['log_d2']\n",
-    "sns.distplot(predicted - actual, rug=True, ax=ax2)\n",
+    "sns.histplot(predicted - actual, stat='density', kde=True, ax=ax2)\n",
+    "sns.rugplot(predicted - actual, ax=ax2)\n",
     "\n",
     "ax2.set_title('histogram')\n",
     "ax2.set_xlim(-.7, .7)\n",
@@ -829,5 +844,5 @@
  ],
  "metadata": {},
  "nbformat": 4,
- "nbformat_minor": 2
+ "nbformat_minor": 4
 }

11_Tuning_and_Pipelines_code.ipynb

@@ -238,7 +238,7 @@
     "knn = neighbors.KNeighborsClassifier()\n",
     "grid_knn = skms.GridSearchCV(knn, \n",
     "                             param_grid = param_grid, \n",
-    "                             iid=False, cv=3)"
+    "                             cv=3)"
    ]
   },
   {
@@ -283,7 +283,7 @@
     "knn = neighbors.KNeighborsClassifier()\n",
     "grid_knn = skms.GridSearchCV(knn, \n",
     "                             param_grid = param_grid, \n",
-    "                             iid=False, cv=2)\n",
+    "                             cv=2)\n",
     "\n",
     "outer_scores = skms.cross_val_score(grid_knn,\n",
     "                                    iris.data, \n",
@@ -432,7 +432,7 @@
    "outputs": [],
    "source": [
     "# iid to silence warning\n",
-    "mod = skms.GridSearchCV(pipe, param_grid, iid=False, n_jobs=-1)\n",
+    "mod = skms.GridSearchCV(pipe, param_grid, n_jobs=-1)\n",
     "mod.fit(diabetes.data, diabetes.target);"
    ]
   },

12_Combining_Learners_Ensemble_Methods_code.ipynb

@@ -28,7 +28,7 @@
    "metadata": {},
    "outputs": [],
    "source": [
-    "base_estimators = [linear_model.LogisticRegression(),\n",
+    "base_estimators = [linear_model.LogisticRegression(max_iter=10000),\n",
     "                   tree.DecisionTreeClassifier(max_depth=3),\n",
     "                   naive_bayes.GaussianNB()]\n",
     "base_estimators = [(get_model_name(m), m) for m in base_estimators]\n",
@@ -273,8 +273,11 @@
    "source": [
     "AdaBC  = ensemble.AdaBoostClassifier\n",
     "GradBC = ensemble.GradientBoostingClassifier\n",
+    "# NOTE: loss=\"deviance\" deprecated\n",
+    "#  see: https://github.com/scikit-learn/scikit-learn/pull/23036\n",
+    "#  and: https://github.com/scikit-learn/scikit-learn/issues/18248\n",
     "boosted_classifiers = {'boost(Ada)' : AdaBC(learning_rate=2.0),\n",
-    "                       'boost(Grad)' : GradBC(loss=\"deviance\")}\n",
+    "                       'boost(Grad)' : GradBC(loss=\"log_loss\")}\n",
     "mean_accs = {}\n",
     "for name, model in boosted_classifiers.items():\n",
     "    model.set_params(n_estimators=max_est)\n",

13_Feature_Engineering_II_Automated_code.ipynb

@@ -504,7 +504,7 @@
     "pipe = pipeline.make_pipeline(ftrsel, linear_model.LogisticRegression())\n",
     "\n",
     "param_grid = {'selectpercentile__percentile':[5,10,15,20,25]}\n",
-    "grid = skms.GridSearchCV(pipe, param_grid=param_grid, cv=3, iid=False)\n",
+    "grid = skms.GridSearchCV(pipe, param_grid=param_grid, cv=3)\n",
     "grid.fit(wine.data, wine.target)\n",
     "\n",
     "print(grid.best_params_)\n",
@@ -770,7 +770,7 @@
     "svc = svm.SVC(kernel='rbf')\n",
     "\n",
     "grid_model = skms.GridSearchCV(svc, param_grid = param_grid, \n",
-    "                               cv=10, iid=False)\n",
+    "                               cv=10)\n",
     "grid_model.fit(digits.data, digits.target);"
    ]
   },
@@ -1077,7 +1077,7 @@
     "\n",
     "cmap = plt.cm.Spectral\n",
     "fig = plt.figure(figsize=(4,4))\n",
-    "ax = plt.gca(projection='3d')\n",
+    "ax = fig.add_subplot(projection='3d')\n",
     "ax.scatter(*data_3d.T, c=color, cmap=cmap)\n",
     "ax.view_init(20, -50)"
    ]
@@ -1103,13 +1103,6 @@
     "data_2d = tsne.fit_transform(data_3d)\n",
     "axes[1].scatter(*data_2d.T, c=color, cmap=cmap);"
    ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "metadata": {},
-   "outputs": [],
-   "source": []
   }
  ],
  "metadata": {},