DOC more informative make_multilabel_classification example

jnothman · jnothman · commit 1815725f785b · 2014-08-21T23:17:48.000+10:00
diff --git a/doc/sphinxext/gen_rst.py b/doc/sphinxext/gen_rst.py
@@ -870,7 +870,7 @@ def generate_file_rst(fname, target_dir, src_dir, root_dir, plot_gallery):
                     my_stdout = my_stdout.replace(
                         my_globals['__doc__'],
                         '')
-                my_stdout = my_stdout.strip()
+                my_stdout = my_stdout.strip().expandtabs()
                 if my_stdout:
                     stdout = '**Script output**::\n\n  %s\n\n' % (
                         '\n  '.join(my_stdout.split('\n')))
diff --git a/examples/datasets/plot_random_multilabel_dataset.py b/examples/datasets/plot_random_multilabel_dataset.py
@@ -10,66 +10,87 @@
 Points are labeled as follows, where Y means the class is present:
 
     =====  =====  =====  ======
-          Classes        Color
-    -------------------  ------
+      1      2      3    Color
+    =====  =====  =====  ======
       Y      N      N    Red
       N      Y      N    Blue
       N      N      Y    Yellow
       Y      Y      N    Purple
       Y      N      Y    Orange
       Y      Y      N    Green
-      Y      Y      Y    Black
+      Y      Y      Y    Brown
     =====  =====  =====  ======
 
-Below the scatter of these data points we show the underlying class
-distribution.
+A star marks the expected sample for each class; its size reflects the
+probability of selecting that class label.
 
 The left and right examples highlight the ``n_labels`` parameter:
-more of the samples in the right plot have 2 or 3 classes.
+more of the samples in the right plot have 2 or 3 labels.
 
+Note that this two-dimensional example is very degenerate:
+generally the number of features would be much greater than the
+"document length", while here we have much larger documents than vocabulary.
+Similarly, with ``n_classes > n_features``, it is much less likely that a
+feature distinguishes a particular cluss.
 """
 
+from __future__ import print_function
 import numpy as np
 import matplotlib.pyplot as plt
 
 from sklearn.datasets import make_multilabel_classification as make_ml_clf
 
-COLORS = np.array(['!', 'r', 'b', 'purple', 'y', 'orange', 'g', 'k'])
+print(__doc__)
+
+COLORS = np.array(['!',
+                   '#FF3333',  # red
+                   '#0198E1',  # blue
+                   '#BF5FFF',  # purple
+                   '#FCD116',  # yellow
+                   '#FF7216',  # orange
+                   '#4DBD33',  # green
+                   '#87421F'   # brown
+                   ])
+
+# Use same random seed for multiple calls to make_multilabel_classification to
+# ensure same distributions
+RANDOM_SEED = np.random.randint(2 ** 10)
 
 
 def plot_2d(ax, n_labels=1, n_classes=3, length=50):
-    X, Y, p_c, p_w_c = make_ml_clf(n_samples=100, n_features=2,
+    X, Y, p_c, p_w_c = make_ml_clf(n_samples=150, n_features=2,
                                    n_classes=n_classes, n_labels=n_labels,
                                    length=length, allow_unlabeled=False,
                                    return_indicator=True,
-                                   return_distributions=True)
-
-    ax.scatter(X[:, 0], X[:, 1], color=COLORS.take((Y * [4, 2, 1]
-                                                    ).sum(axis=1)))
+                                   return_distributions=True,
+                                   random_state=RANDOM_SEED)
+
+    ax.scatter(X[:, 0], X[:, 1], color=COLORS.take((Y * [1, 2, 4]
+                                                    ).sum(axis=1)),
+               marker='.')
+    ax.scatter(p_w_c[0] * length, p_w_c[1] * length,
+               marker='*', linewidth=.5, edgecolor='black',
+               s=20 + 1500 * p_c ** 2,
+               color=COLORS.take([1, 2, 4]))
     ax.set_xlabel('Feature 0 count')
     return p_c, p_w_c
 
 
-_, ((ax1, ax2), (ax3, ax4)) = plt.subplots(2, 2,
-                                           sharex='row',
-                                           sharey='row')
-plt.subplots_adjust(hspace=.3)
+_, (ax1, ax2) = plt.subplots(1, 2, sharex='row', sharey='row', figsize=(8, 4))
+plt.subplots_adjust(bottom=.15)
 
+p_c, p_w_c = plot_2d(ax1, n_labels=1)
 ax1.set_title('n_labels=1, length=50')
 ax1.set_ylabel('Feature 1 count')
-p_c, p_w_c = plot_2d(ax1, n_labels=1)
+
+plot_2d(ax2, n_labels=3)
+ax2.set_title('n_labels=3, length=50')
 ax2.set_xlim(left=0, auto=True)
 ax2.set_ylim(bottom=0, auto=True)
 
-ax3.bar([1, 2, 3], p_c, color=COLORS.take([1, 2, 4]))
-ax3.set_ylabel('Class prior probability')
-
-ax2.set_title('n_labels=2, length=50')
-p_c, p_w_c = plot_2d(ax2, n_labels=2)
-
-ax4.bar([1, 2, 3], p_c, color=COLORS.take([1, 2, 4]))
-ax4.set_ylim(0, 1)
-ax4.set_xticks([])
+plt.show()
 
-###plt.show()
-plt.savefig('/tmp/foo.pdf')
+print('The data was generated from (random_state=%d):' % RANDOM_SEED)
+print('Class', 'P(C)', 'P(w0|C)', 'P(w1|C)', sep='\t')
+for k, p, p_w in zip(['red', 'blue', 'yellow'], p_c, p_w_c.T):
+    print('%s\t%0.2f\t%0.2f\t%0.2f' % (k, p, p_w[0], p_w[1]))
