Label Propagation digits active learning

Demonstrates an active learning technique to learn handwritten digits using label propagation.

We start by training a label propagation model with only 10 labeled points, then we select the top five most uncertain points to label. Next, we train with 15 labeled points (original 10 + 5 new ones). We repeat this process four times to have a model trained with 30 labeled examples. Note you can increase this to label more than 30 by changing max_iterations. Labeling more than 30 can be useful to get a sense for the speed of convergence of this active learning technique.

A plot will appear showing the top 5 most uncertain digits for each iteration of training. These may or may not contain mistakes, but we will train the next model with their true labels.

Out:

Iteration 0 ______________________________________________________________________
Label Spreading model: 10 labeled & 320 unlabeled (330 total)
              precision    recall  f1-score   support

           0       0.00      0.00      0.00        24
           1       0.51      0.86      0.64        29
           2       0.83      0.97      0.90        31
           3       0.00      0.00      0.00        28
           4       0.00      0.00      0.00        27
           5       0.85      0.49      0.62        35
           6       0.84      0.95      0.89        40
           7       0.70      0.92      0.80        36
           8       0.57      0.76      0.65        33
           9       0.41      0.86      0.55        37

   micro avg       0.62      0.62      0.62       320
   macro avg       0.47      0.58      0.50       320
weighted avg       0.51      0.62      0.54       320

Confusion matrix
[[25  3  0  0  0  0  1]
 [ 1 30  0  0  0  0  0]
 [ 0  0 17  7  0  1 10]
 [ 2  0  0 38  0  0  0]
 [ 0  3  0  0 33  0  0]
 [ 8  0  0  0  0 25  0]
 [ 0  0  3  0  0  2 32]]
Iteration 1 ______________________________________________________________________
Label Spreading model: 15 labeled & 315 unlabeled (330 total)
              precision    recall  f1-score   support

           0       0.00      0.00      0.00        24
           1       0.51      0.75      0.61        28
           2       0.91      0.97      0.94        31
           3       0.00      0.00      0.00        28
           4       0.00      0.00      0.00        27
           5       0.84      0.97      0.90        33
           6       1.00      0.95      0.97        40
           7       0.75      0.92      0.83        36
           8       0.46      0.81      0.59        31
           9       0.43      0.78      0.56        37

   micro avg       0.66      0.66      0.66       315
   macro avg       0.49      0.61      0.54       315
weighted avg       0.53      0.66      0.58       315

Confusion matrix
[[21  0  0  0  0  6  1]
 [ 1 30  0  0  0  0  0]
 [ 0  0 32  0  0  0  1]
 [ 2  0  0 38  0  0  0]
 [ 0  3  0  0 33  0  0]
 [ 6  0  0  0  0 25  0]
 [ 0  0  6  0  0  2 29]]
Iteration 2 ______________________________________________________________________
Label Spreading model: 20 labeled & 310 unlabeled (330 total)
              precision    recall  f1-score   support

           0       1.00      1.00      1.00        22
           1       0.67      0.71      0.69        28
           2       0.94      0.97      0.95        31
           3       0.00      0.00      0.00        28
           4       0.85      0.92      0.88        24
           5       0.89      0.97      0.93        33
           6       1.00      0.95      0.97        40
           7       1.00      0.92      0.96        36
           8       0.50      0.81      0.62        31
           9       0.67      0.78      0.72        37

   micro avg       0.81      0.81      0.81       310
   macro avg       0.75      0.80      0.77       310
weighted avg       0.76      0.81      0.78       310

Confusion matrix
[[22  0  0  0  0  0  0  0  0]
 [ 0 20  0  1  0  0  0  6  1]
 [ 0  1 30  0  0  0  0  0  0]
 [ 0  1  0 22  0  0  0  1  0]
 [ 0  0  0  0 32  0  0  0  1]
 [ 0  2  0  0  0 38  0  0  0]
 [ 0  0  2  1  0  0 33  0  0]
 [ 0  6  0  0  0  0  0 25  0]
 [ 0  0  0  2  4  0  0  2 29]]
Iteration 3 ______________________________________________________________________
Label Spreading model: 25 labeled & 305 unlabeled (330 total)
              precision    recall  f1-score   support

           0       1.00      1.00      1.00        22
           1       0.68      0.85      0.75        27
           2       1.00      0.90      0.95        31
           3       1.00      0.77      0.87        26
           4       1.00      0.92      0.96        24
           5       0.89      0.97      0.93        33
           6       1.00      0.97      0.99        39
           7       0.95      1.00      0.97        35
           8       0.66      0.81      0.72        31
           9       0.97      0.78      0.87        37

   micro avg       0.90      0.90      0.90       305
   macro avg       0.91      0.90      0.90       305
weighted avg       0.91      0.90      0.90       305

Confusion matrix
[[22  0  0  0  0  0  0  0  0  0]
 [ 0 23  0  0  0  0  0  0  4  0]
 [ 0  1 28  0  0  0  0  2  0  0]
 [ 0  0  0 20  0  0  0  0  6  0]
 [ 0  1  0  0 22  0  0  0  1  0]
 [ 0  0  0  0  0 32  0  0  0  1]
 [ 0  1  0  0  0  0 38  0  0  0]
 [ 0  0  0  0  0  0  0 35  0  0]
 [ 0  6  0  0  0  0  0  0 25  0]
 [ 0  2  0  0  0  4  0  0  2 29]]
Iteration 4 ______________________________________________________________________
Label Spreading model: 30 labeled & 300 unlabeled (330 total)
              precision    recall  f1-score   support

           0       1.00      1.00      1.00        22
           1       0.68      0.85      0.75        27
           2       1.00      0.87      0.93        31
           3       0.92      1.00      0.96        23
           4       1.00      0.92      0.96        24
           5       0.97      0.94      0.95        33
           6       1.00      0.97      0.99        39
           7       0.95      1.00      0.97        35
           8       0.81      0.81      0.81        31
           9       0.94      0.86      0.90        35

   micro avg       0.92      0.92      0.92       300
   macro avg       0.93      0.92      0.92       300
weighted avg       0.93      0.92      0.92       300

Confusion matrix
[[22  0  0  0  0  0  0  0  0  0]
 [ 0 23  0  0  0  0  0  0  4  0]
 [ 0  1 27  1  0  0  0  2  0  0]
 [ 0  0  0 23  0  0  0  0  0  0]
 [ 0  1  0  0 22  0  0  0  1  0]
 [ 0  0  0  0  0 31  0  0  0  2]
 [ 0  1  0  0  0  0 38  0  0  0]
 [ 0  0  0  0  0  0  0 35  0  0]
 [ 0  6  0  0  0  0  0  0 25  0]
 [ 0  2  0  1  0  1  0  0  1 30]]

print(__doc__)

# Authors: Clay Woolam <[email protected]>
# License: BSD

import numpy as np
import matplotlib.pyplot as plt
from scipy import stats

from sklearn import datasets
from sklearn.semi_supervised import label_propagation
from sklearn.metrics import classification_report, confusion_matrix

digits = datasets.load_digits()
rng = np.random.RandomState(0)
indices = np.arange(len(digits.data))
rng.shuffle(indices)

X = digits.data[indices[:330]]
y = digits.target[indices[:330]]
images = digits.images[indices[:330]]

n_total_samples = len(y)
n_labeled_points = 10
max_iterations = 5

unlabeled_indices = np.arange(n_total_samples)[n_labeled_points:]
f = plt.figure()

for i in range(max_iterations):
    if len(unlabeled_indices) == 0:
        print("No unlabeled items left to label.")
        break
    y_train = np.copy(y)
    y_train[unlabeled_indices] = -1

    lp_model = label_propagation.LabelSpreading(gamma=0.25, max_iter=5)
    lp_model.fit(X, y_train)

    predicted_labels = lp_model.transduction_[unlabeled_indices]
    true_labels = y[unlabeled_indices]

    cm = confusion_matrix(true_labels, predicted_labels,
                          labels=lp_model.classes_)

    print("Iteration %i %s" % (i, 70 * "_"))
    print("Label Spreading model: %d labeled & %d unlabeled (%d total)"
          % (n_labeled_points, n_total_samples - n_labeled_points,
             n_total_samples))

    print(classification_report(true_labels, predicted_labels))

    print("Confusion matrix")
    print(cm)

    # compute the entropies of transduced label distributions
    pred_entropies = stats.distributions.entropy(
        lp_model.label_distributions_.T)

    # select up to 5 digit examples that the classifier is most uncertain about
    uncertainty_index = np.argsort(pred_entropies)[::-1]
    uncertainty_index = uncertainty_index[
        np.in1d(uncertainty_index, unlabeled_indices)][:5]

    # keep track of indices that we get labels for
    delete_indices = np.array([])

    # for more than 5 iterations, visualize the gain only on the first 5
    if i < 5:
        f.text(.05, (1 - (i + 1) * .183),
               "model %d\n\nfit with\n%d labels" %
               ((i + 1), i * 5 + 10), size=10)
    for index, image_index in enumerate(uncertainty_index):
        image = images[image_index]

        # for more than 5 iterations, visualize the gain only on the first 5
        if i < 5:
            sub = f.add_subplot(5, 5, index + 1 + (5 * i))
            sub.imshow(image, cmap=plt.cm.gray_r, interpolation='none')
            sub.set_title("predict: %i\ntrue: %i" % (
                lp_model.transduction_[image_index], y[image_index]), size=10)
            sub.axis('off')

        # labeling 5 points, remote from labeled set
        delete_index, = np.where(unlabeled_indices == image_index)
        delete_indices = np.concatenate((delete_indices, delete_index))

    unlabeled_indices = np.delete(unlabeled_indices, delete_indices)
    n_labeled_points += len(uncertainty_index)

f.suptitle("Active learning with Label Propagation.\nRows show 5 most "
           "uncertain labels to learn with the next model.", y=1.15)
plt.subplots_adjust(left=0.2, bottom=0.03, right=0.9, top=0.9, wspace=0.2,
                    hspace=0.85)
plt.show()

Total running time of the script: ( 0 minutes 0.992 seconds)

Gallery generated by Sphinx-Gallery