W_test = pca.transform(X_test)
params = {'C': 1, 'gamma': 2e-4, 'kernel': 'linear'}
ovr = OVR(**params)
ovr.fit(W_train, y_train)
y_hat = ovr.predict(W_test[::-1]).ravel()
done = {'success': False, 'failure': False}
fig, axes = plt.subplots(ncols=2)
for y, t, w in zip(y_hat, y_test.T.ravel(), W_test.T):
if y == t and done['success'] is False:
x_hat = pca.reconstruct(w)
axes[0].imshow(x_hat.reshape(SHAPE).T, cmap=plt.get_cmap('gray'))
axes[0].set_title(
'Successful SVM Classification\nPredicted Class: %d, True Class: %d'
% (y, t),
color=_palette[1])
done['success'] = True
elif y != t and done['failure'] is False:
x_hat = pca.reconstruct(w)
axes[1].imshow(x_hat.reshape(SHAPE).T, cmap=plt.get_cmap('gray'))
axes[1].set_title(
'Failed SVM Classification\nPredicted Class: %d, True Class: %d' %
(y, t),
color=_palette[2])
done['failure'] = True
elif done['failure'] is True and done['success'] is True:
ovr = OVR(**params)
ovr.fit(W_train, y_train)
n_idx = 4
np.random.seed(13)
fig, axes = plt.subplots(nrows=2, ncols=n_idx)
idx = np.array(list(ovr.classifiers.keys()))[np.random.choice(
len(ovr.classifiers.keys()), n_idx)]
for l, i in zip(idx, range(n_idx)):
w_hat = ovr.classifiers[l].support_vectors_[0]
x_hat = pca.reconstruct(w_hat)
axes[0, i].imshow(x_hat.reshape(SHAPE).T, cmap=plt.get_cmap('gray'))
axes[0, i].set_title('OVR\nSupport Vector for label %d\n' % l)
ovo = OVO(**params)
ovo.fit(W_train, y_train)
idx = np.array(list(ovo.classifiers.keys()))[np.random.choice(
len(ovo.classifiers.keys()), n_idx)]
for l, i in zip(idx, range(n_idx)):
w_hat = ovo.classifiers[tuple(l)].support_vectors_[0]
x_hat = pca.reconstruct(w_hat)