elif dec_name == 'quadratic':
feat_calc = soil.QuadraticSurfaceCalculator(
2,
np.array([X[:, 0].min(), X[:, 0].max(), X[:, 1].min(), X[:, 1].max()],
dtype='float32'),
random_seed=rs)
cls = []
lm = []
for i in range(n_trees):
cls.append(
soil.ThresholdDecider(
feat_sel_prov, feat_calc,
soil.RandomizedClassificationThresholdOptimizer(
suggestions,
n_classes,
soil.EntropyGain(soil.ShannonEntropy()),
random_seed=rs)))
lm.append(soil.ClassificationLeafManager(n_classes))
forest = soil.Forest(depth, 1, 2, n_trees, cls, lm,
soil.ClassicTraining(soil.NoBagging()))
forest.fit(X, Y)
if INTERACTIVE:
plt.figure()
point_prob_plot(forest, X, Y, plotx, ploty)
plt.title('Random Forest $(T=%d, D=%d, |\Theta_j|=%d)$' %
(n_trees, depth, suggestions * 2))
plt.savefig('spiral_rf.png')
plt.show()
assert accuracy_score(Y, np.argmax(forest.predict(X), axis=1)) > 0.95
X, Y = make_spiral(n_arms=n_classes, noise=.4)
##############################################################################
parameters = {'kernel': ['rbf'],
'C': [1, 10, 100, 1000, 10000, 100000],
'gamma': [10 ** x for x in range(-5, 3)],
'probability':[True]}
svr = svm.SVC()
clf = grid_search.GridSearchCV(svr, parameters)
clf.fit(X, Y.ravel())
svmp = probaproxy(clf.predict_proba)
plt.figure()
point_prob_plot(svmp, X, Y, plotx, ploty)
plt.title('RBF kernel SVM $(\gamma=%d, C=%f)$' % (clf.best_params_['gamma'], clf.best_params_['C']))
plt.savefig('spiral_svm.png')
plt.show()
##############################################################################
parameters = {'C': [1, 10, 100, 1000, 10000],
'probability': [True]}
svr = svm.SVC(kernel='linear', max_iter=20000)
clf = grid_search.GridSearchCV(svr, parameters)
clf.fit(X, Y.ravel())
svmp = probaproxy(clf.predict_proba)
plt.figure()
point_prob_plot(svmp, X, Y, plotx, ploty)
n_trees = 200
# These variables will contain the classifiers and leaf managers for each tree.
cls = []
lm = []
for i in range(n_trees):
# Classifier:
cls.append(soil.ThresholdDecider(
soil.StandardFeatureSelectionProvider(1, 2, 2, 2, random_seed=1+i),
soil.LinearSurfaceCalculator(400, random_seed=1+i),
soil.RandomizedClassificationThresholdOptimizer(
True,
2,
soil.EntropyGain(soil.ShannonEntropy()), random_seed=1+i)))
# Leaf manager:
lm.append(soil.ClassificationLeafManager(2))
forest = soil.Forest(depth,
1,
2,
n_trees,
cls,
lm,
soil.ClassicTraining(soil.NoBagging()))
forest.fit(X, Y)
assert accuracy_score(Y, np.argmax(forest.predict(X), axis=1)) > 0.95
if INTERACTIVE:
plt.figure()
point_prob_plot(forest, X, Y, plotx, ploty)
plt.savefig('custom.png')
plt.show()
##############################################################################
parameters = {
'kernel': ['rbf'],
'C': [1, 10, 100, 1000, 10000, 100000],
'gamma': [10**x for x in range(-5, 3)],
'probability': [True]
}
svr = svm.SVC()
clf = grid_search.GridSearchCV(svr, parameters)
clf.fit(X, Y.ravel())
svmp = probaproxy(clf.predict_proba)
plt.figure()
point_prob_plot(svmp, X, Y, plotx, ploty)
plt.title('RBF kernel SVM $(\gamma=%d, C=%f)$' %
(clf.best_params_['gamma'], clf.best_params_['C']))
plt.savefig('spiral_svm.png')
plt.show()
##############################################################################
parameters = {'C': [1, 10, 100, 1000, 10000], 'probability': [True]}
svr = svm.SVC(kernel='linear', max_iter=20000)
clf = grid_search.GridSearchCV(svr, parameters)
clf.fit(X, Y.ravel())
svmp = probaproxy(clf.predict_proba)
plt.figure()
point_prob_plot(svmp, X, Y, plotx, ploty)
h = SimpleNode(sx, 2, 4)
h2 = SimpleNode(h, 4, 2)
out = SimpleNode(h2, 2, 1, nlin=pynnet.nlins.sigmoid)
cost = errors.mse(out, sy)
theano.config.blas.ldflags = ''
eval = theano.function([sx], out.output)
test = theano.function([sx, sy], cost.output)
train = theano.function([sx, sy],
cost.output,
updates=get_updates(cost.params, cost.output, 0.01))
print("Error at start:", test(X, Y))
for i in range(200000):
train(X, Y)
print("Error after 200000:", test(X, Y))
def pfunc(x):
return 1. - eval(x)
clp = probaproxy(pfunc)
plt.figure()
point_prob_plot(clp, X, Y, plotx, ploty)
plt.title('ANN (2 hidden layers, 4, 2)')
plt.savefig('spiral_ann.png')
plt.show()
#%% Build ANN.
sx = theano.tensor.matrix('x')
sy = theano.tensor.matrix('y')
h = SimpleNode(sx, 2, 4)
h2 = SimpleNode(h, 4, 2)
out = SimpleNode(h2, 2, 1, nlin=pynnet.nlins.sigmoid)
cost = errors.mse(out, sy)
theano.config.blas.ldflags=''
eval = theano.function([sx], out.output)
test = theano.function([sx, sy], cost.output)
train = theano.function([sx, sy], cost.output,
updates=get_updates(cost.params, cost.output, 0.01))
print("Error at start:", test(X, Y))
for i in range(200000):
train(X, Y)
print("Error after 200000:", test(X, Y))
def pfunc(x):
return 1. - eval(x)
clp = probaproxy(pfunc)
plt.figure()
point_prob_plot(clp, X, Y, plotx, ploty)
plt.title('ANN (2 hidden layers, 4, 2)')
plt.savefig('spiral_ann.png')
plt.show()