def main():
layers_conf = [[784, 10], [784, 100, 10], [784, 100, 100, 10],
[784, 100, 100, 100, 10]]
results = []
for layers in layers_conf:
print(layers)
clf = TFDeep(layers, 5e-3, 1e-4)
clf.train_mb(mnist.train.images, mnist.train.labels, 25)
for digit in digits:
probs = clf.eval(np.array([[*digit]]))
print(np.argmax(probs, axis=1))
Y_ = np.argmax(clf.eval(mnist.test.images), axis=1)
Y = np.argmax(mnist.test.labels, axis=1)
# acc, pr, m = data.eval_perf_multi(Y, Y_)
# print(acc)
# print(sum([p for p, r in pr]) / 10, [p for p, r in pr])
# print(sum([r for p, r in pr]) / 10, [r for p, r in pr])
# print(m)
# results.append(
# (
# acc,
# (sum([p for p, r in pr]) / 10, [p for p, r in pr]),
# (sum([r for p, r in pr]) / 10, [r for p, r in pr]),
# m
# )
# )
Y_best = np.argmax(clf.eval_with_weights(mnist.test.images, clf.best_W,
clf.best_b),
axis=1)
acc, pr, m = data.eval_perf_multi(Y, Y_best)
print(acc)
print(sum([p for p, r in pr]) / 10, [p for p, r in pr])
print(sum([r for p, r in pr]) / 10, [r for p, r in pr])
print(m)
results.append(
(acc, (sum([p for p, r in pr]) / 10, [p for p, r in pr]),
(sum([r for p, r in pr]) / 10, [r for p, r in pr]), m))
if clf.N == 1:
W = clf.best_W
for weights in W[0].T:
plt.imshow(weights.reshape((28, 28)),
cmap=plt.get_cmap('gray'),
vmin=min(weights),
vmax=max(weights))
plt.show()
for l, r in zip(layers_conf, results):
print(l)
for i in r:
print(i)
def svm_sol():
for kernel in ['linear', 'rbf']:
print(kernel)
clf = ksvm_wrap.KSVMWrap(mnist.train.images,
np.argmax(mnist.train.labels, axis=1),
kernel=kernel)
Y_ = clf.predict(mnist.test.images)
Y = np.argmax(mnist.test.labels, axis=1)
print(Y, Y_)
acc, pr, m = data.eval_perf_multi(Y, Y_)
print(acc)
print(sum([p for p, r in pr]) / 10, [p for p, r in pr])
print(sum([r for p, r in pr]) / 10, [r for p, r in pr])
print(m)
def main():
# inicijaliziraj generatore slučajnih brojeva
np.random.seed(100)
tf.set_random_seed(100)
# instanciraj podatke X i labele Yoh_
X, Y_ = data.sample_gmm_2d(6, 2, 10)
Yoh_ = data.class_to_onehot(Y_)
# izgradi graf:
tflr = TFDeep([2, 3, 2, 2], 0.1, 1e-4)
tflr.count_params()
# nauči parametre:
tflr.train(X, Yoh_, int(1e4))
# dohvati vjerojatnosti na skupu za učenje
probs = tflr.eval(X)
Y = np.argmax(probs, axis=1)
# ispiši performansu (preciznost i odziv po razredima)
accuracy, recall, precision = data.eval_perf_multi(Y, Y_)
print('acc', accuracy)
print('recall', recall)
print('precision', precision)
# iscrtaj rezultate, decizijsku plohu
rect = (np.min(X, axis=0), np.max(X, axis=0))
data.graph_surface(lambda x: np.argmax(tflr.eval(x), axis=1),
rect,
offset=0.5)
# graph the data points
data.graph_data(X, Y_, Y, special=[])
data.plt.show()
# instanciraj podatke X i labele Yoh_
X, Y_ = data.sample_gauss_2d(3, 100)
Yoh_ = data.class_to_onehot(Y_)
# izgradi graf:
tflr = TFLogreg(X.shape[1], Yoh_.shape[1], 0.5, 1e-2)
# nauči parametre:
tflr.train(X, Yoh_, 1000)
# dohvati vjerojatnosti na skupu za učenje
probs = tflr.eval(X)
Y = np.argmax(probs, axis=1)
# ispiši performansu (preciznost i odziv po razredima)
accuracy, recall, precision = data.eval_perf_multi(Y, Y_)
print('acc', accuracy)
print('recall', recall)
print('precision', precision)
# iscrtaj rezultate, decizijsku plohu
rect = (np.min(X, axis=0), np.max(X, axis=0))
data.graph_surface(lambda x: np.argmax(tflr.eval(x), axis=1),
rect,
offset=0.5)
# graph the data points
data.graph_data(X, Y_, Y, special=[])
data.plt.show()