def question10():
perceptron = Perceptron()
svm = SVM()
lda = LDA()
repeat = 500
mean_accuracies = np.empty((3, len(all_m)))
accuracies = np.empty((3, repeat))
for i, m in enumerate(all_m):
for j in range(repeat):
X, yx = draw_points_until_two_classes(m) # training set
Z, yz = draw_points_until_two_classes(k) # test set
perceptron.fit(X, yx)
svm.fit(X, yx)
lda.fit(X, yx)
accuracies[0, j] = perceptron.score(Z, yz)["accuracy"]
accuracies[1, j] = svm.score(Z, yz)["accuracy"]
accuracies[2, j] = lda.score(Z, yz)["accuracy"]
mean_accuracies[:, i] = accuracies.mean(axis=1)
models = ["Perceptron", "SVM", "LDA"]
colors = ['blue', 'red', 'green']
fig = plt.figure()
for i, model in enumerate(models):
plt.plot(all_m, mean_accuracies[i, :], color=colors[i], label=model)
plt.legend()
plt.title("Q10: mean accuracy as function of m")
fig.savefig("q10.png", bbox_inches='tight', pad_inches=0.2, dpi=fig.dpi)
def question9():
perceptron = Perceptron()
svm = SVM()
fig = plt.figure()
plt.suptitle("Q9: True vs. Perceptron vs. SVM hyperplanes")
for i, m in enumerate(all_m):
X, y = draw_points_until_two_classes(m)
svm.fit(X, y)
ax = fig.add_subplot(2, 3, i + 1)
ax.scatter(X[y == -1, 0], X[y == -1, 1], color="blue",
label="y=-1") # first class, labeled -1
ax.scatter(X[y == 1, 0], X[y == 1, 1], color="red",
label="y=1") # second class, labeled 1
xmin, xmax = plt.xlim()
xx = np.linspace(xmin, xmax)
true_hyperplane = a * xx - (b / w[1])
# perceptron
perceptron.fit(X, y)
w_perc = perceptron.model[:-1]
a_perceptron = -w_perc[0] / w_perc[1]
b_perceptron = perceptron.model[-1] / perceptron.model[1]
perceptron_hyperplane = a_perceptron * xx - b_perceptron
w_svm = svm.model.coef_[0]
a_svm = -w_svm[0] / w_svm[1]
b_svm = svm.model.intercept_[0] / w_svm[1]
SVM_hyperplane = a_svm * xx - b_svm
ax.plot(xx, true_hyperplane, color="black", label="true hyperplane")
ax.plot(xx,
perceptron_hyperplane,
color="green",
label="perceptron hyperplane")
ax.plot(xx, SVM_hyperplane, color="orange", label="svm hyperplane")
ax.title.set_text("m=" + str(m))
if i == 0: plt.legend()
fig.savefig("q9.png", bbox_inches='tight', pad_inches=0.3, dpi=fig.dpi)
# Y_train = (Y_train-1)/2.0
X[:, 0:X_train.shape[1]] = X_train
Y[:, 0:Y_train.shape[1]] = Y_train
Y = to_categorical(Y)
X = np.reshape(X, (X.shape[0], X.shape[1], 4))
#train
model = SVM((X.shape[1], 4))
model._get_distribution_strategy = lambda: None
json_string = model.to_json()
open('model.json', 'w').write(json_string)
checkpointer = ModelCheckpoint(path_best_weights,
verbose=1,
monitor='val_loss',
mode='auto',
save_best_only=True)
tbCallback = TensorBoard(log_dir='./logs',
histogram_freq=0,
write_graph=True,
write_images=True,
profile_batch=100000000)
model.fit(X,
Y,
epochs=N_epochs,
batch_size=batch_size,
verbose=2,
shuffle=True,
validation_split=0.2,
callbacks=[checkpointer])
model.save_weights(path_last_weights, overwrite=True)
results0 = np.zeros(3000)
len_files = len(FILES)
for i in range(len_files):
γ = gamma_list[i]
λ = lambda_list[i]
X_train, Y_train, X_test = load_data(i,
data_dir=DATA_DIR,
files_dict=FILES)
kernel = GaussianKernel(γ)
clf = SVM(_lambda=λ, kernel=kernel)
clf.fit(X_train, Y_train)
y_pred = clf.predict(X_test)
results0[i * 1000:i * 1000 + 1000] = y_pred
# SAVE Results
save_results("results_SVM_gaussian.csv", results0, RESULT_DIR)
print("1/3 Ending SVM with Gaussian kernel...")
#####################################
# 2) SVM with Convolutional kernel #
#####################################
print("2/3 Starting SVM with Convolutional kernel...")
# Define parameters lists
sigma_list = [0.31, 0.31, 0.3]
k_list = [9, 10, 11]
lambda_list = [1e-5, 1e-9, 1e-9]
