def load_data_fourth_record():
i, p = fourth_record() #i, interictal;p, preictal
#arbitrarily assign 1,0 to preictal, 0,1 to interictal
pre = np.array((1, 0)).reshape(1, 2)
inter = np.array((0, 1)).reshape(1, 2)
#we have 630 interictal, 431 preictal
prem = np.repeat(pre, len(p), axis=0)
interm = np.repeat(inter, len(i), axis=0)
labels = np.concatenate((prem, interm), axis=0)
data = np.concatenate((p, i), axis=0)
idx = np.random.permutation(data.shape[0])
x, y = data[idx, :22, :], labels[idx, :]
# x = util.normalize(x)
x = np.abs(np.fft.fftn(x, axes=(1, 2)))
x = util.normalize(x)
x_training = x[:len(x) // 2, :, :]
y_training = y[:len(y) // 2, :]
x_val = x[len(x) // 2:len(x) // 2 + len(x) // 4, :, :]
y_val = y[len(y) // 2:len(y) // 2 + len(y) // 4, :]
x_test = x[len(x) // 2 + len(x) // 4:len(x), :, :]
y_test = y[len(y) // 2 + len(y) // 4:len(y), :]
return x_training, y_training, x_val, y_val, x_test, y_test
def pca_test():
d = [5, 10, 15, 20, 30, 40, 50, 75, 100]
acc = np.zeros((9))
sen = np.zeros((9))
spe = np.zeros((9))
for i, di in enumerate(d):
print(f"Given d = {di}:")
tr = util.normalize(xt_f, axs = 1)
te = util.normalize(xte_f, axs = 1)
tr = PCA.pca(tr, di)
te = PCA.pca(te, di)
knn.fit_model(tr, yt_f)
acc[i], spe[i], sen[i] = knn.test(te, yte_f, verbose = True, pp=2)
print("") # skip one line
def p_test():
pii = [1.5,2, 2.5, 3,3.5]
acc = np.zeros((6))
sen = np.zeros((6))
spe = np.zeros((6))
for i, pi in enumerate(pii):
print(f"Given p = {pi}:")
knn.set_k(1)
tr = util.normalize(xt_f, axs = 1)
te = util.normalize(xte_f, axs = 1)
tr = PCA.pca(tr, 10)
te = PCA.pca(te, 10)
knn.fit_model(tr, yt_f)
acc[i], spe[i], sen[i] = knn.test(te, yte_f, verbose = True, pp = pi)
print("") # skip one line
def k_test():
ks = [1,3,5,7,9]
acc = np.zeros((5))
sen = np.zeros((5))
spe = np.zeros((5))
for i, ki in enumerate(ks):
print(f"Given k = {ki}:")
knn.set_k(ki)
tr = util.normalize(xt_f, axs = 1)
te = util.normalize(xte_f, axs = 1)
tr = PCA.pca(tr, 10)
te = PCA.pca(te, 10)
knn.fit_model(tr, yt_f)
acc[i], spe[i], sen[i] = knn.test(te, yte_f, verbose = True, pp = 2)
print("") # skip one line
tr = util.normalize(xt_f, axs = 1)
te = util.normalize(xte_f, axs = 1)
tr = PCA.pca(tr, 10)
te = PCA.pca(te, 10)
knn.fit_model(tr, yt_f)
acc[i], spe[i], sen[i] = knn.test(te, yte_f, verbose = True, pp = pi)
print("") # skip one line
knn.set_k(1)
tr = util.normalize(xt_f, axs = 1)
te = util.normalize(xte_f, axs = 1)
tr = PCA.pca(tr, 10)
te = PCA.pca(te, 10)
knn.fit_model(tr, yt_f)
knn.test(te, yte_f, verbose = True, pp = 3)
# import matplotlib.pyplot as plt
# plt.plot(p,acc, label="acc")
# plt.plot(p,spe, label="spe")
# plt.plot(p,sen, label="sen")
# plt.legend(loc = "upper right")
C = 320, #10
N = 927)
ss.add(svm)
ss.set_model_parameters(learning_rate = 0.00001, #0.001
rho = 0.975,
optimizer = "SGD+Momentum",
loss = "SoftMargin")
xtf,ytf,xvf,yvf,xtef,ytef = util.extract_linear_features(xt,yt,xv,yv,xte,yte)
#normalize...
xtf = util.normalize(xtf, axs = 1)
xvf = util.normalize(xvf, axs = 1)
xtef = util.normalize(xtef, axs = 1)
xtf = util.add_bias_term(xtf)
xvf = util.add_bias_term(xvf)
xtef = util.add_bias_term(xtef)
ss.fit_model(xtf, ytf, #for training
xvf, yvf, #for validation
epoch = 1000,
verbose = True,
perEpochVerbose = 100,
weightSavePerEpoch = 201)
# ss.load_weights(9801)
