def my_supervised_mnist(X, y=None, l1=.1, **kwargs):
if (X.shape[1] != 28 * 28):
print("Error Dataset")
return range(len(X.shape[1]))
rrfs = RRFS(28 * 28, hidden=2)
codes = getCodes(X.reshape(X.shape[0], 28, 28, 1))
score = rrfs.train_fs_network(X, rep=codes, l1=l1, epochs=300, loss='mse')
idx = np.argsort(score)[::-1]
return idx
def my_se(X, y=None, l1=.1, n_components=2, **kwargs):
rrfs = RRFS(X.shape[1], hidden=n_components)
model = SpectralEmbedding(n_components=n_components)
codes = model.fit_transform(X)
codes = (codes - np.min(codes)) / (np.max(codes) - np.min(codes))
#rrfs.train_representation_network(x_train, name=dataset+'_rep.hd5', epochs=1000)
score = rrfs.train_fs_network(X, rep=codes, l1=l1, epochs=300, loss='mse')
# sort the feature scores in an ascending order according to the feature scores
idx = np.argsort(score)[::-1]
return idx
def my_autoencoder(X, y=None, l1=.1, n_components=2, **kwargs):
rrfs = RRFS(X.shape[1], hidden=n_components)
input_tensor = Input(shape=(X.shape[1], ))
l1 = Dense(10, activation='relu')(input_tensor)
l2 = Dense(n_components, activation='relu')(l1)
l3 = Dense(10, activation='relu')(l2)
output_tensor = Dense(X.shape[1])(l3)
model = Model(input_tensor, output_tensor)
encoder = Model(input_tensor, l2)
model.compile(optimizer='adam', loss='mse')
model.fit(X, X, epochs=300)
codes = encoder.predict(X)
codes = (codes - np.min(codes)) / (np.max(codes) - np.min(codes))
#rrfs.train_representation_network(x_train, name=dataset+'_rep.hd5', epochs=1000)
score = rrfs.train_fs_network(X, rep=codes, l1=l1, epochs=300, loss='mse')
# sort the feature scores in an ascending order according to the feature scores
idx = np.argsort(score)[::-1]
return idx
test_indexes += list(c_test_indexes)
x_test = X[test_indexes,:]
y_test = Y[test_indexes]
x_train = X[train_indexes,:]
y_train = Y[train_indexes]
t_max = np.max(x_train)
t_min = np.min(x_train)
#scaler = preprocessing.StandardScaler()
x_train = (x_train.astype(float)-t_min)/(t_max-t_min)
x_test = (x_test.astype(float)-t_min)/(t_max-t_min)
#x_test = scaler.transform(x_test)
rrfs = RRFS(dim, loss='mse')
rrfs.train_representation_network(x_train, name=dataset+'_rep.hd5', epochs=1000)
ps = [2, 4, 6, 8, 10, 20 , 30, 40 , 50, 60, 70, 80, 100]
l1s = [0.0001,.001,.005,.01,.1,1]
accs_ps = np.zeros(len(ps))
max_features = []
all_features ={}
for i, p in enumerate(ps):
num_features = int(p*dim/100)
print('NUM === %d'%num_features)
accs_l1 = np.zeros(len(l1s))
acc_max = 0
for index,l1 in enumerate(l1s):
x_test = X[test_indexes, :]
y_test = Y[test_indexes]
x_train = X[train_indexes, :]
y_train = Y[train_indexes]
#scaler = preprocessing.MinMaxScaler()
#x_train = scaler.fit_transform(x_train)
#x_test = scaler.transform(x_test)
mmax = np.max(x_train)
mmin = np.min(x_train)
x_train = (x_train - mmin) / (mmax - mmin)
x_test = (x_test - mmin) / (mmax - mmin)
rrfs = RRFS(dim, hidden=2)
tsne = TSNE()
tsne_codes = tsne.fit_transform(x_train)
tsne_codes = (tsne_codes - np.min(tsne_codes)) / (np.max(tsne_codes) -
np.min(tsne_codes))
#rrfs.train_representation_network(x_train, name=dataset+'_rep.hd5', epochs=1000)
ps = [2, 4, 6, 8, 10, 20, 30, 40, 50, 60, 70, 80, 100]
l1s = [1e-5, 0.0001, .001, .005, .01, .05, .1]
accs_ps = np.zeros(len(ps))
fatures_ps_l1s = {}
accs_ps_l1s = {}
for i, p in enumerate(ps):
num_features = int(p * dim / 100)
accs_l1 = np.zeros(len(l1s))