def run(filename, gamma, m, k, ktype, outfig, critoutfig, testfile):
digitsdat = Data()
digitsdat.load_svmlight(filename,
gamma=gamma,
docalkernel=False,
savefile=None,
testfile=testfile,
dobin=False)
if ktype == 0:
digitsdat.calculate_kernel()
print "Running Kernel type : global "
else:
digitsdat.calculate_kernel_individual()
print "Running Kernel type : local "
# selected = greedy_parallel(digitsdat.kernel, m)
# print np.sort(selected)
selected = greedy_select_protos(
digitsdat.kernel, np.array(range(np.shape(digitsdat.kernel)[0])), m)
# print np.sort(selected)
# critselected = select_criticism(digitsdat.kernel, selected, k)
selectedy = digitsdat.y[selected]
sortedindx = np.argsort(selectedy)
critselected = None
if outfig is not None:
plotfigs2(digitsdat.X[selected[sortedindx], :], selectedy[sortedindx],
outfig)
if k > 0:
critselected = select_criticism_regularized(digitsdat.kernel,
selected,
k,
is_K_sparse=False,
reg='logdet')
critselectedy = digitsdat.y[critselected]
critsortedindx = np.argsort(critselectedy)
if critoutfig is not None:
plotfigs2(digitsdat.X[critselected[critsortedindx], :],
critselectedy[critsortedindx], critoutfig + reg)
return selected, critselected, digitsdat
data_record = data_record.reset_index()
feature_columns = [c for c in data_record if c.startswith('machine_')]
X = data_record[feature_columns].as_matrix()
y = data_record['sleep_stage'].as_matrix()
data = Data()
data.load_data(X, y, gamma=0.026, docalkernel=False, savefile=None, testfile=None, dobin=False)
# Calculate global kernel
data.calculate_kernel()
# Calculate local kernel
#data.calculate_kernel_individual()
num_examples_to_select_per_class = 10
num_examples_to_select = 40
num_prototypes_to_select = num_examples_to_select
prototypes_selected = greedy_select_protos(data.kernel, np.array(range(np.shape(data.kernel)[0])), num_prototypes_to_select)
prototypes_selected_y = data.y[prototypes_selected]
prototypes_selected_wake = prototypes_selected[prototypes_selected_y == 0][:num_examples_to_select_per_class]
prototypes_selected_N1 = prototypes_selected[prototypes_selected_y == 1][:num_examples_to_select_per_class]
num_criticisms_to_select = num_examples_to_select
criticisms_selected = select_criticism_regularized(data.kernel, prototypes_selected, num_criticisms_to_select, is_K_sparse=False, reg='logdet')
criticisms_selected_y = data.y[criticisms_selected]
criticisms_selected_wake = criticisms_selected[criticisms_selected_y == 0][:num_examples_to_select_per_class]
criticisms_selected_N1 = criticisms_selected[criticisms_selected_y == 1][:num_examples_to_select_per_class]
examples_selected[data_record_metadata['file_name']] = {
'prototypes': {
'Wake': data_record['epoch_start_in_seconds'][prototypes_selected_wake].tolist(),
'N1': data_record['epoch_start_in_seconds'][prototypes_selected_N1].tolist(),
},
def prototype_selection(X, subsample=20, kernel='rbf'):
return greedy_select_protos(
pairwise_kernels(X, metric=kernel), np.array(range(X.shape[0])),
subsample) if subsample > 1 else np.array(range(X.shape[0]))