def roc_curves(n_samples, p_value=False):
"""
"""
data = generate_data(n_samples=n_samples)
X, Y = data[:, :-1], data[:, -1]
if p_value:
I = np.abs(0.5 - compute_values(X))
else:
I = mutual_information(X, Y)
tau_min, tau_max = I.min(), I.max()
if not p_value and tau_min < 0:
tau_min = 0
print "warning - tau_min < 0"
step = 500
erange = [tau_min + i * (tau_max - tau_min) * 1. / step
for i
in range(step)]
roc_x = []
roc_y = []
for tau in erange:
pos = I > tau
tp = pos[1000:].sum()
tn = (1 - pos[:1000]).sum()
roc_x.append(float(tp) / (200))
roc_y.append(1 - float(tn) / (1000))
return np.array(roc_x), np.array(roc_y), np.array(erange)
def compute_ratios(tau, p_value=False):
data = generate_data(n_samples=1000)
X, Y = data[:, :-1], data[:, -1]
if p_value:
I = np.abs(0.5 - compute_values(X))
else:
I = mutual_information(X, Y)
pos = I > tau
fp = pos[:1000].sum()
fn = (1 - pos[1000:]).sum()
tp = pos[1000:].sum()
tn = (1 - pos[:1000]).sum()
return fp, fn, tp, tn