What if there are multiple knee points? about kneed

[arvkevi]

Hello @shineisagoodgirl, thanks for your question. There is a unit test that checks the knee points of a sine wave. The trick is that direction and curve need to be modified to detect different types of knees in one line. Kneed can't guess which curve is the one the user is trying to identify.

You could use the code from the unit test to help get you started:

x = np.arange(0, 10, 0.1)
y_sin = np.sin(x)

sine_combos = [
    ('decreasing', 'convex'),
    ('increasing', 'convex'),
    ('increasing', 'concave'),
    ('decreasing', 'concave'),
]
expected_knees = [4.5, 4.9, 7.7, 1.8]
detected_knees = []
for direction, curve in sine_combos:
    kl_sine = KneeLocator(
        x, y_sin, direction=direction, curve=curve, S=1, online=True
    )
    detected_knees.append(kl_sine.knee)
assert np.isclose(expected_knees, detected_knees).all()

You could plot this:

plt.style.use("fivethirtyeight")
plt.figure(figsize=(8, 8))
plt.plot(x, y_sin)
plt.vlines(
    detected_knees[0],
    min(y_sin),
    max(y_sin),
    linestyles="--",
    color="#fc4f30",
    label="convex decreasing",
)
plt.vlines(
    detected_knees[1],
    min(y_sin),
    max(y_sin),
    linestyles="--",
    color="#e5ae38",
    label="convex increasing",
)
plt.vlines(
    detected_knees[2],
    min(y_sin),
    max(y_sin),
    linestyles="--",
    color="#6d904f",
    label="concave increasing",
)
plt.vlines(
    detected_knees[3],
    min(y_sin),
    max(y_sin),
    linestyles="--",
    color="#8b8b8b",
    label="concave decreasing",
)
plt.legend()

Notice that the concave increasing knee could be at two locations in the figure. If there is more than one potential knee point for a line, kneed can show you all the candidate knee points by accessing the attribute .all_knees.

kl_sine = KneeLocator(
    x, y_sin, direction='increasing', curve='concave', S=1, online=True
)

print(kl_sine.all_knees)
{1.4000000000000001, 7.7}