User:Caron/memo

Graphviz

import graphviz # type: ignore

dgraph = graphviz.Digraph(**kwargs)
for i, _ in enumerate(self._data):
    dgraph.node(self._gn(i))
for i, _data in enumerate(self._data):
    label = str(_data[1]-_data[0])
    for nxt in np.unique(_data[8:14])[:-1]: # 255は除外
        dgraph.edge(self._gn(i), self._gn(nxt), label=label)

if render:
    dgraph.render(view=view)
elif view:
    dgraph.view()

Smoothing:Catmull–Clark

faceは不要なのでCatmull-clarkの途中処理まで
not tested yet

counts = round(counts) if isinstance(counts, float) else counts
for _ in range(counts):
    data_size = len(target)
    smoothed = create_new_data(self._sectionname, 2, 2*data_size-1)

    smoothed[::2,:] = target # type: ignore

    # Center position
    smoothed[1::2,0:3] = ((target[:-1,0:3] + target[1:,0:3])/2)
    # Average((cur + center_1 + center_2)/3)
    smoothed[2:-2:2,0:3] = (smoothed[1:-2:2,0:3] + smoothed[3::2,0:3] + smoothed[2:-2:2,0:3])/3

    # update settings
    smoothed = update_settings(
        iter = range(1, len(smoothed)-1),
        old_array=target,
        new_array=smoothed
        )

    # clone array for next iteration.
    target = smoothed.copy()

Smoothing:Spline interpolation

こっちの方が元の曲線に近くフィットしてくれる
not tested yet

pos = target[:, :3].T
m = len(pos[0])

# create knots vector
tck, _ = interpolate.splprep(pos, s=m+np.sqrt(m), k=4)
u_counts = np.linspace(0,1,round(m*counts))

# fitting
fitted = np.vstack(interpolate.splev(u_counts, tck)).T

### update settings: start
data_size = len(fitted)
update_idx = np.zeros([0]*data_size)
# find nearest positions
for idx in range(m):
    old_vec = np.tile(pos.T[idx], (data_size, 1))
    norm = np.linalg.norm(old_vec - fitted, axis=-1)
    update_idx[idx] = np.argmin(norm)
smoothed = create_new_data(self._sectionname, 2, data_size)
smoothed[:,:3] = fitted
# POTI Setting is start at target[3:], but ENPT, ITPT are target[4:].
start = 4 if self._sectionname != 'POTI' else 3 
smoothed[update_idx,start:] = target[:,start:]
smoothed = update_settings(
    iter=range(1, m),
    old_array=target,
    new_array=smoothed,
    spline_index=update_idx
)

Convex Quadrilaterals or not

緑の角度の総和が2πになればよい。

def is_convex(pt1: np.ndarray, pt2: np.ndarray) -> bool:
    pos = np.vstack([pt1.reshape(2,2), pt2.reshape(2,2)[::-1]])
    flat = pos.flatten()
    if flat.shape[0] != np.unique(flat).shape[0]:
        return False

    # vector
    pos = pos - np.roll(pos, 2)
    # concatnate
    pos = np.hstack([pos, np.roll(pos, 2)])
    # \vec{A} \bullet \vec{B}
    _inner = np.apply_along_axis(
        lambda x: np.inner(x[:2], x[2:4]), 1, pos)
    # |\vec{A}| \cdot |\vec{B}|
    _len = np.apply_along_axis(
        lambda x: np.linalg.norm(x[:2])*np.linalg.norm(x[2:4]), 1, pos)
    # \sum_i \frac{\vec{A_i} \bullet \vec{B_i}}{|\vec{A_i}| \cdot |\vec{B_i}|}
    if round(abs(np.sum(_inner/_len)*360/np.pi)) != 0:
        return True
    return False

Fix Non-convex Quadrilaterals

これでいけそう

非凸の頂点...簡単。反時計回りで行くなら時計回りになっている点を見つければそれが非凸の頂点。
二等分線...非凸の頂点と隣の2点から二等分線を求める
移動距離...これが問題、素直に100ずつ移動させるか、一気に移動させる方法がないか模索中。

Internal nodes

チェックポイントのパス構造には内部ノードは存在してはいけない(自身のコース製作で経験済み)。CPUとアイテムルートは別にパス関係が木構造になっていてもok.