时序数据切片 :param data: shape [B, ...] :param window: :param horizon: :return: X is [B, W, ...], Y is [B, H, ...]

r""" Samples from the Gumbel-Softmax distribution (`Link 1`_ `Link 2`_) and optionally discretizes. Args: logits: `[..., num_features]

Computes the sampling probability for scheduled sampling using inverse sigmoid. :param global_step: :param k: :return

compute a list of chebyshev polynomials from T_0 to T_{K-1} Parameters ---------- L_tilde: scaled Laplacian, np.ndarray, shape (N, N)

Parameters ---------- distance_df_filename: str, path of the csv file contains edges information num_of_vertices: int, the number of

Disable cudnn to maximize reproducibility

compute \tilde{L} Parameters ---------- W: np.ndarray, shape is (N, N), N is the num of vertices Returns ---------- scale

:param data: [B, *] :param val_days: :param test_days: :param interval: interval (15, 30, 60) minutes :return:

inputs : x : input feature maps( B X C X W X H) returns : out : self attention value + input

Initialize the learning rate scheduler

Initialize optimizer

compute \tilde{L} Parameters ---------- W: np.ndarray, shape is (N, N), N is the num of vertices Returns ---------- scale

Normalized graph Laplacian function. :param W: np.ndarray, [n_route, n_route], weighted adjacency matrix of G. :return: np.matrix, [n_rou