a basic conv in MSDNet, two type :param nIn: :param nOut: :param type: normal or down :param bottleneck: use

a basic conv in MSDNet, two type :param nIn: :param nOut: :param type: normal or down :param bottleneck: use

type = ["split", "cross"]

log-likelihood for mixture of discretized logistics, assumes the data has been rescaled to [-1,1] interval

numerically stable log_softmax implementation that prevents overflow

numerically stable log_sum_exp implementation that prevents overflow

Computes the error@k for the specified values of k

like concatenated ReLU (http://arxiv.org/abs/1603.05201), but then with ELU

log-likelihood for mixture of discretized logistics, assumes the data has been rescaled to [-1,1] interval

lnL = -1/2 * { ln|Var| + ((X - Mu)^T)(Var^-1)(X - Mu) + kln(2*PI) } k = 1 (Independent) Var = logs ** 2

compute sample mean and precision (inverse of covariance) return: sample_class_mean: list of class mean precision: list of preci

Replaces x^i with q^i(x) = U(x, x + 1.0 / 256.0). Args: x: 4-D Tensor of shape (NCHW) n_bits: optional. Returns: x: x