# coding: utf-8
#import cupy as cp
import numpy as cp
import numpy as np

def cross_entropy_error(y, t):
    if y.ndim == 1:
        t = t.reshape(1, t.size)
        y = y.reshape(1, y.size)

    # 教師データがone-hot-vectorの場合、正解ラベルのインデックスに変換
    if t.size == y.size:
        t = t.argmax(axis=1)

    batch_size = y.shape[0]
    return -cp.sum(cp.log(y[cp.arange(batch_size), t])) / batch_size